FuzzyUPWELL System v2.2 User Manual - centria
Contents
1. Ar specific colorbar scale E E Color Temperature adjustment pe Option 1 AP FCM Visualize result SST an Mode fomato Frontine type crap E FCM SST wi cluster borders Titt norte au Dee AP C2 Last cluster s contributi 2 Number of Clusters 6 Custer Extension sza y TOM South 539 x10 5 Threshold 01 x10 2 Filter value 6 Cloud Noise 532 Option2 Load g Re Apply Algorithm Figure 8 Upwelling front is detected using the Information Gain algorithm The Upwelling front minimum and maximum temperatures and mean cluster s temperatures are also shown 12 There are 3 modes for detecting the upwelling front 1 In Experimental mode the parameter values have been manually fixed such that the results were best matched with the analysis of the oceanographers These fixed parameter values works correctly in almost all images 2 In Information Gain mode the threshold value of each feature TDiff and Cluster Extension has been established using an entropy based attribute discretization procedure Nascimento and Franco 2009b 3 In Custom mode the user sets the parameter values The Parameters are TDiff South TDiff North Cluster Extension and Cloud Noise 1 TDiff South denotes the threshold value of relative temperature difference between consecutive cluster prototypes in the southern region below Cabo Espichel or 38 42 N latitude This threshold is used to compute the transition cluster2. Last clusters contributi Apply Number of Custers 6 Cluster Extension 524 9 TDI South 5 39 x10 5 Threshold 04 x102 Fiter value 6 Cloud Noise s32 Apply Reset i Load Clustering Result Apply Algorithm Onion Load Custerng Resut Agph Algorithm view First 3 Frontines EEE Figure 7 Visualizing the first 3 fuzzy borders along with crisp boundaries The range of mean temperatures of the fuzzy borders is displayed on the right hand side in Frontline Temperature panel 11 Third Component Upwelling Frontline Detection The Upwelling Frontline Detection panel is used to set parameters for identifying the upwelling frontlines A default parameter setting is available which has been tested for the years on 1998 and 1999 This current parameter value computes the upwelling front with high accuracy for SST images of these two years Naturally the user has the option to change these default values BD Fast mer Feature Panel ond New se Load Felder Save Results Extract image UpwelingArea Upweling Front Save info Upweling Front Original SST Image 19980802 14 26 Computed Result 2 r Mean Temperature C Cluster 1 1485 Cluster 2 16 55 D Custer 3 1812 Cluster 4 19 58 Cluster 5 20 52 Cluster 6 23 8 1t Frontline Temperatures C 192 a m
3. Result E r Mean Temperature C Cluster 1 14 85 Cluster 2 16 55 D Cluster 1812 Cluster 4 19 58 Cluster 5 20 52 Cluster 6 23 8 14 Frontline Temperatures C T 2 Ir EB E 1 specific colorbar scale has Calor Temperature adjustment Option 1 o AP FCM Visualize result Vpmeling Front Detection fede information Frontinetype crisp X F i SST w cluster borders a TDiff North 447 x10 5 AP C2 Last cluster s contributi Number of Clusters 6 Custer Extension sza Diff South 5 39 x 10 5 Threshold 0 1 x10 2 Fitter value 6 Cloud Noise 532 Option 2 Load Clust o Resut Figure 1 A screenshot of FuzzyUPWELL tool First Component Initial Settings WEST tele tes Load New SST map Load Folder Original SST Image 19980802 14 26 a Option 2 Load Clustering Result M Last cluster s contributi AP FC AP C2 Threshold x10 2 Figure 2 Setting the initial parameters for a loaded image and applying a clustering algorithm 1 2 The dataset can be loaded using the push button Load New SST Map The loaded image is displayed in the left hand axis A color bar corresponding to the loaded SST image is displayed on its right The color bar temperature range is set to default 12 C to 24 C By pressing the Specific colorbar scale button the color bar temperature range w
4. each cluster The colors in front of the clusters numbers identify the mean temperature of the corresponding cluster and correspond to the color bar in the Result Axis 2 The Frontline Temperature panel displays the minimum and maximum temperatures of the fuzzy frontlines between neighboring clusters Once again the color assigned to the temperature values correspond to the color by which the borders are visualized on the Result Axis 15 3 The Feature Panel consists on several options to visualize the clustering results and extract useful information Specifically i Save Results button which saves the segmentation result of a clustering algorithm to a file ii Extract Image button which provides a screenshot of the image such that it can be analyzed separately iii Upwelling Area button which shows the upwelling structure retrieved from the segmented image and after the Front detection iv Upwelling Front button which allows to visualize the upwelling front in the right axis individually v Save Info Upwelling Front button which saves into a file the spacial and geographical coordinates of the pixels in the upwelling front as well as their distance to the coast line Bret e e 3 i EUR Feature Panel oad f id se Load Fokker Save Results Extract Image e Upweling Front Save info Upweling Front Original SST Image 19980802 14 26 Computed Res
5. of Clusters 8 Threshold 0 1 x10 2 Option2 LoadClusterigResut Apply Algorithm Figure 4 Visualizing a fuzzy membership map after applying a fuzzy clustering algorithm to an SST image 2 Cluster Borders The cluster borders can be visualized using different criteria 2 1 Crisp criteria displays the original SST image with cluster borders marked on it sia els ts inso ER Feature Panel jew ap Load Folder Save Resuts Extractimage UpwelingArea Upweling Front Save Info Upweling Front_ Original SST Image 19980802 14 26 Computed Result 2 2 r Mean Temperature C Cluster 1 1485 Cluster2 16 55 D Cluster3 1812 Cluster 4 19 58 Cluster 5 20 52 Cluster amp 23 8 12 f 1 specific colorbar scale Color Temperature adjustment Option 1 E 5 UpwielingEcontnetectin Mode Information G Frontline type crisp AP FCM X isualize result SST w cluster borders 1 AP C2 Last cluster s contributi v Apply Number of Clusters 6 cluster Extension 52 4 TDiff South 5 39 x 105 Threshold 01 x10 2 Fitter value 6 Cloud Noise 532 Option2 Load Clustering Resutt Apply Algorithm Figure 5 Visualizing Crisp cluster frontlines 2 2 The other criteria are the uncertainty measures which enable to visualize the classification uncertainty with which pixels are assigned to clusters after defuzzifi
6. of the upwelling region in the south This parameter setting is specifically for detecting upwelling in Coastal Portugal For other regions these parameters could be manually set by the user using the Custom mode ii TDiff North denotes the threshold value of relative temperature difference between consecutive cluster prototypes in the northern region above Cabo Espichel or 38 42 N latitude This threshold is used to compute the transition cluster of the upwelling region in the north This parameter setting is specifically for detecting upwelling in Coastal Portugal For other images these parameters could be manually set by the user iii Cluster Extension denotes the cardinality of the pixels belonging to the upwelling region This parameter is used to set the upper limit on the percentage of the area that can be covered by upwelling iv Cloud Noise denotes the number of pixels in the upwelling region neighboring the clouds This parameter sets the upper limit of the number of pixels in the upwelling region that can border the clouds If the number of pixels 13 surrounding clouds is greater than Cloud Noise the upwelling transition cluster number is decreased by one After delineating the upwelling total area it can be visualized and analyzed using either Feature Panel push buttons or using View Upwelling Borders and View Upwelling Front options under Fuzzy Ignorance Uncertainty Fuzzy Exaggeration Uncertaint
7. where the datasets are stored using the Load Folder button in the top left corner and a directory where the clustering results are to be saved The default directory is recommended to gain full advantage of the Apply button EJ FuzzyUpwell Lo es Load New SST Map Option 1 AP FCM X AP C2 Last cluster s contributi v Threshold x10 2 Figure 3 How to apply a specific clustering algorithm to a set of SST images Second Component Visualization of Segmentation Result Once the SST image is segmented and the result is available it can be visualized and analyzed in following ways 1 Fuzzy Membership Map This option displays the fuzzy membership map assigned to a clustering segmentation These degrees of memberships are visualized by assigning a color value to the data pixels based on their maximum membership value Nascimento and Franco 20092 Blei aa Feature Panel T Lond Now SST Map oad Coker save Results Extractimage UpwelingArea UpwelingFront SaveinfoUpwelingFront Original SST Image 19980802 14 26 Computed Result Mean Temperature C Cluster 1 14 85 Cluster 2 16 55 Cluster 3 18 12 Cluster 4 19 58 Cluster 5 20 52 Cluster6 238 Owl Dido podia DORMIDO SOAR OeKOD GMA EDI ARMII Color Temperature adjustment E E pg ITI Le AP FCM Visualize resul Membership Map AP C2 Last cluster s contributi v Apply Number
8. 9 LNCS 5788 Springer Verlag pp 543 553 Burgos Spain e P Franco 2009 MSc Thesis Fuzzy clustering n o supervisionado na detec o autom tica de regi es de upwelling a partir de mapas de temperatura da superf cie oce nica Faculdade de Ci ncias e Tecnologia Universidade Nova de Lisboa in Portuguese 18
9. FuzzyUPWELL System v2 2 Computacional system for the automatic detection of upwelling from sea surface temperature SST images via Fuzzy Clustering User Manual 1 1 2 Yashu Chamber Susana Nascimento cerea Centre of Artificial Intelligence Universidade Nova de Lisboa Departamento de Inform tica Faculdade de Ci ncias e Tecnologia Universidade Nova de Lisboa 2011 Contents Fuzzy UPWELL User Manual itt tte eene e Eee petivit ra eta 3 First Component Initial Settings essen ren rennen nnne 4 Second Component Visualization of Segmentation Result sse 7 Third Component Upwelling Frontline Detection eese 12 Complementary Functionalities sssri siiras nii anaoiit emen nennen 15 ze E 18 FuzzyUPWELL User Manual The FuzzyUPWELL tool has three main components 1 The first component corresponds to dataset loading clustering algorithm selection parameter setting and algorithm execution 2 The second component provides functionalities on visualization of clustering results 3 The third component corresponds to the upwelling frontlines detection their visualization and analysis Bl FuzzyUpwell Ee er ps m pM Feature Panel lew SST Map Load Folder Gave Resulis Extractimage UpwelingArea Upweling Front Save info Upweling Front Original SST Image 19980802 14 26 Computed
10. cation these are 1 Fuzzy Ignorance Uncertainty ii Fuzzy Exaggeration Uncertainty iii Fuzzy Confusion Ratio iv Fuzzy Confusion Difference Bl FuzzyUpwell Es Feature Panel Load New SST Map Load Folder Save Resutts Extract image UpwelingArea Upweling Front _ Save info Upweling Front Computed Result Original SST Image 19980802 14 26 24 Mean Temperature C Cluster 1 1485 Cluster 2 16 55 J Cluster3 1812 Cluster 4 19 58 Cluster 5 20 52 Cluster 6 238 Fuzzy Exaggeration Uncertainty Fuzzy Confusion Ratio Fuzzy Confusion Diff f Specific colorbar scale Color Temperature adjustment Option 1 T T Upwelling Front Detection E Mode Information G Frontline type Fuzzy ignorance Uncert x Visualize result n AP FCM SST wi cluster borders armies TUE ems AP C2 Last cluster s contributi w Number of Clusters 6 Cluster Extension 524 o i South 539 x1055 T ne e Q3 1 ds ES Crisp Frontlines Threshold 04 x10 2 Filter value amp Cloud Noise 532 Apply Reset 7 Opaqueness 03 b Oplion2 LosdCusterngResut Apply Algorithm View AllFrontines Se Figure 6 Visualizing the fuzzy frontlines using the ignorance uncertainty measure ii Fuzzy Ignorance Uncertainty is an entropy measure which is used to measure ignorance uncertainty a
11. he second highest membership The lower this value the lower the fuzziness of data entity The fuzzy boundaries obtained by the uncertainty measures described above can be visualized by distinct levels of opaqueness on the segmented SST image by using different parameters The Alpha value slider enables to change the lower threshold for identifying the frontline pixels the Opaqueness slider enables to set the degree of opaqueness of the fuzzy frontline pixels Separate frontlines can be viewed using the options pa View All frontlines 2 View First 3 frontlines 3 View a specific frontline 4 View Upwelling frontlines or 5 View Upwelling Front The last two options 4 amp 5 become available only after the upwelling front detection routine is run as described in the next component 10 Feature Panel Load New SST Map Load Folder r l f l J Save Resuts Extractimage Upwelling Area UpwellingFront Save info Upweling Front Original SST Image 19980802 14 26 Computed Result Ea 2 Mean Temperature C Cluster1 1485 Cluster2 1855 B 7 Custera 1812 Custer 4 1958 5 Clusters 2052 i owes 238 WB 16 u Frontline Temperatures C o Color Temperature adjustment Option 4 Ra dl m O Et m AP FCM ualize resuk SST w cluster borders roer North za xtoss AP c2
12. ill be set according to the maximum and minimum temperature values of the loaded dataset A slider below this axis changes the color palette of the loaded image Once the SST dataset is loaded the user is either required to select a clustering algorithm and set the associated parameters using Option 1 panel or to load a result file if available using Load Result File push button This file contains the fuzzy segmentation of the clustering algorithm applied to the loaded SST image which was saved for this dataset during a previous run This functionality is useful since loading the result file is much faster than applying the clustering algorithm to the SST image 3 The clustering algorithms currently present in the FuzzyUPWELL tool are i Anomalous Pattern Fuzzy Clustering AP FCM ii Fuzzy C means FCM and Gi Histogram Thresholding i The Anomalous Patter Fuzzy c Means AP FCM is the novel fuzzy clustering algorithm described in the paper Nascimento and Franco 2009b The user has to select a termination criteria for this algorithm The available options are 1 AP C1 which ensures that the clustering terminates only when all data points have been clustered 2 AP C2 which terminates the clustering when the contribution to the data scatter of the last cluster obtained becomes smaller than a pre defined threshold An empirically tested threshold value is already set however the user has the option to change to a diffe
13. pe Crisp Upwelling Front Detection Y Visualize resul ARFCM v isualize result SST w cluster borders mbiff North 447 x10 5 AP C2 Last cluster s contributi x Number of Clusters 5 Cluster Extension sza itt Southe 530 x10 5 Threshold o1 xto2 Fitervalue 6 Cloud Noise 532 Option2 Load Clustering Result Apply Algorithm Figure 11 After pressing the Upwelling Front button the user visualizes the upwelling front boundary apart from the remaining cluster frontlines The Restore button allows the user to view the previous image 17 Bibliography e Susana Nascimento Pedro Franco F tima Sousa Joaquim Dias Filipe Neves 2012 Automated computational delimitation of SST upwelling areas using fuzzy clustering Computers amp Geosciences Volume 43 pp 207 216 Elsevier June 2012 http dx doi org 10 1016 j cageo 2011 10 025 e S Nascimento P Franco 2009b Unsupervised Fuzzy Clustering for the Segmentation and Annotation of Upwelling Regions in Sea Surface Temperature Images in J Gama eds Discovery Science LNCS 5808 Springer Verlag Vol 5808 2009 Pag 212 226 Porto Portugal October 2009 e S Nascimento P Franco 2009a Segmentation of Upwelling Regions in Sea Surface Temperature Images via Unsupervised Fuzzy Clustering in H Yin and E Corchado Eds Proc of the Intelligent Data Engineering and Automated Learning IDEAL 200
14. rent value 3 AP C3 which halts clustering when number of clusters has reached a pre defined maximum value ii The Fuzzy c Means FCM is the second clustering algorithm Dunn 1973 Bezdek 1981 For this algorithm the user has to pre specify the number of clusters to be found iii The Histogram Thresholding Tobias and Seara 2002 is the third clustering algorithm On choosing this algorithm the user also has to pre specify the number of clusters to be found 4 After an algorithm is selected the clustering of loaded SST image can be started using the following push buttons a Apply button recommended which starts by searching for a possible existing clustering result on the default directory according to the algorithm selected if present in the default directory it will load the result which is useful since it is faster than applying the algorithm If not the tool will execute the chosen clustering algorithm saving the clustering result in the default directory b Apply Algorithm button applies the chosen algorithm to the SST dataset independently of whether the clustering result is present or not in the default directory Alternatively if there is a large number of SST images in the same directory and the user wishes to apply a specific clustering algorithm to all at once and interruptedly the user can choose the Apply for all Images button In this case the user selects a clustering algorithm a directory
15. ssociated with the defuzzification process assigned to each pixel of an image This measure will be higher for values where the memberships are highly dispersed for all clusters such as membership values of 0 3 0 37 0 33 and lower for memberships where the entity is highly associated with a single cluster such as 0 1 0 84 0 06 iii Fuzzy Exaggeration Uncertainty is a measure associated with the hardening of a classification It means that this is the uncertainty associated with the maximum membership of each entity This way the measure will be higher for entities with lower maximum membership values exaggerating the fuzziness of segmentation Both of the above measures lie between O and 1 The O value means that each entity has full membership to the cluster it is assigned to i e zero uncertainty The 1 value means that each entity has an equal degree of membership to all K clusters i e complete uncertainty iv Fuzzy Confusion Ratio is the ratio between the second highest membership and the highest membership for each pixel For example if the highest membership of a pixel is to cluster 1 with value 0 823 and the second highest membership of the same pixel is to cluster 2 with value 0 156 then the Fuzzy Confusion Ratio for that pixel would be 0 1560 0 823 0 1896 The lower this value for a pixel the lower it s fuzzy nature v Fuzzy Confusion Difference is equivalent to 1 Difference between the highest membership and t
16. ult a E 24 cll 4 z z aL 4 Mean Temperature C Cluster1 1485 E t 4 Cluster 2 1655 z m Custer 1812 5 wp J Custer4 1958 Clusters 2052 W Cluster 6 la m Q custers ns HN mL 4 16 16 sab 4 wt 4 e u Frontline Temperatures C al 4 gt S 1 1 1 i i i f 1 f m wm a am o m x o a s T sf E gt specific colorbar scale EB Colar Temperature adjustment E Option 1 T AP FCM Visualize result pd inrer Mode information G Frontline type Crisp SST wi cluster borders TNT axis AP C2 Last cluster s contributi v Apply Number of Clusters 8 Custer Extension Deze a Of Sousa x 10 CORPS Threshold 0 1 x10 2 Fitervalue 6 Cloud Noise 532 Apply Reset Option 2 Load Clustering Result Apply Algorithm Figure 10 After pressing the Upwelling Area button the user visualizes the upwelling area The Restore STT Image allows the user to view to the previous image 16 Bl FuzyUpwell e e js mere errem Feature Panel w SST Map Load Folder Gave Results Extractimage Upweling Area Emm Save Info Upweling Front Original SST Image 19980802 14 26 Computed Result D E E Mean Temperature C Cluster1 1485 Bg Cluster 2 16 55 3 Cluster 3 18 12 Clster4 1958 Custers 2052 W Cuser amp 238 W it Frontline Temperatures C 18 5 Mode information G v Frontlinety
17. y Fuzzy Confusion Ratio Fuzzy Confusion Diff gt Border Type 14 Complementary Functionalities The tool provides a set of additional functionalities which are B Fuzzypwel ee fes Loaanew ssTuap a Feature Panel Lead New SST Uee Lond Folder Save Results CExtractimage UpwelingArea Restore Save info Upweling Front Original SST Image 19980802 14 26 Specific colorbar scale Color Temperature adjustment 2 Cluster 1 Cluster 2 m Cluster 3 Cluster 4 Cluster 5 Cluster 6 ETE Option 1 z S F Upwelling Front Detection information AP FCM isuelzeresut SST wi cluster borders pitt north an ees AP C2 Last cluster s contributi v Apply Number of Clusters 8 Case Exton IU sou ESSB REPT S D apee 105 Threshold 0 1 x10 2 Filter value 6 Cloud Noise 532 E app Reset Option 2 Load Clustering Resut Apply Algorithm 2s 0 Opaqueness 03 View Upweling Front v Mean Temperature C 1485 16 55 1812 19 58 20 52 238 14 Frontine Temperatures C View Crisp Frontlines 1 Figure 9 Visualizing fuzzy upwelling front boundary Upwelling front minimum and maximum temperatures and mean cluster temperatures are also shown Useful options for manipulating the image is also highlighted in Feature Panel 1 The Mean Temperature panel which displays the mean temperature of
Download Pdf Manuals
Related Search