Home

- R

1. For the current user only Create a Quick Launch icon Other tasks Associate files Zoo PuytoImacr 4 User MANUAL 8 Another screen of the Zoo PhytoImage assistant You can create desktop and quick launch icon in the quick launch bar t is very important to associate files with Zoo PhytoImage those files have special extensions and it will not be possible to open them by a double click in the Windows explorer if you don t select this option So leave this option checked unless you have good reason to change it At the end of the installation you should have a ZooImage entry in the start menu and possibly a ZooImage icon on your desktop if you left that option checked E Mes documents Bureau e 5 Poste de travail aa a Mozilla Firefox zooimage Mozilla Thunderbird RD Eli Sciviews R Console Favoris r seau g Corbeille a zoolmage tl Ad Aware SE Personal Adobe Reader 7 0 AVG Free ae an td OpenOffice org Writer is d marrer oe e 6B An example desktop with the Zoolmage icon a little blue copepod currently selected Zoo PuytoIMacE 4 User MANUAL 9 4 First use oF Zoo PHyToIMAGE This quick tutorial will show you how to analyze the Spain _Bioman example images installed with the software When you double click on the Z00Image icon on the desktop or select the ZooImage R entry in the start menu two windows appe
2. Key Section Comment Id Description The short identifiant of the series Name Description A longer name for this series Description Description A short description of the series Contact Description The name of a responsible person of this series Email Description The email address of the contact Label Samples The complete label of the sample as in the file names Code Samples A code for this sample Date Samples The data of sampling in yyyy mm dd format Latitude Samples The latitude of sampling in x xx Longitude Samples The longitude of sampling in x xx 20 Don t overlook these metadata they will allow you to calculate abundances and biomasses per water volume in the field to locate your samples in space or time for further analysis etc Zo0 PuytolImMace 4 User MANUAL 45 Considering the large amount of fields in this file tt would be convenient to reimplement it in a database Any volunteer to reprogram this part of the software in an Open Source database like MySQL out there The MetaEditor displays the Description zis template Description zis Sc1 File Edit Search Yiew Tools Options Description Id Cruises Code Stations Code m Samples Labe Name ShipName Location Code Language Help Project ShipType Latitude S69 Institution ShipCallSign Longitude Series Country PortDeparture Start Cruise Location PortReturn
3. End Station Contact Captain Frequency Date You have to fill it to obtain something like this Description_Test zis Sc1 Name Bioman series Institution 4ZTI Technalia Objective Description Contact Xabier Irigoien Email xirigoien pas azti es Name Project Institution Country Location Contact Bioman AZTI Technalia Spain Bay of Biscay Xabier Irigoien ShipName ShipType ShipCallSign PortDeparture PortReturn Captain Stations Code Location Latitude Longitude Start End Frequency Samples Label Code scs Series Cruise Station Date BIO 2000 05 05 p72 P72 BIO BIO 2000 05 05 BIO 2000 05 08 p123 p123 BIO BIO 2000 05 08 You can just close the window and your changes are saved automatically 12 2 Calculating samples To process all samples in one series use the menu entry Analyze gt Process samples the shortcut ctrl s or click on the tenth button in the toolbar To process all samples in a given series click on the tenth button on the Zoo mage assistant toolbar Z00 PHYTOIMAGE 4 USER MANUAL 46 and select the corresponding zis file Select your Description zis file The program then asks to select a classifier Select your train lada object You have also to specify the limits for the different size classes to consider for the size spectra The default value creates a regular sequence from 0 25mm to 2mm with a class width of 0 1mm seq 0 25 2 by 0 1 If you clear this
4. Matlab etc lthough you can export your results to analyze them in a different software you dont have to do so Zoo PhytoImage operates in a R session and the thousands of R functions are available for producing even the most sophisticated statistical analyses and graphs without leaving Zoo PhytoImage R yana Display the PDF version of the user s manual 5 The confusion matrix is shown both in tabular and in graphical presentations 6 ECD Equivalent Circular Diameter 7 Spatial representations are not handled yet in this version but they are planned in future versions Z00 PHYTOIMAGE 4 USER MANUAL 12 5 CQUIRE DIGITAL IMAGES OF ZOOPLANKTON OR PHYTOPLANKTON Zoo PhytoImage is not a digitizing software It is only designed to analyse existing digital images However for convenience it binds to your favorite external acquisition software it should be hardware specific As an example if you use a digital camera with a dedicated capture software you can specify that software in Zoo PhytoImage and start it from the ZooImage assistant in one click Zoo PhytoImage can be used with Vuescan an excellent and very capable software to acquire pictures from more than 400 commercial flatbed scanners and from more than 100 different RAW formats of digital cameras Here we explain how to use Vuescan with a flatbed scanner to get digital zooplankton images but it should be clear that it is just an example you are fr
5. button in the toolbar The first part of your analysis import and process of your images is almost done You have now to create the zidb files These are special ZooImage DataBase files that contain all you need for the rest of the analysis but saves as much disk space as possible Those zidb files represent a convenient solution to keep all required data of even long series thousands of samples on a standard hard disk of 100 300Gb In such a case high resolution raw images consume litterally terabytes of space and cannot be all kept on the hard disk at the same time Just process your series bit by bit and backup raw images from time to time to solve the problem Now click on the fourth button in the ZooImage assistant This shows the following dialog box foolmage data processing You should have processed all your images now The next step is to Finalize the zid files o0Image Data files There will be one data File per sample and it is all you need For the next part of your work Once this step succeed you can free disk space by 1 Transferring all raw images for zip files in the taw subdirectory to DYDs 4pps gt CD DVD Burner 2 Safely delete the whole work subdirectory possibly after verification of the process of the images 3 Remove also zim files after making a backup At the end vou should have only zid Files remaining in your Working directory Click OK to proceed select working direct
6. B3 respectively for the six pictures related to the sample Zoo Puyrolmace 4 User Manuva 16 DP Vmage sample Sde Gibon Ardaya Fatt Oe GS riire C F jo netadi gt Dosos Agera O D rage tamote v Ei A Nom Tate Type a Gestion des meges ere Dooper de fates MIAO 2000 05 05 prHA pa Tbk insys PEG P the naaa GGE w00 otek w9 Shee asp wee Cprwrenieler Ges of piters vies viene 4 paler bs as w Ogee hos s rwr vers io CL Now click on the second button on the toolbar the one with the following icon P Zoo PhytoImage asks you for the images that should be imported Select both images Select images to import 2000 05 05 p72 4 jpg Mes documents recents Poste de travail Nom du fichier BIO 2000 05 05 p72 A ipg BIO 2000 05 08 Favoris r seau Fichiers de type Jpeg image files jpa we It is then supposed to check that image formats and names are correct and possibly propose to change or convert them but that feature is not implemented yet It then checks if metadata files files with zim extensions are associated Since you did not copy these files with your images they are not found and Zoo PhytoImage creates them It also displays their content in the built in metadata editor Sc1 each file in turn Z00 PHYTOIMAGE 4 User MANUAL 17 Prececeree cl Linge LJ Dirag Ea o Author Handware EPSON 427C hor Tale Typa Sofiwa
7. E 5 training set i H alter a Annand C haakn T gt hgt You are then prompted for a name to give to the ZITrain object that will be created Question Name for the l Train object Call your object simply training and click oK Zoo PhytoImage processes the tree it takes a while for large training sets and then displays basic statistics about your training set that is the number of vignettes in each group in the R Console window Zoo PuytoIMacE 4 User MANUAL 37 I R Console Fichier Edition Misc Packages Aide Zoolmage Tapez demo pour des d monstrations help pour l aide en ligne ou help start pour obtenir l aide au format HTML Tapez q pour quitter Loading required package Loading Tel Tk interface Loading required package Loading required package Loading required package Loading required package Loading required package Loading required package Loading required package Loading required package randomForest 4 5 15 Type rfNews to see new Loading required package Loading required package Manual training set data Classification stats Ra ECItK done teltk2 svMisc svDialogs Mass graphics grDevices stats randomForest features changes bug fixes utils svllidgets collected in training Annelida Appendicularia badfocus hadobject 8 1 3 11 Chordata other Cnidaria Copepoda Crustacea other 19 2 100 72 marine
8. a very useful diagnostic to determine if segmentation and detection of the objects was correct So leave this option checked now The show outlined objects option works only for the last picture rocessed so either uncheck process all items in this directory or be repared to wait for the last picture to get this diagnostic image You should zoom in the image Image gt Zoom gt 100 entry menu and pan it by selecting the hand button and dragging the image content in the window to best see the result When you start the process by clicking ok on the dialog box ImageJ do the following work Itopens a Log window and reports its activity in it It opens each image in turn process it and possibly measure particles and extract vignettes You can follow the process on the screen Note that a scale bar is added in the top right corner of each vignette for convenience It possibly displays the outlined objects of last picture if it was requested also the last table of measurements is left open for inspection If the process failed somewhere look if your images are of the right type if they are not too big for the RAM memory allocated and if the correct lugin parameters set and calibration set where selected Look at the log ile and the images produced in the work directory to help you track the roblem Always check the log file seeking for errors and take the habit to inspect outlines objects and table of measurements at
9. convention imposed by Zoo PhytoImage which is Mos fl OD DS TFE eae With this convention the images are easily identifiable in a long series both by the software and by the human In particular sorting files alphabetically results in a chronologial sorting of the images according to sampling dates 1 SCS is the identifying code of the Series Cruise Station Use three to four letters to identify the point within all you series cruises stations data 9 Once the images are completelly processed you just need the resuting zidb files somewhere on your hard disk So you can delete original pictures after making a backup on DVDs or external hard disks and save a lot of disk space Zoo PuytoImMace 4 User MANUAL 15 2 YYYY MM DD is the date of sampling in year month day format If for some reasons the day or the month is unknown use 00 3 SS is a code to uniquely identify each sample useful when several samples are taken the same date at the same station 4 PP is the image identifier Zoo PhytoImage manages different images per sample and even images of different fractions at different dilutions of the same sample Zoo PhytoImage will carry all required calculations including collecting together results from the six images in a single zid file calculating abundances and biomasses per m taking into account the two fractions at different dilutions etc 5 EXT is the file extension according to the fil
10. dialog box gives access only to a couple of them Moreover in order to simplify the process only default values are given for parameters The solution you will obtain is thus often suboptimal any machine learning algorithms should be put in the do not try this at home category It means that you need a trained biostatistician to get the best from them and to analyze results to make sure they produce consistent reliable and accurate identification of your plankton items Everything was voluntary simplified in the Zoo PhytoImage dialog box just to give a flavor of these algorithm to everybody and to allow a round trip process of your data in an easy way Don t be fooled by the apparent simplicity of the process using Zoo PhytoImage dialog boxes For serious analyses consider to fine tune your classifier with a biostatistician that will use all the functions provided by R he will rogram code in R s native language instead of just clicking with the mouse on a few options in the dialog box There is no warranty on the results and we would not endorse responsability of the consequences for alse results published after using uncertified toy classifiers 10 1 Training a classifier p To train an automatic classifier with you manual training set use the menu entry Analyze gt Make classifier the shortcut Ctrl C or click on the seventh button in the toolbar Having a ZITrain object in me
11. i 0 Oot ps oo 1 79o DAU S 0e i s 660 OGL 9d 3 4 x gt MoM m Yrs ee Co 60 JomD Ct ce p 31h a m ashe e Bio IO os gs m 309 Fie z gt 2824 ce P 660 POD t O8 2h ye 4904 t Ho 7O St aL were oO oes 37 P2099 es 2 80 WO OF O8 oi 1 79 19 we j lt pte 9 a UO WO ot ot xm io zy ta tt 0 m t co gs gt z eem i gt oer srs amp t 010 WoO ps 3 ede you i 4498 os i 660 TOOD O6 O8 pi ims 304 418 s0 4207 es es i 0 7OD 06 08 gli D Sbt IDO 1 4 7 GOO es o it 80 7 O6 O8 217 1 7079 7 e 112 es v 1 uo 7eoe o4 o8 o 1 4727 woe ties 400 e i 00 oot o paz 96 Oe tor e4 ss O60 JOA giroa 3 286 wb re oO120 ce oa to Bio WOOO 0s gt Dairi sv 2th reve l 43 HO 7000 ot 0e jipet se we gt 1543 te e i HO 7008 Of 2h Lore 491 27799 ts 63 860 WOO Of O8 oh7 gt 1 bebe gt imi ye ree es a 80 700c o 3 ore e rer e gt 69 e 650 W ot o ote 374 2 oh er e id Once you have done with your image processing you can close ImageJ and return to Zoo PhytoImage either restore the ZooImage assistant window or restart the program depending if you minimized or close it when you started ImageJ Zoo PuyToIMaGE 4 User MANUAL 26 8 CREATE ZID FILES To finalize your images import process you must now build zid files In the ZooImage assistant use the menu entry Analyze gt Make zid files the shortcut ctrl z or click on the fourth
12. in the same file and thus you reload them all at once in this case The RData files can be exchanged between computers even on different platforms for instance RData files generated on Windows are totally compatible with those made on Linux Unix or MacOS X e Objects gt Save gives you the opportunity to select one or more ZIxxx objects Zoo PhytoImage specific objects present in memory and to save them in a file e Objects gt List prints the list of all Zoo PhytoImage objects currently in memory e Objects gt Remove permanently deletes one or several objects from memory Consider using this command to free memory if you created a lot of objects that you don t need any more The RData files are very convenient to exchange training sets and toroughly tested classifiers with your colleagues Everything is included in the RData files to reuse those manual training sets and or these classifiers on a different computer R has a mechanism to save and restore automatically all objects in memory when you quit the program and restart it from the same active directory When you quit R File Exit on the R Console or click the close button of the R Console you have a question Save workspace image that appears If you click No R exists without saving anything If you click yes it saves the data in the file RData in the current active directory the one reported in the status bar of the ZooImage assitant window
13. least for the last image in your series The plugins created several subdirectories in your process directory 1 A _raw subdirectory contains raw images that were successfully processed 2 A _work subdirectory contains temporary intermediary images left there for further inspection and diagnostic Once you are satisfied with the treatment you can delete the whole work subdirectory to save Zoo PuytolmMacE 4 User MANUAL 25 space on your hard disk 3 One separate subdirectory for each sample bearing the sample name everything before the sign in the images zim file names This subdirectory contains all the vignettes for the sample possibly combining various images and or fractions and dat1 zim file s with metadata plus measurements for each image p Urmet samp ie tw Edn Atehege Fauts Quik CD rrectderee C j Petade Dosos tht Apert O Oc wrage tamcie A ion e Spr a hes Moges Du rt TIe 2000 05 05 072 9870 2000 05 05 5122 se 020 2000 05 05 of A Astres ciple coments gt waco 200438 512A Let ats E Gestion des Ac heers Here is how a dat1 zim file looks like Notice that you have two new sections appended at the end of your metadata Process that gives information on the processing parameters used and Data with a table of measurements don on each particle e ow t lt i t HO FOOO Ob O8 ot 1 7842 W co A r te 2 80 70D O8 O8 ata pre os rT a 4677 ta 2
14. make your training set rapidly starting with a long historical series already available in your laboratory it could be interesting to first choose the representative samples that will be used in the training set and digitize them in priority That way you do not have to wait that all the samples in the series are digitized and processed to make your training set Also if different people are digitizing the sample technicians and making the training set specialized taxonomists and biostatisticians you could have work done in parallel once the few samples required for the training set are digitized To experiment with our example images create first an empty directory dedicated to this training set You can create it anywhere on your hard disk but if you create a subdirectory in your process directory D image sample make sure you prepend its name with an underscore like train for instance That way ZooImage will ignore it in further processing of your images Of course do not use rawor work for the name of this subdirectory since these names are reserved for the image processing treatment see importing images Create now an empty train subdirectory in you processing dir Now click on the fifth button on the ZooImage assistant toolbar A dialog box with instructions appears on screen Zoo PuytoImMacE 4 User MANUAL 31 ooImage prepare training set This step prepares a directory in the hard disk where you will ha
15. object that is about to be created Question Name for the lClass object to create Enter train ida and click ox The algorithm learns how to recognized your zooplankton based on your manual training set When it is done its performances are assessed using a method called 10 fold cross validation Then a summary of the results total accuracy and error by group is reported to the R Console If you want you can now test and compare other algorithms with the same training set Also if you notice that one or several groups have Z00 PHYTOIMAGE 4 USER MANUAL 40 consistently high errors it means they are not well separated Could you consider reworking them in the context of your analysis Look also at the confusion matrix hereunder for further diagnostic tools 10 2 Analyzing classifier performances D Further diagnostic tools are provided to study the performances of your classifier use the menu entry Analyze gt Analyze classifier the shortcut Cctr1 n or click on the eigth button in the toolbar Having a ZIClass object in memory you should calculate a 10 fold cross validated confusion matrix between your manual and the automatic classification The confusion matrix is a square matrix that compares all groups af the manual classification with all groups of the automatic classification The number of items in each cell corresponds to the counting of objects The diagonal from top left to bott
16. on the window Then paste this graph in Word If required you can resize the graph window first to adjust the size of the graph relative to the size of the text If you have lots of graphs on the same page you are better to maximize the graph window rst You can open several graph windows simultaneously for comparison In the Utilities menu of the ZooImage1 assistant you have three entries in the R Graphs submenu New Activate next and Close all They are self explicit The Utilities gt R Graphs gt Activate next switches the active flag to the next graph window Indeed there is only one active graph window at a time It is the window that will receive the next graph s Its name ends with ACTIVE The name of all other graph windows if any end with inactive To send the next graph in a different window as the active one use the Activate next menu entry until the target window becomes active Z00 PHYTOIMAGE 4 User MANUAL 49 12 4 Analyzing results in R All Zoo PhytoImage objects inherit from data frames which are the basic case by variable type in R Consequently all the analysis and graphing functions of R can also be used without change on Zoo PhytoImage objects Look at the abundant litterature and the more than 5000 additional R packages available on CRAN http cran r project org to perform your analyses Look in particular at the task views about environmetrics graphics machine learning spatial spat
17. or individuals in rows Additional names for columns and or for rows are allowed Such data can be stored in plain text being ASCII UTF 8 encoded using a predefined field separator and one row per line The most commonly used format is CSV for comma separated values It uses either the comma Zoo PuytolImMace 4 User MANUAL 3 English version or a semi colon French version Another frequent variant is the TSV format which uses tabulations as field separators CSV or TSV are readable by all software making them the most universal storage format for case by variable data Excel or other spreadsheet formats can be used as well but they are a little bit less widely recognized CSV or TSV are not the most efficient formats when it comes to memory usage or speed Since numbers as stored as character strings they consume much more memory than their binary counterpart It is also impossible to retrieve some data in the middle of a table without reading all previous data since the offset in memory where those data are stored in not predictable Another shortcoming of the CSV or TSV format is the impossibility to associate metadata in addition the main two dimensional table Yet this format remains one of the best to store small to mid sized raw datasets and make sure they will be most readable in the future n Zoo PhytoImage we use a variant of the TSV format where the two dimensional table of features measured on ea
18. r project org Finally recent additions made in version 4 do complete the set of features Main changes are Updated code for running on latest R version 3 Complete internal refactoring to make it compatible with Linux and Mac OS X in addition to Windows Version 3 also supports Windows Vista 7 and 8 in addition to Windows XP A new storage format called ZIDB that is much faster to retrieve vignettes Routines to build sort and use test sets similarly to training sets Functions to display vignettes directly insode R graphs using R scripts Improved handling of confusion matrices with the possibility to change prior probabilities of the classes and inspect how this changes the shape of the confusion matrix A battery of summary statistics for the confusion matrix recall precision F score specificity New and improved graphs for the confusion matrix including F score plots and dendrograms depicting hierarchical classification of the classes according to their confusion 2 1 New data storage format Among all those change the most important one for end users is probably the new storage format named ZIDB for Zoo PhytoImage DataBase Data storage format is a key aspect for data analysis software In statistics there is a consensus towards a case by variable format that is suitable for most but simplest datasets It presents the data in a two dimensional table with variables in columns and cases
19. recognize zooplankton taxa on the basis of images measurements done in Zoo PhytoImage you have to make a manual training set In Zoo PhytoImage you can have a relatively complex organization of the different groups taxa ecological groups or any other grouping of the plankton that suits your needs in a hierarchical tree Hence you have relationship between the groups for instance Sapphirina intestinata and Sapphirina ovatolanceolata are collected together in the Sapphirina sp group Copilia sp and Sapphirina sp form your Sapphirinidae group Sapphirinidae together with Oncaeidae and Corycaeidae which contain also corresponding subgroups are collected together in the Poecilostomatoida etc Up to the top group called Copepoda You can also decide to make other groupings like ecological groups or even mix the styles You are here 100 free of the groups you create but there are a couple of constraints 1 make logical hierarchy of your groups and subgroups 2 keep in mind the parameters abundances biomasses and partial size spectra that you want to calculate on these groups 3 make only groups where you can actually classify vignettes with a reasonable accuracy solely on the visual inspection of these vignettes 4 it is useless to make groups for very rare items you need at least ten to fifteen example vignettes in each group in your training set 30 to 50 is even better 5 ultimately the most pertinent grouping is the on
20. required unless you use a RAID system 3 2 Download of the software The software is available for download on the ZooImage website http www sciviews org zooimage It can also be installed from within R through CRAN run install packages Zooimage from the R console Linux or Mac OS X users will have no problems installing R and then Zoo PhytoImage that way The following section details installation from the Windows installer as it exists since early versions screenshots not updated 3 3 Installation of Zoo Phytolmage under Windows Zoo PhytoImage will uses about 400Mo of space on your hard disk when installed You just have to execute the ZooImage_ x y z Setup exe Zo0 PuytoIMacE 4 User MANUAL 7 file that you downloaded and to follow the installer s instructions step by step Default values for the options should be fine if you don t understand them ie Setup oolmage Welcome to the 7ooImage Setup Wizard This will ingtall oolmage version 0 3 2 on your camputer his recommended that you close all other applications before continuing Click Hest to continue or Cancel to exit Setup The first screen of the ZooImage installation assistant is Setup Zoolmage Oo led Select Additional Tasks Which additional tasks should be performed Select the additional tasks you would like Setup to perform while installing oolmage then click Nest Additional icons For all users
21. snow multiple part scratch 30 unknown 17 Proportions per class 26 13 T Annelida Appendicularia badfocus badobject Tere tens 00 2840909 08522721 3 1250000 Chordata other Cnidaria Copepoda Crustacea other ES r BATATAS 0 5681818 28 4090909 20 4545455 marine snow multiple part scratch B S22 203 7 3863636 3 6931818 1 9866364 unknown 4 86295455 Chaetognatha 37 Egg al shadow 5 Chaetognatha 10 5113636 Egg 00 2840909 shadow 1 4204545 If you see that you have too much or too few items in some groups like here only one Appendicularia and a hundred Copepoda go back to XnView and rework them before rereading your training set Note that you have too few samples available in the examples for filling each group with enough items For the rest of the demonstration you can read the example training set installed with Zoo PhytoImage as well Z00 PHYTOIMAGE 4 User MANUAL 38 10 MAKING AND ANALYZING AN AUTOMATIC CLASSIFIER In Zoo PhytoImage classifier algorithms used range in a category called machine learning Basically you feed the algorithm with example identifications together with measurements done on the same objects and the algorithm learns how to recognize the groups according to the measurements It is a very simple scheme but it has proven efficient in many situations Many algorithms exist and many are implemented in R over which Zoo PhytoImage is running The Zoo PhytoImage
22. software Use YueScan Another software You have a dialog box that let you choose the program to start either Vuescan or another one Select Vuescan and click ok Vuescan is opened Once the software is registered you can switch in Advanced mode by clicking on the corresponding button at the bottom if Vuescan is started in Guide me mode You have to parameterize Vuescan for your acquisition device digital camera or flatbed scanner and the type of images you want Vueimage allows you to record both uncompressed TIFF files with 16bit gray levels and JPEG 24bit color files These two types of files correspond respectively to the Grayl 6 bits 2400dpi and Color24bits 600dpi plugins in ImageJ see hereunder Vuescan offers a wide range of options for digitizing your pictures A couple of options are very sensitive in the context of your image analysis dditional documents are in preparation to list best Vuescan options for several digitizing devices Zooscan You are also welcome to contribute your own recipe Z00 PHYTOIMAGE 4 User MANUAL 14 6 IMPORT IMAGES P Once your images are stored on your hard disk you must prepare them for use in Zoo PhytoImage Use the menu entry Analyze gt Import images the shortcut Ctrl I or click on the second button in the toolbar Zoo Phytomage image importation is indeed performing several tasks to make sure your pictures are in correct formats and all requir
23. the amount of RAM memory required by the image process compared to the one you got on your computer The small example pictures we are dealing with do not require much RAM So if you have something like 512Mb on your machine you should be safe to keep both Zoo PhytoImage and ImageJ opened simultaneously If you analyze very large pictures you should close Zoo PhytoImage and all other running programs before starting your image processing in ImageJ As an example 16bit gray pictures of 60 million pixels for instance 10000x6000 pixels require 900Mb of RAM allocated to ImageJ You need at least 1Gb of actual RAM in your computer for dealing with such images The maximum amount of RAM you can allocate to ImageJ is system dependent On 32bit system do not try to allocate more than 1 6Gb to ImageJ or the program will crash Of course you need at least 2Gb of actual RAM in your machine to use that maximum value Although we did not tested the Grayl bits 2400dpi plugin with images larger than 10000x6000 pixels the maximum allocatable RAM value should work with images of about 100 million pixels Thus currently the largest 16bit gray images you can deal with in ImageJ is something like 10000x10000 ixels At 2400dpi it is a little bit less than 10x10cm of cell size If you have larger cell area just take several separate pictures and both ImageJ and Zoo PhytoImage will take them into account you just loose measurement on objects that are cu
24. 00 05 05 p72 A zim sc1 L Bf File Edit Search View Tools Options Language Help Z00 PHYTOIMAGE 4 User MANUAL Il Image Author Author 4uthor Kevin Denis Author Naiara_ Serrano ay 2400dpi Author Philippe Grosjean ere Code 4 Min 1 Mlax 1 Subsample SubPart 1 SubMethod volumetry CellPart 1 00 Replicates 1 YolIni olPrec 19 If needed you can enter additional metadata Just use the key value syntax If you want to create another topic enter it in a separate line in square brackets like topic Zoo PhytoImage does not create separate zim files fo each picture It only create separate zim files for each fraction of each sample So if you have a lot of pictures related to the same sample and fraction this is likely to be the case if you work with FlowCAM or VPR images you just have to fill one zim file for all of them You can customize both the default entries in the metadata and the list of roposed values are customizable Just edit those files bin MetaEditor templates default zim and zim api from the base Zoo PhytoImage directory Note that you cannot use spaces in the list of suggestions in the zim api file Use the underscore instead ZooImage will convert it in a space in due time So Author Alfred Hitchcock should be entered in the list of possible completions instead of Author Alfred HEECHCOCK Meaning of the metadata entries Entry Topic Explanation Z This
25. It also saves the history of commands in a Rhistory file in the same directory The next time you start R you can restore this RData file if you like It is far better to use the Objects menu and selectively save restore given objects than to systematically rely on this mechanism This way you can also choose a meaningful name and directory where you store your data So if you save your objects using the Objects menu of Zoo PhytoImage you can systematically answer No to Save workspace image when you quit R ZooImage Z00 PHYTOIMAGE 4 USER MANUAL 42 12 CALCULATING VISUALIZING AND EXPORTING SERIES This section supposes that you have already made zidb files from your raw images part I and that you have a valid ZIClass object in memory part II either that you just created or that your reloaded from a RData file Up to now all treatments were made at the sample level You never had more than one sample loaded in memory A sequence of samples or images was always treated one by one by Zoo PhytoImage possibly reporting long processes in a log file so that you can leave the software unattended doing the calculation and come back later to see the results it seems that the koffee room will be more crowded that usual This is a feature Zoo PhytoImage is not designed as a toy program that would be just able to calculate a couple of demo examples but that will crash with an out of memory message with any serio
26. PhytoImage Results They are most suitable for the space time analysis at the series level which can be done in R ZooImage directly or you can export the tables to analyze them in another software like Matlab for instance 12 1 Creating and documenting a series EG A series is a collection of samples plus a few additional metadata To edit a series description file zis file use the menu entry Analyze gt Edit samples description the shortcut Ctrl D or click on the nineth button in the toolbar Until now your zid files had independent lives totally ignoring each other It is now time to tell to Zoo PhytoImage which zid files you want to collect together in a space time series This is done by editing a samples description file with a zis extension You can create as many zis files as you like making thus different series for instance a variation in time at a single station for one series a spatial coverage of the area at a given time for another series etc As an illustration of this principle you will create now a mini series collecting together the two example samples we are analyzing Click on the nineth button on the ZooImage assistant toolbar The following dialog box appears with an explanatory message and a single option 19 Note that Zoo PhytoImage does not have yet a mechanism to incrementally add data to a ZlRes object but that mechanism is planned for future versions Z00 PHYTOIMAGE 4 USE
27. R MANUAL 44 foolmage edit samples description Samples are about to be analyzed and collected together to Form a series zoolmage needs to know which samples should be collected into the same series and you must provide additional metadata information especially date and time of collection location of the sampling stations or possibly temperature salinity turbidity etc that were recorded at the same time as these samples 4 zis File by default Description zis needs to be created and edited for each of the considered series You can here edit or create 4 new samples description file From the template Click OK to edit a samples description File now a New description file From template You can either create a new description file from the template check the option or edit an existing one uncheck it Create a new file and thus leave the option checked and click OK now After telling where you want to store the description file the MetaEditor opens a template You have to fill it in order to tell to Zoo PhytoImage which samples are included in the series The zidb files corresponding to all samples included in the series are supposed to be in the same directory as the zis files themself A complete description of data and metadata in zis files is found in the annexes You do not have to fill all field Also you can add additional keys if you want Major fields that you have to fill correctly are
28. Started 200S5 1 24 10i422255 Verification BIO 2000 05 05 p72 OK BIO 2000 05 08 pi23 OK K no Error EFEound Compression BIO 2000 05 05 pra OF Bimi z0uu g0s5 07a Pls UK OF no error found Cleaning the hard disk at the end of the process Once all your zidb files are created it is time to save space on your hard disk You should do the following from time to time 1 Delete the work subdirectory once you are confident with the image processing of all your samples 2 Back up your original images in the raw subdirectory the corresponding zim files on DVDs external hard disk tapes etc Always back up your raw image files you would Zoo PuytoImacr 4 User MANUAL 28 perhaps have to redo your analysis with a better algorithm in the future and zidb files do not contain required data for reprocessing the images Once it is done delete the raw subdirectory and all remaining zim files in the treatment directory to free disk space 3 Check this in your processing directory you should only have zidb files remaining one per sample no mather how many pictures you had for each sample and no additional subdirectories or files except perhaps zis files and manual training sets if you already build them see later in the manual Zoo PuytoImMacE 4 User MANUAL 29 9 MANUALLY CLASSIFYING VIGNETTES In order to train the computer to semi automatically
29. Zo0 PhytoImage a free Open Source software to analyze plankton digital images http www sciviews org zooimage Zoo PHYTOIMAGE VERSION 4 Computer Assisted Plankton Images Analysis User MANUAL The ZooImage development team February 2014 Ph Grosjean amp K Denis Numerical Ecology of Aquatic Systems UMONS Belgium X Irigoien G Boyra amp I Arregi AZTI Tecnalia Spain A Lopez Urrutia Centro Oceanogr fico de Gij n IEO Spain M Sieracki amp B Tupper FlowCAM plugin 1 INTRODUCTION Zooplankton or phytoplankton samples analysis is traditionally associated with long and boring sessions spent counting fixed plankton items under the binocular with formaldehyde vapors floating around Although this picture of a planktonologist will probably remain for a while there seems to be another way to gather data about zooplankton computer assisted analysis of plankton digital images A whole suite of hardware to take pictures of our animals both in situ and or from fixed samples is now available Flowcam laser OPC VPR Zooscan more to come with Holocam Sipper Zoovis HAB Buoy not forgetting the use of a digital camera on top of a binocular or with a macro lens But digital images of zooplankton are barely usable as such they must be analyzed in a way that biologically and ecologically meaningful features are extracted from the pixels A software doing such an analysis is thus indispensable Zoo P
30. ages are computer intensive processes and you will likely analyze lots of objects typically hundreds of thousands or millions of them Thus you need a recent and powerful computer to run Zoo PhytoImage decently Consider especially A fast and recent multicore and multithreaded processor 4Gb of RAM memory or more Depending on the size of the images you want to analyze you may need even more Very large images issued from a flatbed scanner require at least 1Gb of RAM Zooscan images may require even more Nowadays it is very easy to use 16Gb or 32Gb of RAM on 64bit systems so consider this option seriously After the processor speed and the RAM the next most important part of your computer to work with images is the graphic card and the screen Chose a rapid optimized graphic card capable of displaying 1280x1024 or 1600x1200 pixels or more with 24 32bit color depth millions of colors associated with a high quality screen of no less than 19 Dual screen configuration can help too since it gives more space for displaying side by side images and plots Although Zoo PhytoImage optimizes disk space by compressing all files dealing with lots of high resolution pictures is consumming a lot of space on disk You will need a fast hard disk of at least 2 4Tb of capacity One small SSD disk greatly improves the speed of the analysis when used to store the few samples currently manipulated Finally a good backup system is also
31. alternate configurations Currently alternate configurations are hard coded in the plugin but users will be able to edit them freely in future versions Parameters set defines minimum and maximum particle size to consider which measurement is done which threshold is used for separating particles from background etc leave the default value now e The calibration set drop down list is similar to parameters set but define calibration data i e pixel size and calibration curves for grayscales and or color channels possibly depending on the lighting staining of the sample etc leave the default value now e Zip images rewrites the pictures in a zip compressed TIFF format This is not useful for JPEG images because they are already compressed So uncheck this option now e Analyze particles do the measurements on the particles after processing the images leave this option checked now Z00 PHYTOIMAGE 4 USER MANUAL 24 e Make vignettes estract small images for each identified object called vignettes in Zoo PhytoImage s terminology leave this option checked now e Sharpen vignettes optionally applies a sharpen filter on the pictures in the vignettes This often enhances the quality of the vignettes but is not necessary for some kinds of pictures leave this option checked now e Show outlined objects displays a composite image with the detected object outlines superposed to the grayscale image This is
32. and to menu in Zoo PhytoImage 4 Vignettes are accessible directement within R and can be included anywhere in R plots or displayed as a gallery The code to do so looks like this Lazy loading data from one ZIDB file in R db1 lt zidbLink path_to_zidb Contains data in _dat1 and vignettes in _nn items1 lt Is db1 vigs1 lt items1 grep dat1 items1 Display a 5 5 thumbnail of the first 25 vignettes Fig 13 3 zidbPlotNew The 25 first vignettes in MTPS 2004 10 20 H1 for i in 1 25 zidbDrawVignette db1 vigs1 i item i nx 5 ny 5 The summary method of a ZIClass object a classifier displays a lot of summary statistics like recall precision specificity F score balanced accuracy etc These statistics are calculated group by group See the help page of the ZIClass object ZIClass The ZIClass object has a confusion method that creates a confusion matrix with four specific plots image barplot stars and dendrogram The barplot is a new view of F score called F score by group plot See confusion and the example in the R script The star plot can also be used to compare two classifiers applied to the same test set There are also complements about the way Zoo PhytoImage calculates abundances and biomasses biovolumes You can calculate these quantities at different detail levels and indicate which groups are out of interest e g marine snow and zooplankton if your study focu
33. any you will have to get a license for XnView before you can use it Zoo PuyrolmacE 4 User MANUAL 34 Tree If you do not have a thumbnail view in the XnView browser main window you can have an icon list or tabular view of the files as well select View gt View As gt Thumbnails Both the folder three and the main window in thumbnail mode are required for the rest of the work Now begin to classify the vignettes manually by moving them in the corresponding directory in the tree by drag amp drop with the mouse It is easier to move vignettes first in top directories all copepods in Copepoda all appendicularians and chaetognathes in Append Chaeto etc Then you open the Copepoda subdirectory and classify vignettes from there to deeper levels Gymnoplea or Podoplea etc Of course this work should be done by or with help of trained taxonomists t makes sense to ask different taxonomists to classify the same vignettes independently so that you can check unmatching results and build a consensus that is supposed to bear less errors than a single manual training set We may add tools for analyzing and building consensus training sets in the future in ZooImage but it is not the case yet in the current version You are not restricted to the groups and subgroups already made You can freely modify the structure of the tree change directories add or delete other ones In the tree panel of XnView browser you right click in a dire
34. ar on screen the R console and the ZooImage assistant s 4 OF _ Zoolmage assistant Mes documents Bureau DARED EBS ve Bi y Readly D image sample R Console Tapez contributors pour plus d information et citation pour la fagon de le citer dans les publi Tapez demo pour des demonstrations helpi pour en ligne ou help starti pour obtenir l aide au for Tapes qfi pour quitter R Loading required package tcltk Loading Tel Tk interface done Loading required package tcltke Loading required package syMisc Loading required package esvDialogs Loading required package MASS Loading required package graphics Loading required package grbevices Loading required package stats Loading required package randomForest randomForest 4 5 15 Type ciNews to see new features changes bug fixes Loading required package utils Loading required package syvWidgets gt E 4 2 demarrer EOS sa o Zoolmagel assistant The two first windows appearing when you start Zoo PhytoImage At bottom left the R console you can interact with R there and at top right the ZooImage assistant window The R console allows you to control R directly through command lines You should not worry about this window unless you are familiar with the R language However it logs important results and messages from your actions in Zoo PhytoImage So you are better not to minimize it The ZooImage assistant window is a
35. ater it offers more flexibility like for instance the possibility to define an alpha channel defining the transparency of the pixels in addition to their color GIF can only tag one color as being fully transparent and the other ones are fully opaque In Zoo PhytoImage the PNG format can be used in addition to JPEG for vignettes if a lossless compression is required and the image has a chance to be further analyzed at a later stage after the vignette is created This alternate format for vignettes was introduced in version 3 of Z00 PhytoImage In Zoo PhytoImage data for one sample contain three components 1 A case by variable table containing fetures extracted by image analysis on each blob particles or individual items in the images This table is stored in TSV format and in binary R own format Rdata containing a data frame for quick loading in R 2 A list of metadata information about the sample contained in a plain test file with an ini file organization with one key value per line The same information is also stored as attributes of the R s data frame 3 A series of vignettes that are cropped subsections of the initial images containing the picture of a single particle and enhanced for visual identification These vignettes were stored as JPEG images in version 1 and 2 but PNG format is now also accepted from version 3 Since there can be easily thousands of particles and thus vignettes in one single sample
36. ch particle is prepended by a section defining associated metadata in a key value pairs set It is the dat1 zim file The same data is also duplicated in own binary R format RData which is much faster to load than the original TSV file Zoo PuytoImMacE 4 User MANUAL 4 For the images there is a large number of formats availables The most widespread used ones are TIFF JPEG GIF and PNG TIFF is the most versatile one but the number of subformats that exist makes it difficult to read on some software for the most exotic configurations It is the preferred format for RAW plankton images to be processed by Zoo PhytoImage JPEG is a lossy compression format that is restricted to RGB 24 bit images only It is the most efficient lowest size of the file for images that should only be viewed However the compression algorithm introduces artifacts in the picture that cannot be reliable analyzed when compression factor is too large This format is reserved for vignettes small images containing only one particle when they are only used for visual classification of the particles no further image analysis on them GIF and PNG are image format that use lossless compression algorithms PNG was proposed as an alternative to the older GIF format because of patent and licence problems the non free licence of the GIF format was a problem in the past but now the patent has expired and this format can be freely used However PNG being defined l
37. contains very few particles 1 20th of the whole sample Yet you almost doubled the fraction of erroneous particles at that step Run it a third time AD Beck eo Dor a BE ArMix 2009 04 29 300A4X_01_5 Step3 Suspects and error corrected at each iteration is Zoo PuytoImMace 4 User MANUAL 57 On this sample the algorithm predicts a relatively low amount of suspect items on other samples with a higher proportion of initial error this fraction can easily reach 80 to 90 Nevertheless the fraction of erroneous particles has increased a little bit more You are now concentrating the error more efficiently Continue with a few sets iwetee ee a oe l E SE Artec 2000 04 29 OAS ao OL aim ee a a Nexg gt gt Done BE ArMix 2009 04 29 300A4X_01_2 Step7 SUSPects and error corrected af each iterabon r e Sec ore Here after step 7 you notice two important things First the detection of suspects now closely matches actual error Detection improves with the fraction of sample already validated that can be used for training the detection algorithm Second residual error drop to less than 10 From this moment on you know that you have manually validated all erroneous particles down to about 5 But since the model is also used to calculate a correction factor for the remaining items the calculation of abundances per classes will become quite good Also remember that par
38. ctory and select New Folder Delete or Rename entries to rework the tree Make sure all people that build the training set or similar training sets have the same perception of each group Define clearly which kind of object should go in which group print these directives and keep them on your desk for reference when you classify your vignettes Also if you plan to build a consensus training set collecting together independently trained data or if you want to build similar training sets for different series you must work in two stages e First define the structure of the tree with all concerned people and define clearly which vignette should go in each group At the end of the process it should be useful to have a definition file with a zic extension off this reworked tree Distribute this zic file to all collaborators and ask them to make their training sets with the same tree without modifications e Second build your manual training set with the tree and groups you just defined When you classify your vignettes you should try as much as possible to classify them down to the most detailed subgroups If there are many vignettes you cannot classify deeper than a certain level although your tree has more detailed groups it means that you were too ambitious in the level of details you want to reach in the tree Rework your tree and eliminate problematic subgroups where you cannot classify those vignettes Zoo PuytoImMacE 4 Use
39. d of the import process you should get a report in a ZooImage log window that pops up It should look like this IF Zoolmage log E Bjk Fichier Edition foolmage log started 2005 11 22 11 36 24 D image samp ler BIO 2000 05 05 pv2t a zim creating the file D image sample BIO 2000 05 08 piz234 i fim creating the file OF MO error found Take care that you should have the OK no error found message at the end of the log For only two pictures this log is not very useful but imagine the advantage of logging individual error if you import thousands of pictures and when all the checkings file names formats etc will be activated Now you D image sample directory should look like this Open a zim file Mes documents EE 000 05 06 5 p123ea jPa r cents E BIO 2000 05 08 p123 4 zim zri E Bureau Mes documents Poste de travail e Nom du fichier BID 2000 05 05 p 2 4 zim i Favoris r seau Fichiers de type Tous les fichiers ka Z00 PHYTOIMAGE 4 USER MANUAL 21 7 PROCESS IMAGES To process your images use the menu entry Analyze gt Process images the shortcut Ctrl g or click on the third button in the toolbar Zo0 PhytoImage will now switch to ImageJ a free image processing software Before doing so a dialog box proposes to close Zoo PhytoImage Whether you can leave Zoo PhytoImage open at the same time as ImageJ or not depends on
40. der hierachy as the one used in the original training set with their vignettes pre sorted according to the automatic prediction done by the chosen classifier This serves as two purposes 1 to visually check the quality of the classifier through the vignettes identifications and 2 to allow for further manual correct validation of that classification In this case you can read the test set back as you do with a training set and you obtain a fully validated classification of your sample Zo0 PuytoIMacE 4 User MANUAL 5O Validate classification is a new tool that combines advanced statistical tools and a new user interface to easy partial validation of classification The tools detects so called suspect items and present them first step by step so that optimisation procedure is more efficient Typically validation of only one third of all vignettes yields same level of error correction than a 90 95 random validation procedure It is also combined with tools to model the error specifically for that sample and to perform statistical correction according to that model The combination of suspect detection and error correction provides even faster improvement of the validation by manually validating 15 20 only of the vignettes one gets abundance by groups calculations with typically less than 10 of error for all groups 12 7 Smart validation of classification Here is how to use the validate classification tool First make su
41. e format It must be tif lowercase for TIFF images and jpg lowercase for JPEG pictures You do no have to conform to the Zoo PhytoImage naming convention of the images However the minimum is to use NAME PP EXT with whatever string you want that uniquely identify one sample being at least A if you have only one image per sample and EXT as above Thus as a minimum TIFF images should end with A tif and JPEG images with A jpg That say we will now practice on the example pictures 1 Prepare an empty directory on your hard disk let s say D image sample but you can freely choose another partition or directory name 2 Swith the active directory there using the Utilities gt Change active dir menu and select that directory 3 Copy the two example pictures BIO 2000 05 05 p72 A jpg amp BIO 2000 05 08 p123 A 4pg that are located in the examples raw subdirectories of your Zoo PhytoImage installation directory by default it is C Program Files ZoolImage on English versions of Windows in that directory Do not copy corresponding zim files 4 You should have something like this without the treatment subdirectory 10 For instance you filter you sample on a 1000um sieve and apply different dilutions for the large fraction and the small one Just decide to call your large fraction A and your small fraction B Now if you make three pictures for each fraction PP will be A1 A2 A3 B1 B2
42. e graphs of your results click on the eleventh button on the ZooImage assistant toolbar a and select the ZIRes object you just created results You have then a list of possible graphs Zoo PuytoIMacE 4 User MANUAL 47 Select 1 to 12 graphs Biovolume Temperature Salinity Chla Abd total Abd Chaetognatha 46d Chordata other Abd Copepoda 46d Crustacea other 46d marine snow Bio tatal Bio Chaetognatha Bio Chordata other Bio Copepoda Bio Crustacea other Bio marne snow spectrum of BIO 2000 05 05 p72 spectrum of BIO 2000 05 08 p1 23 As the title of the list says you can select between 1 and 12 graphs to draw If you select the two spectra at the bottom of the list the program asks also if you want to plot the spectra of a given taxa in red superimposed on top of the total size spectra Select Copepoda in the list and click ox You should obtain a composite graph similar to this one Z00 PHYTOIMAGE 4 User MANUAL 48 R Graphics Device 2 ACTIVE DER Fichier Historique FRedimensionnement BIO 2000 05 05 p72 Frequency 0 25 0 35 0 85 0 95 145 1 55 BIO 2000 05 08 p123 Frequency 0 25035 0 85095 145 155 You should experiment with the different possible options here You can copy these graphs in Word Just use File Copy to clipboard gt As Metafile in the graph window menu or use Copy as metafilein the context menu after right clicking
43. e that the computer can actually discriminate with a reasonable accuracy You have to classify all kinds of items Even those you are not interested in may be bubbles marine snow phytoplankton if you are only interested by zooplankton etc Indeed you have to recognize those items to elimine them from the countings and you need a group in the training set for that You don t need to classify all vignettes When you have about 50 items in a group and you think it is well representative of the overall variability in shapes of that group you don t need to add more vignettes Also fuzzy objects unrecognizable ones multiple or part except for VPR images rare taxa etc do not need to be classified Abberrant individuals which are not likely to occur often in your samples should be eliminated too You have a special top group named in the hierarchy for all these items AH vignettes in the _ top group or any of its subgroups will not be considered in the training set For biomasses calculations it could be useful to further split groups depending on the orientation of the animals conversions formulas could be different for lateral or dorso ventral views of the same animals Make subgroups for them if you want to take advantage of these different conversion formulas Ex Oithona sp lateral versus Oithona sp dorsal Make sure you use unique names for all levels of all groups Do not use a Classification l
44. ed metadata are associated In the current version the function just checks the presence of metadata files but more exhaustive control and processes are planned in future versions It means you have to do the rest of housekeeping manually Here is what you should do e Make sure that all the images you want to process are in one directory on your hard disk Do not mix pictures you want to process with other ones on the same directory Keep them separate For instance have one d ImageProcess directory where you store your fresh images and place them in one d ImageDone directory as soon as they are processed ince Zoo PhytoImage always starts from the current active directory when you have to browse for files and subdirectories it saves time to switch it to the one where you store your raw images The active directory is displayed in the status bar of the ZooImage assistant window To change it use the Utilities Change active dir menu entry e Make sure your images are in a correct format uncompressed TIFF with 16bit gray scale preferably with a resolution of 2400dpi for the Gray16bits 2400dpi plugin and 24bit color JPEG preferably with a resolution of 600dpi and with the lowest compression level for the Color24bits 600dpi plugin Other file formats will be accepted in the future Use general graphic utilities like Imagemagick or XnView to convert image that are not in one of these formats e Make sure you respect the naming
45. ed samples scanned with the Zooscan for instance The general framework of Zoo PhytoImage is designed in a way that the software is capable of dealing effectively with images of various origins and characteristics Consequently it is not a streamlined and rigid system It is rather made of a collection of different and customizable applications collected together in a single system This user s manual will guide you in your first use of Zoo PhytoImage This manual describes current version of ZooImage 4 0 0 which is not a public version It is geared towards early adoption among our artners UMONS IFREMER BelSpo ULCO and LISIC The functions resented here will eventually land in the next public version 5 However 4 5 of the code is commonwith version 3 whichi 1s public and downloadable from CRAN http cran r project org 1 The current version is developed mainly on Mac OS X but is also tested on Windows and Linux Ubuntu Zo0 PuytoIMacE 4 User MANUAL 2 2 CHANGES FROM VERSION 1 AND 2 Zoo PhytoImage version 1 2 was the latest public version distributed on http www sciviews org zooimage until now Version 2 of the software was not public and contained several developments made for us UMONS university and our main partners IFREMER in France and Belspo Belgian Science Policy in Belgium Version 3 of ZooImage collects most of these developments into a relifted system and it is distributed on CRAN http cran
46. ee to use any hardware software combination you like to acquire your images Vuescan is not a free software It is a shareware distributed in two versions personal and professional You need the professional version Its license is about 89 and you have to register your license with the author of Vuescan see instruction in the Vuescan online help We got the right to redistribute the trial version with ZooImage but you have to unleach full eatures by entring your license code before you can use it in production 5 1 The acquire images tool n this manual we use examples images installed with the software So you do not need to acquire your own image to practice with Zoo PhytoImage As an example we show you how you can get your own images using Vuescan To start your image acquisition software from the ZooImage assistant window use the menu entry Analyze gt Acquire images the shortcut Ctr1 a or click on the first button in the toolbar 8 For instance Canon or Nikon digital reflex camera are bundled with specific capture software that you can use to save directly your picture on your hard disk Zoo PuytolmMacE 4 User MANUAL 13 foolmage picture acquisition To acquire digital zooplankton images WOU Can use a specialized equipment or a digital camera on top of a binocular or a flatbed scanner To pilot a scanner or rework RA digicam images WOU Can use Wuescan You can also specify to use your own
47. entry the program understands that you do not want to calculate size spectra for these samples Keep default values and click OK now The last question is a name for the ZIRes object to create Give results and click oK now We still have to implement the table of parameters for the biomass conversion in the program Zoo0 PhytoImage calculate each sample in turn and generate a log file Once the process is done you should get a log file indocating that there is no error Your ZIRes object is now created if no error occur look at the log If there are arrors the most probable cause is a problem in the Description zis file or corresponding zid files that are not located in the same directory as the zis file Make the corrections and start the analysis again 12 3 Visualizing results To visualize you series use the menu entry Analyze gt View results the shortcut Ctrl v or click on the eleventh button in the toolbar Having now calculated a ZIRes object that contains abundances biomasses and size spectra one can visualize graphs or composite graphs with up to 12 graphs on the same page of that series Currently the program proposes only a limited number of graphs and you cannot customize colors titles etc These graphs are sufficient for a rapid inspection of time series but spatial components are not handled yet Graphs in R are very flexible and you can visualize your data in many other ways To mak
48. hytoImage aims to provide a powerful and feature rich software solution to use zooplankton or phytoplankton pictures of various origins and turn them into a table of usable measurements i e abundances total and partial size spectra total and partial biomasses Zoo PhytoImage is not tight with any of the previously cited devices and it is not going to be a commercial product It is distributed for free GPL license distributed through its web site http www sciviews org zooimage and it is open meaning it provides a general framework to import images analyze them and export results from and to a large number of systems So everybody can use Zoo PhytoImage but better yet every developer can also contribute to it The Open Source approach of wiring many willing developers around the world in a common project has already shown its efficiency Linux Apache but also R or ImageJ in the field of statistics and image analysis respectively are good examples of it Zoo PhytoImage is based on ImageJ and R and it runs on Linux but it can also be run on Windows Mac OS or various Unixes Zoo PhytoImage s best qualifying is reusability It is born by reusing various features of great existing software like ImageJ or R and it provides itself reusable components for the benefit of both users and developers Zoo PhytoImage can be used on images acquired in different situations in situ like VPR or HAB Buoy or in the lab fix
49. ike Nauplius subgroup in Copepoda and Nauplius subgroup in Malacostraca Indeed the program will manipulate groups Zoo PuytoImMacE 4 User MANUAL 30 independently for some treatments and how to differente Nauplius from Nauplius then when you don t use the grouping hierarchy Correct presentation should be Copepoda nauplius in Copepoda versus Malacostraca nauplius in Malacostraca acc Seg does not check uniqueness of group names for the moment you have to care about this by yourself 9 1 Preparing a manual training set from zidb files To install files and directories required for making a manual training set use the menu entry Analyze gt Make training set the shortcut ct r1 v or click on the fifth button in the toolbar You must first decide which samples you will use in the training set Select a couple of samples i e a couple of zidb files that are representative of the whole variability in your series Choose samples that span on the whole time scale possibly several years and the whole considered geographic area Choose also samples collected at different seasons if this applies Depending on the number of groups you want to make you will need a couple of hundred vignettes to a couple a thousands of them maximum 10 to 20 000 items for very detailed training sets Knowing the average number of vignettes you have in a sample you can determine how many samples you need usually a couple a tens If you want to
50. in future versions e The third part uses this classifier and the measurements done on all objects identified in your pictures first part to calculate automatically abundances biomasses and size spectra in all your samples You can then visualize results or export them 1 Edit samples description Series of samples are identified by a list written in a specific Zoo PhytoImage format This list contains also further metadata about the series and you have the opportunity to append various other measurements to the samples data temperature salinity fluorescence etc 2 Process samples This is the workhorse function that process each sample of a given series one after the other 1 identifying all individuals using your automatic classifier 2 computing abundances per taxa 3 calculating size classes in total and in each taxa for size spectra representations and studies and 4 computing biomasses in total and per taxa using a table of conversion from ECD to cabon content dry weight etc Data are converted per m if suitable dilution information is available in the metadata View results Graphically present results You can draw composite graphs up to 12 different graphs on the same page either time series of abundances or biomasses C or size spectra of given samples tf Export results Results are written on the hard disk i in ASCII format This format is readable by any other software Excel
51. ince this first sample is purely randomly selected Thus you know that you have a totla of about 15 error and that you already corrected 1 20th of that error If you continue to validate random subsamples you still have to look at the remaining 19 20th of the sample If you decide to accept a remaining error of less than 5 of the total you will still need to validate 2 3 that is roughly 12 20th of the whole sample But wait doing so do not guarantee that you have less than 5 error in all groups Typically you will leave far more error in the rarest groups Thus you are better to validate everything or The smart validator provides a much more efficient way of validating your sample with this goal in mind of less than 5 error in all groups To reach this goal a statistical model and a bayesian probability is calculated for each particle telling if it has a chance to be suspect understand probably wrongly classified or not Zoo PuytoIMace 4 User MANUAL 54 The model also considers several additional aspects e The probability returned by the classifier for the second class predicted for the particle is compared with the probability for the first selected class The idea is that if the difference between those two probabilities is small one should consider the particle is close to the border between the two classes and should be checked The number of particles classified in the same class for the whole sam
52. io temporal and time series for further tools that can be useful to analyse you plankton samples or series 12 5 Exporting results To write the result tables as ASCII files use the menu entry Analyze gt Export results the shortcut Ctr1 s or click on the twelve button in the toolbar If despite all the potentials of R to analyze your series right in the current environment you want to export data you can do it easily Click on the forelast button on the ZooImage assistant toolbar elect your ZIRes object in the dialog box and indicate a directory preferrably empty where to place the tables Zoo PhytoImage exports one table for abundances and biomasses and then it exports a separate table with size spectra for each sample These are tabulation delimited ASCII files They should be easy to read from any other software Microsoft Excel Matlab Python with Numpy Scipy Pandas Julia 12 6 Further work with training test sets Version 3 of Zoo PhytoImage introduces additional tools that add more flexibility in building training sets visualizing how vignettes are automatically classified and managing test sets These tools are accessible through the Analyze menu Add vignettes to training set allows to complete existing tra ining sets by adding more vignettes to them without breaking the training set structure Automatic classification of vignettes allows to select one sample and to represent the same fol
53. is not an entry It just tells it is a ZooImage1 file Author Image Who digitized the picture Hardware Image Device used to digitize the picture Software Image Acquisition software and version Type of image For instance trans 16bits gray 2400dpi ImageType Image means image acquired in transparency of 16bit gray scales and a resolution of 2400dpi Code Fraction The same fraction identifier as in the file name A B etc Minimum mesh size used to retrieve this fraction in um Use Min Fraction i 1 if none A Maximum mesh size used to retrieve this fraction in um Use Max Fraction 1 for none Part of the sample that was digitized If the picture contains SubPart Subsample only 10 of the organisms in your sample SubPart 0 1 for instance Method used to get the part volumetry Motoda Falsom SubMethod Subsample etc Part of the cell containing the plankton that was actually CellPart Subsample digitized Z00 PHYTOIMAGE 4 User MANUAL 20 If you did replicated images with the same protocol for that fraction how many replicates do you have Note ZooImage with average results among raplicates instead of summing them Replicates Subsample The volume of seawater that was sampled in m This is Volini Subsample required to calculate abundances and biomasses per m The precision on the sampled volume estimate in m This VolPrec Subsample will be used for error evaluation not implemented yet At the en
54. it is not convenient to keep all these items in separate files on disk In version 1 and 2 Zoo PhytoImage did compressed zipped these component in a single archive file with the ZID extension for Zoo PhytoImage Data This approach is simple and ensures readability of the data since the unzip program required to extract the components is widely available However unzipping the archive to access the vignettes is a Zoo PuytoImMacE 4 User MANUAL 5 slow operation This format prevents thus a fluid and fast display of the vignettes for best user interaction and experience Starting from version 3 Zoo PhytoImage now uses a custom binary format called ZIDB for Zoo PhytoImage DataBase This format is indeed a hash table followed by binary versions of the different components Fast C functions are used to access the different components for very fast retrieval of any vignettes the features or the metadata This format is a little bit less portable but is easily accessible from R and R itself is now widely available In term of disk storage the new ZIDB format is marginally usually around 5 less compressed So you need an little extra storage space too Of course a series of function have also be added to import data from the old ZID format and to convert back and forth between the two formats Zoo PuytroImace 4 User MANUAL 6 3 INSTALLATION 3 1 Hardware requirement Image analysis and automatic classification of im
55. les A short description for this station A short note concerning this station The complete label of the sample as in the file names A code for this sample The scs for that sample The series code to which that sample belongs The cruise code corresponding to the sample if any The station code The data of sampling in yyyy mm dd format The time of sampling in hh mm ss The time zone lag from GMT in x hours The latitude of sampling in x xx The longitude of sampling in x xx Precision of lat long radius in m Who collected this sample The type of gear used to collect the sample The opening area if collected with a net in m8 For a net only size of the mesh in um Minimum depth of sampling in m Maximum depth of sampling in m Volume of seawater sampled in m Precision of sampled volume in m8 Type of tow vertical horizontal oblique etc Speed during tow in m s Weather conditions during sampling Preservative used for instance buffered formaldehyde 4 Staining used if any Zoo PuytoImMacE 4 User MANUAL 63 Biovolume Temperature Salinity Chla Note Samples Samples Samples Samples Samples Samples Rough estimation of the biovolume after sedimentation in mm Temperature of the water at sampling in degree Celcius Salinity of sampled water in per thousands Chlorophyll alpha in the sampled water A short note ab
56. mory you can now create a ZIClass object that is an automatic classifier that learns how to recognize your zooplankton based on the examples you give in your manual training set Click on the seventh button on the ZooImage assistant toolbar The next dialog box appears It displays a warning message about the simplified learning phase and proposes a variety of machine learning algorithms to use Zoo PuyrolMmacE 4 User MANUAL 39 Zoolmage make classifier This is a simplified version of the classifiers where you just need to select one algorithm Warning Many algorithms have parameters to be fine tuned before efficient use and this must be done For each specific data set Here only default parameters that have proven efficient with zooplankton are applied automatically Some methods already work pretty well that way Learn using an algorithm linear discriminant analysis recursive partitioning tree CO k nearest neighbour C learning vector quantization neural network random Forest C support vector machine Choose the one you want to use Now we will use the simplest algorithm linear discriminant analysis select it and click ox The program asks then which ZITrain object he should use You have probably only one training set in memory the training object you just created Choose one ITrain obj Select it and click ox The program then asks for a name for the ZIClass
57. om right corresponds to cells where both identifications are the same This is thus the counting of correctly predicted items All cells outside of the diagonal depict disagreement in both classifications They are usually attributed to errors done by the automatic classifier starting form the hypothesis that there is no error in the manual training set To calculate and display the confusion matrix for your classifier click on the eigth button on the ZooImage assistant toolbar and select your ZIClass object in the dialog box You probably have only one so select it and click ox According to those analyses you could decide to rework the groups that are difficult to separate in your manual training set to reread it and train a new classifier with these optimized groups Other diagnostic tools are also accessible from the same dialog box in version gt 4 Experiment by yourself with it and discover the different diagnostic plots available here Z00 PHYTOIMAGE 4 User MANUAL 41 11 MANIPULATING ZOO PHYTOIMAGE OBJECTS You don t have of course to read manual training sets and train classifiers again and again each time you launch Zoo PhytoImage You can Save and restore existing objects The Objects menu provides functions to do so e Objects gt Load reloads one or several objects form a RData file The RData file is a binary format that is used by R to save its variables You can save several objects
58. on All the vignettes can be freely drag and dropped everywhere Thus you can rearrange the vignettes in order to performs required corrections For very long grids with tens or even hundreds of columns you can use a special yellow area on the left named Unclassified to temporary store items that you want to relocate in a distant position in the grid However you cannot leave items in that special area when you validate your work Zoo PuytoImMacE 4 User MANUAL 53 For all particles that you cannot recognize or that do not belong to the pre specified classes you have a special class other at the extreme right of the grid Once you have done with the validation of these vignettes click on the Validate button A report of the validation process done during that first step is displayed i m nate ENE 29 200441 OL Miep gt i l Nax gt gt 3 Done m BE ArMix 2009 04 29 300A4X_01_78 Step1 Suspects and error corrected at each iteration Of haces s gt It present a barplot with gray bars representing the proportion of suspect items in the fraction just validated During the first step no model is calculated yet so all items are considered as suspect A red bar at its right indicates the fraction of items that were incorrectly classified and that you just corrected In the present case it amounts at around 15 This is a very good indication of the overall error in that classification s
59. ory Update also comments of _raw inages zip Files Instructions should be clear By clicking OK you compute zidb files for your processed samples The option update also comments of _raw images zip files add zim data as comments to zipped image files if you selected that option in the process Since we did not zipped images we should uncheck that option now and click OK You are prompted for a directory where treated data are located give you working directory D image sample treatment 14 You reach easily a compression factor close to 100 or more starting with uncompressed 16bit TIFF images 6 times 120Mb of raw images that is 720Mb compresses to 4 10Mb in the corresponding zidb files Zoo PuytolmMace 4 User MANUAL 27 Rechercher un dossier Please choose a directory then select OK D inage sample treatment econum i gt images image sample Gl eas gt raw gt work cy BIO 2000 05 05 p 2 BIO 2000 05 08 p1i23 ao LA ire see sree j i a Zoo PhytoImage computes zidb files and issues a report at the end of the process For convenience it first quicky checks if all files are corrects Stay in front of the computer during checking Once it succeed you can take a coffee break during the process that can be long if you processed a lot of samples Make sure there is no error reported once the compression is done Zoolmage log Fichier Edition 200 Image log
60. out this sample You can add any additional measurement done on the sample here Z00 PHYTOIMAGE 4 USER MANUAL 64
61. pe Tous les fichiers ka The reasons why you have to select the zim file instead of the corresponding image are e Weare sure you have metadata associated with the image s e As explained here above you could have several images for the same sample fraction The plugin will process all images associated with the selected zim file not only one In the example we have only one image for each zim file but that feature is designed with FlowCAM or VPR images in mind You then have a dialog box with parameterization of your process Zoo PuytoIMacE 4 User MANUAL 23 foolmage1 Image Processor Selected item BIO 2000 05 05 pre A zim Process all items in this directory Read from directory Cem Parameters set default 0 75 10 Calibration set 600dpi Haematoxilyn Zip images Analyze particles Make vignettes Sharpen vignettes Show outlined objects e The name of the selected zim file is displayed e You can process all items in this directory all images that have associated zim files or only that one keep this checked now e You can optionally read images from a different directory This function is useful if you saved your large images on DVDs or external disks You just have to copy the small associated zim files in your process directory and you point to the directory that contains the images on your DVD leave this blank now e The parameters set drop down list allows you to select
62. ple If there are few of them it is a rare group It implies two consequences 1 the probability of false positive increases and 2 the class has more probabilities to contains no particles for that sample because that taxonomic group is absent there at that time So the probability to be suspect increases with the scarcity of particles classified in the same class e The information from the confusion matrix is used to determine which classes tend to be less good discriminated Again that information increases the probability of the corresponding particles to be suspect e Possibly biological information can be supplied too not from the menu dialog box but by calling correctError directly in the R console see its help page at correctError That biological information should indicate if a given class has chances or not to be found in that sample Say you know from the geographic location from the time of the year from the water temperature or simply from a quick inspection of the sample under the microscope that class A is very unlikely to be present and class B is certainly there Just indicate a low value say 0 01 to class A and a high value say 0 99 to class B Note that the numbers you provide are not necessarily restricted between o and 1 but the concept is easier to consider if you look at these weight like pseudo probabilities of occurrence of the class in your sample Zoo PhytoImage use the first set of pa
63. ptain Name s of the scientific coordinator s on board Name s of additional scientific staff on board Date of departure in yyyy mm dd Date of arrival at the final destination in yyyy mm dd Southmost latitude reached in x xx degree decimal Westmost longitude reached in x xx Northmost latitude reached in x xx Eastmost longitude reached in x xx The project to which this cruise belongs An optional URL pointing to a web page that further describes this cruise A short comment about this cruise A code for this station The name of location of this station The latitude of the station in x xx The longitude of the station in x xx The date at which sampling started at the station in yyy mm dd The date at which sampling was stopped if any in yyyy mm dd The frequency of sampling in no of samples per day The maximum depth at the station location in m Zoo PuyrolmacE 4 User Manuva 62 Description Note Label Code SCS Series Cruise Station Date Time TimeZone Latitude Longitude CoordsPrec Operator GearType OpeningArea MeshSize DepthMin DepthMax SampVol SampVolPrec TowType Speed Weather Preservative Staining Stations Stations Samples Samples Samples Samples Samples Samples Samples Samples Samples Samples Samples Samples Samples Samples Samples Samples Samples Samples Samples Samples Samples Samples Samples Samples Samp
64. r MANUAL 35 A final pass is required before you can use your training set you must rework or eliminate rare subgroups were you have too few items in them let s say less that 8 10 vignettes Two alternatives 1 Merge them with other subgroups making less detailed groups but with enough vignettes 2 Decide not to include these rare groups in the training set Keep them but move the directories to the top folder remember that this top folder contains all subgroups and vignettes that will be ignored in the classification ever forget that including rare groups in your training set will only have the consequence to reduce the total identification accuracy and the accuracy of other major groups due to missclassification of other items in these rare groups The only exceptional situation where you would like to keep a rare group is when you are specifically interested by tracking target rare organisms in your whole set of images When you rework your groups make sure you do not have also too many vignettes in the most abundant ones It is useless to have hundreds or thousands of items in one group If it is the case randomly eliminate vignettes you can create the same group under the _ top folder and move the vignettes there so that you keep them correctly classified but do not take them into account in the learning stage Consider that if you have more than 50 vignettes in a group you can begin to eliminate randomnly item
65. raining set kz DK Mom Taille Type Gestion des fichiers fo Dossier de Fichiers C alter Dossier de fichiers s C Append Chaeto Dossier de fichiers 5 eu ce dossier sur le CO Copepoda Dossier de Fichiers gt Gelatinous Dossier de Fichiers C Zooplankton other Dossier de Fichiers R BIG 2000 05 05 p72_dati R R Workspace R BIG 2000 05 08 p123_ dati R Wiorkspace Cr er un nouveau dossier E Partager ce dossier Autres emplacements Lj treatment Mes documents Lj Documents parbag s F Poste de travail J Favoris r seau D tails LnAView Bowie D Umage sanoit ae seti t ET 00 a cane Jeg F O mhp O aat 2 weronm 60 20000508 pt2 4_12h pe L oer 10 Sor Cueto O Cwpl teres LD Dopico ote a VAL 8 00 20005 C8 pl IIA An pg 300 odjeci J 1 les soloctad 1 05 kD EIO 20O Opi 23 A 22 bo 7O True Coo 145D 100 XnView is a free software for non commercial use It is both an image viewer manager and an image converter Here we only use its ability to work with thumbnails of images in directories and manage them We don t use all its features Depending how you organize XnView windows the browser has a tree of directories a thumbnail of images and a preview panel for the currently selected picture You can change XnView configuration in Tools gt Options If the directories tree is not visible select view gt Folder 15 Ifyou are in a private comp
66. re gt VisesScan 2 0 10 Drreowere Deotsey d ficteers im epeType trans Lobvits Gray 24003 LE Tm SOOM EE p72 p Zi6fo Deeg PEG BPE 200 406 083 11 23 A BO 14to Drage PEG a Fraction J 000 SOO pl 2 eon Lio oora Metadata oders z Mer lt 7 gt Maxz Sub sarnple subPart subMethod volanetry COEPart 1 Ge You are supposed to fill these data correctly Here is how you can use the metadata editor Itis a plain text editor Type your text as usual You don t have to save your changes When you close the window changes are automatically saved and Zoo0 PhytoImage switches to the next file If you want help about a given entry type an opening parenthesis just after the equal sign You got a tip with information about that entry Zoo Puyrolmace 4 User Manuva 18 fates erase BIO 2000 05 05 p 2 A zim 5c1 Siz Eg File Edit Search View Tools Options Language Help Il Image Author Author myname Who digitized this picture Software YueScan 8 0 10 ImageTypestrans 16bits gray 2400dpi Fraction Code 4 Min 1 Max 1 Subsample SubPart 1 SubMethod volumetry CellPart 1 00 Replicates 1 YolIni YolPrec gt You can also have a list of proposition for that entry Place the caret just after the equal sign and hit ctri t A list displays default entries This way of entering metadata should be preferred because it avoids typing errors BIO 20
67. re you have created or loaded a suitable classifier ZIClass object Typically you save your classifiers on disk in Rdata files So to retrieve one go to the menu ZooImage Load objects navigate to the folder where you store your classifier s and select the one you need tie ranea Sere Acwere ew HOt en ien of type erT Caroi Now that your classifier object is in memory select Validate classification in the Analyze menu Zoo PuyrolMmacE 4 User MANUAL D1 If there is no classifier found in memory an explicit message invites you to create or load one first Otherwise Zoo PhytoImage asks you now which one of all classifiers found in the current R session you want to use Z00 PuytoImacE 4 User MANUAL 52 Once it is done Zoo PhytoImage creates a web page that presents you a first set of by default 1 20th of the vignettes in the sample EAM 2009 04 79 30044_ 01 7p This page presents a first series of particles randomly selected in the sample as they are sorted automatically by the chosen classifier Each class is represented by one column in the page e g C compressus D brightwellii etc in the example All vignettes classified in one group are presented in the correponding column Moving the cursor on top of one vignette automatically triggers a floating window that dispalys the corresponding particle in full size view for inspecti
68. rticles as a training set to detect suspect items using all features peasured on these particle plus the additional variables described here above Several algorithms can be use but random forest is used by default So when you click Next Zoo PhytoImage presents you another subset of the particles in the sample But this time the subset is not randomly chosen but rather mainly selected in the suspect items As a consequence the proportion of error happens to be higher Thus your validation work is more efficient because you start to focus on the really problematic particles now Zoo PuytoImMacE 4 User MANUAL 55 ee ate 2000 04 78 2008 o enoa AE It is usually quite clear that this second set presents much more errors that the previous one and you will also notice that indeed you got also much more problematic particles hard to recognize particles cropped items blobs with strange forms etc Do not hesitate to use the other group to collect what you cannot place elsewhere but be consistent on what you do here Click Validate when you have done with this second step Zoo PuytoImacr 4 User MANUAL 56 BE ArMix 2009 04 29 300A4X_01_ 36 Step2 Suspects and error corrected at each iterabon Suspect acd corrected errot CR In the report the barplot has now a second series of gray red bars As you can see here the identification of suspect items is mildly efficient recall the training set
69. s down to 50 images per group Making a manual training set is a difficult and time consumming task You have an example training set installed with Zoo PhytoImage You can inspect it in XnView or even read it in Zoo PhytoImage if you like This example training set is located in the examples train subdirectory of your Zoo PhytoImage folder c Program Files ZooImage by default on Windows This training set was build using 29 samples thus more than the two available in you examples subdirectory Look at it to have an idea on how you should balance items in the different groups 9 2 Reading a manual training from disk m To read a training set from directories where vignettes were a manually classified use the menu entry Analyze gt Read training set the shortcut Ctrl T or click on the sixth button in the toolbar Once you are satisfied with your manual training set or after reworking it guided by the inspection of the confusion matrix see hereunder you have to read it in Zoo PhytoImage Click on the sixth button on the ZooImage assistant toolbar Zoo PuytoImMacE 4 User MANUAL 36 i The program asks you for the top folder where your manual training set is located Select now your directory that is D image sample treatment train Rechercher un dossier Select a Zoolmage training set base dir Ce image sample treatment training set O images image sample treatment O raw G work
70. ses on phytoplankton Zoo PuytoImMacE 4 User MANUAL 59 The confusion object can be adjusted for various prior probabilities abundances per groups using the prior function This allows you to visualize the impact of different sample composition in the false positive and false negative rates per groups Do not forget also all the R tools available to manipulate machine learning objects See the machine learning task views at http cran r project org web views MachineLearning html Finally chapter 12 in the Data mining applications with R book presents a collection of bibliographical references 64 most of them pointing on publications whose analyses were done using Zoo PhytoImage This is also an excellent source of inspiration showing in practice how Zoo PhytoImage can be used Zoo Puyrolmace 4 User Manuva 60 14 ANNEXES 14 1 Data and metadata in zis files Here is the explanation of the data and metadata in this description zis file Key Section Comment zu i This is not a key but just an identifiant telling it is a Zoo mage1 file Id Description The short identifiant of the series Name Description A longer name for this series Institution peseanuon he institution that owns the series i e where original iological material is stored if any Objective Description The goal s of this study Description Description A short description of the series Contact Description The name of a respon
71. sible person of this series Email Description The email address of the contact URL Des puon E E a piers pe sng to a Web page that further Note Description A short general comment about this series Code Series The code of a sub series Name Series The name of a sub series Project Series The project in which this sub series is included Institution Series The owner of the sub series as above for the series Country Series Country ies concerned by this sub series Location Series Place s concerned by this sub series Contact Series As above for the sub series Email Series Idem URL Series Idem Note Series Idem 21 Fill these metadata many of these are used by Zoo PhytoImage for its calculations Zoo Puyrolmace 4 User Manuva 61 Code ShipName ShipType ShipCallSign PortDeparture PortReturn Captain Coordinator Investigators Start End SouthmostLat WestmostLong NothmostLat EastmostLong Project URL Note Code Location Latitude Longitude Start End Frequency Depth Cruises Cruises Cruises Cruises Cruises Cruises Cruises Cruises Cruises Cruises Cruises Cruises Cruises Cruises Cruises Cruises Cruises Cruises Stations Stations Stations Stations Stations Stations Stations Stations A code for a cruise The name of the ship The type of the ship Immatriculation of the ship Self explicit Idem Name of the ca
72. t C dropping O econum O images image sample treatment O raw 4 work E training set fos imanar amnla m gt Annuler Select your D image sample treatment train directory finally the program asks you to select the zidb or zid files corresponding to the samples you want to use to build your manual training set they must be all located in the same directory Select now our two example samples BIO 2000 05 05 p72 zidand BIO 2000 05 08 p123 zid Select one or several Zid files Mes documents training set recents BIG 2000 05 05 p72 zid C BIG 2000 05 08 p123 zid E Bureau Mes documents Poste de travail amp Nom du fichier B10 2000 05 08 p123 2id BIO 2000 05 05 p Favoris r seau Fichiers de type lt oolmage data files zid Zoo PhytoImage creates required folders extract data about these samples dat1 Rdata files in the training set s root directory and places all corresponding vignettes in the subdirectory A log file indicates if there were errors creating these files and folders At the end of the process Zo0 PuytoIMacE 4 User MANUAL 33 Zoo PhytoImage starts XnView in the subdirectory If you inspect the files on your computer you should see something like this D image sampleVtreatment training set DER Fichier Edition Affichage Favoris utils Tr Pr c dente amp ei pa Rechercher Wey Dossiers EBk Adresse C D Aimage samplettreatmentit
73. t at the edges of the composite images On 64bit systems you don t have these limitations and should be able to analyze much larger pictures Start now ImageJ by click on the third button Click ok in the dialog box and the ZooImage assistant window is minimized and replaced by the equivalent ImageJ main window as 11 The current configuration of ImageJ installed with Zoo PhytoImage is to allocate a maximum of 900Mb to the program 12 You can change this value in ImageJ with the menu entry Edit Options Memory You have to restart ImageJ for the changes to take effect 13 With a different treatment one could process larger images but silhouette detection would be less accurate and there will be no background elimination Z00 PHYTOIMAGE 4 USER MANUAL 22 File Edit Image Process Analywe Plugins Window Help IBSIGISON lt INAHINIAL Qe Pima Freehand line selections Zoo PhytoImage plugins are collected together in the menu Plugins gt ZooImage For our images we have to select the Color24bits 600dpi plugin The plugin first asks you to select a zim file Do not select on image file here Open a zim file regu ders O f m E BIG 2000 05 05 p72 4 jpg EIG 2000 05 05 p7 2 4 zim Mes documents B10 2000 05 08 p123 A jpg r cents BIO 2000 05 08 p123 4 zim Bureau Mes documents Poste de travail Br Nom du fichier BIO 2000 05 05 p72 4 zim v Favoris r seau Fichiers de ty
74. ticles from rare groups were preferrably selected in the few first sets This ensures you a good prediction for those rare groups otherwise often problematic So with this in mind you can reasonably consider that the validation could end now and that you can trust the correction introduced by this partial validation further helped with the statistical correction by the suspect detection model Click the Done button Look now at the R Console You got the corrected abundance of particles in the different classes at it stand after the last step Moreover the results are saved in the lt sample gt _valid object You can further explore it and of course you can use it as definitive classification of this sample Zoo PuytoImMacE 4 User MANUAL 58 13 Use oF Zoo PHYTOIMAGE AT THE R COMMAND LINE A complete and detailed description of the use of zooimage functions inside the R Console is described in Chapter 12 of the following book Yanchang Zhao and Yonghua Cen Eds Data Mining Applications with R ISBN 978 0124115118 December 2013 Academic Press Elsevier We encourage the interested readers to download the accompanying files from http www sciviews org zooimage Data mining with R There is a fully commented R script and an example dataset that browses the features available at the command line Here is an outline of most important tools in additions to what you can already do using the graphical user interface
75. toolbox with a menu on top and a status bar on bottom It will guide you during the whole process Basically you just have to click on the buttons from left to right to run the various steps of your analysis A Zoo PhytoImage analysis is subdivided in three parts as it the toolbox For each part you have four buttons 2 Ris the statistical software amp environment on which ZooImage is based Z00 PHYTOIMAGE 4 USER MANUAL 10 foolmapel assistant Analyze Objects Apps Functions Utilities Help BARS DLS Eas Ready D fimage sample The three parts of the ZooImage process materialized by three times four buttons The last button shows the ZooImage user s manual e The first part deals with image importation and process Acquire images Start an external acquisition software Vuescan or any other program 2 FA port existing images Possibly convert the format of the images and or rename them If images are already in correct format this function just make sure they have suitable metadata associated i Process images Basically ImageJ is started You are supposed to used one of the ZooImage specific plugins in ImageJ to process your pictures 4 Make zid files Zid files stands for ZooImage Data files They contain all you need for the rest of the treatment i e images of each individual their measurements and the metadata Yet they store this information in a compressed wa
76. us dataset When we speak about serious datasets in the field of zooplankton image analysis it really means e Terabytes of raw images to process Since you can backup your raw images and ZooImage cares about storing highly compressed data in zidb files you can really process very large series containing thousands or even tens of thousands of samples with a simple PC You can store indeed all these tens of thousands zid files in a single hard disk of 200 300Gb e Almost unlimited number of images per sample and also possibly complex samples processes with replicates and with various separate fractions different dilutions or even different processes for each fraction Zoo PhytoImage will perform all the calculations averaging replicates adding data from the fractions after applying corrections for different dilutions and rescaling results to express them per square meter of seawater automatically e Almost unlimited number of objects in each samples the current limit is probably around a few hundreds of thousands items per sample that is the size of a matrix R can store in memory at once with a 2 4Gb RAM computer This is not really a limitation because a few thousands to a few tens of thousands of objects are enough to evaluate the composition of a single sample even for relatively rare taxa with 10 000 objects measured in a sample even rare taxa representing 1 of the sample composition will be represented by abo
77. ut 100 individuals Of course processing time is in proportion with the size of the series but Zoo PhytoImage proposes various mechanisms to recover after a fealure 16 The only limitation is currently the maximum allocatable memory of 1 6Gb in ImageJ under 16bit systems that limits the size of one image to 100 millions of pixels But 64bit systems currently available today overcome that limitation Otherwise Zoo PhytoImage allows an almost unlimited number of images per sample 17 Atypical zid file with 2000 3000 objects weights only about 5Mb 18 For instance using a Zooscan for the large fractions and a FlowCAM for the smaller ones Zoo PuytoImMacE 4 User MANUAL 43 to process a sample and the error is reported in the log file So it is possible to spot the error and to reprocess only the guilty sample s later on So OK it seems relatively easy to accumulated huge amount of data using Zoo PhytoImage But then how do we digest this huge quantity of information The third part of the analysis deals with the calculation of biologically meaningful statistics that summarize each sample abundances biomasses and size spectra total or per taxa Hence from the measurement of a couple of thousands objects in your images you summarize the information into a few tens of numbers for each sample All these numbers are then collected in a single table with one line per sample These tables are stored in ZIRes objects Zoo
78. ve the opportunity to manually classify vignettes in as many taxa as you want The hierarchy of the Folders and subfolders can be used to represent various levels of classification that the software will be able to use subsequently You must specify 1 a grouping scheme to start with 2 a base directory where ta locate the training set 3 a series of zid Files as source of vignettes Use the Following grouping scheme Basic Detailed very detailed Another config You have to select a config file That file will create the initial hierarchy of groups as a series of subdirectories in your training set folder You can choose Basic Detailed and Very detailed or select a different config file with a zic extension Choose now the Basic configuration and click ox nitial groups config files are customizable and you can save other ones everywhere on your hard disk Just respect their simple syntax and save them with a zic extension Basic zic Detailed zic and Very Detailed zic files are located in the subdirectory bin R R 2 2 0 library zooimage etc of the ZooImage root dir usually Cr Program files ZoolImage You now have to select the base empty directory where you want to install files and folders for your new manual training set Zoo PuytoIMacE 4 User MANUAL 32 Rechercher un dossier Please choose a directory then select OK De image sample treatment training se
79. y e The second part help you to make an automatic classifier optimized for your zooplankton series 1 Make a training set This function prepares a directory with a hierarchy of subdirectories representing your manual classification you can freely modify this structure at will and extract vignettes from the samples you want to use for making your manual training set You then have to manually classify them on screen by moving them to their respective directories with the mouse 2 Read training set Once you manually sorted the vignettes this function collect this information into ZooImage Statistics about you calssification number of vignettes in each group is the displayed 3 Make classifier Use a manual training to train an automatic classifier You have the choice of various algorithms You got some statistics at the end of the process to evaluate performances of your classifier cross validation 3 These particular images are called vignettes in ZooImage terminology 4 Ifyou started with uncompressed high resolution 16bit grayscale pictures in TIFF format you usually end up with zid files that weight about 100 times less than the original pictures Z00 PHYTOIMAGE 4 USER MANUAL 11 Oaze classifier Further analyses of your classifiers performances Currently only the confusion matrix showing differences between manual and automatic classification is calculated Other diagnostic tools will be added

- R

Contents

Download Pdf Manuals

Related Search

Related Contents