Tarabalka_IGARSS.pdf

354 views

Published on

Published in: Technology, Education
0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total views
354
On SlideShare
0
From Embeds
0
Number of Embeds
12
Actions
Shares
0
Downloads
6
Comments
0
Likes
0
Embeds 0
No embeds

No notes for slide

Tarabalka_IGARSS.pdf

  1. 1. Best Merge Region Growingwith Integrated Probabilistic Classification for Hyperspectral Imagery Yuliya Tarabalka and James C. Tilton NASA Goddard Space Flight Center, Mail Code 606.3, Greenbelt, MD 20771, USA e-mail: yuliya.tarabalka@nasa.gov July 28, 2011
  2. 2. Introduction Proposed spectral-spatial classification scheme Conclusions and perspectivesOutline 1 Introduction 2 Proposed spectral-spatial classification scheme 3 Conclusions and perspectives Yuliya Tarabalka and James C. Tilton (yuliya.tarabalka@nasa.gov) Best merge region growing with integrated classification for HS data 2
  3. 3. Introduction Proposed spectral-spatial classification scheme Conclusions and perspectivesHyperspectral image Every pixel contains a detailed spectrum (>100 spectral bands) + More information per pixel → increasing capability to distinguish objects − Dimensionality increases → image analysis becomes more complex ⇓ Advanced algorithms are required! λ pixel vector xi samples x1 x2 x3 xi xi x2i Tens or xn hundreds of bands lines wavelength λ Yuliya Tarabalka and James C. Tilton (yuliya.tarabalka@nasa.gov) Best merge region growing with integrated classification for HS data 3
  4. 4. Introduction Proposed spectral-spatial classification scheme Conclusions and perspectivesSupervised classification problem AVIRIS image Ground-truth Task Spatial resolution: 20m/pixSpectral resolution: 200 bands data 16 classes: corn-no till, corn-min till, corn, soybeans-no till, soybeans-min till, soybeans-clean till, alfalfa, grass/pasture, grass/trees, grass/pasture-mowed, hay-windrowed, oats, wheat, woods, bldg-grass-tree-drives, stone-steel towers Yuliya Tarabalka and James C. Tilton (yuliya.tarabalka@nasa.gov) Best merge region growing with integrated classification for HS data 4
  5. 5. Introduction Proposed spectral-spatial classification scheme Conclusions and perspectivesClassification approaches Only spectral information Pixelwise approach Spectrum of each pixel is analyzed SVM and kernel-based methods → good classification accuracies Yuliya Tarabalka and James C. Tilton (yuliya.tarabalka@nasa.gov) Best merge region growing with integrated classification for HS data 5
  6. 6. Introduction Proposed spectral-spatial classification scheme Conclusions and perspectivesClassification approaches Only spectral information Pixelwise approach Spectrum of each pixel is analyzed SVM and kernel-based methods → good classification accuracies Spectral + spatial information Info about spatial structures is included Because neighboring pixels are related How to extract spatial information? How to combine spectral and spatial information? Yuliya Tarabalka and James C. Tilton (yuliya.tarabalka@nasa.gov) Best merge region growing with integrated classification for HS data 5
  7. 7. Introduction Proposed spectral-spatial classification scheme Conclusions and perspectivesOur previous research Segment a hyperspectral image into homogeneous regions Each region = adaptive neighborhood for all the pixels within the region Spectral info + segmentation map → classify image 1 1 1 1 1 1 1 1 1 2 2 2 Segmentation 1 1 2 2 2 2 map (3 spatial regions) 1 1 2 2 2 2 1 1 3 3 3 2 1 3 3 3 3 3 3 3 3 3 3 3 Combination of Result of Pixelwise segmentation and spectral-spatial classification map pixelwise classification (dark blue, white classification results (classification and light grey (majority vote within map after majority classes) 3 spatial regions) vote) Yuliya Tarabalka and James C. Tilton (yuliya.tarabalka@nasa.gov) Best merge region growing with integrated classification for HS data 6
  8. 8. Introduction Proposed spectral-spatial classification scheme Conclusions and perspectivesOur previous research Segment a hyperspectral image into homogeneous regions Each region = adaptive neighborhood for all the pixels within the region Spectral info + segmentation map → classify image 1 1 1 1 1 1 1 1 1 2 2 2 Segmentation 1 1 2 2 2 2 map (3 spatial regions) 1 1 2 2 2 2 1 1 3 3 3 2 1 3 3 3 3 3 3 3 3 3 3 3 Combination of Result of Pixelwise segmentation and spectral-spatial classification map pixelwise classification (dark blue, white classification results (classification and light grey (majority vote within map after majority classes) 3 spatial regions) vote) Unsupervised segmentation: dependence on the chosen measure of homogeneity Yuliya Tarabalka and James C. Tilton (yuliya.tarabalka@nasa.gov) Best merge region growing with integrated classification for HS data 6
  9. 9. Introduction Proposed spectral-spatial classification scheme Conclusions and perspectives Our previous research: Marker-controlled segmentationProbabilistic pixelwise Markers = the Marker-SVM classification most reliably controlled region classified pixels growing Classification map Probability map ⇒ ⇒ Yuliya Tarabalka and James C. Tilton (yuliya.tarabalka@nasa.gov) Best merge region growing with integrated classification for HS data 7
  10. 10. Introduction Proposed spectral-spatial classification scheme Conclusions and perspectives Our previous research: Marker-controlled segmentationProbabilistic pixelwise Markers = the Marker-SVM classification most reliably controlled region classified pixels growing Classification map Probability map ⇒ ⇒ Drawback: strong dependence on the performance of the selected probabilistic classifier Yuliya Tarabalka and James C. Tilton (yuliya.tarabalka@nasa.gov) Best merge region growing with integrated classification for HS data 7
  11. 11. Introduction Proposed spectral-spatial classification scheme Conclusions and perspectivesObjective Perform segmentation and classification concurrently → best merge region growing with integrated classification   ⇑ ⇑ Dissimilarity criterion? Convergence criterion? Yuliya Tarabalka and James C. Tilton (yuliya.tarabalka@nasa.gov) Best merge region growing with integrated classification for HS data 8
  12. 12. Introduction Proposed spectral-spatial classification scheme Conclusions and perspectivesInput Hyperspectral image Preliminary probabilistic classification Each pixel = B-band hyperspectral image one region X = {xj ∈ RB , j = 1, 2, ..., n} While not converged B ∼ 100 Find min(DC) between all pairs of spatially adjacent regions (SAR) Merge all pairs of SAR with DC = min(DC). Classify new regions Spectral-spatial classification map Yuliya Tarabalka and James C. Tilton (yuliya.tarabalka@nasa.gov) Best merge region growing with integrated classification for HS data 9
  13. 13. Introduction Proposed spectral-spatial classification scheme Conclusions and perspectivesPreliminary probabilistic classification Kernel-based SVM classifier* → well suited for Hyperspectral image hyperspectral images Preliminary probabilistic classification Output: Each pixel = one region • classification map • for each pixel xj : While not converged L = {Lj , j = 1, ..., n} a vector of K class Find min(DC) between all pairs of spatially probabilities adjacent regions (SAR) Merge all pairs of SAR {P (Lj = k|xj ), with DC = min(DC). Classify new regions k = 1, ..., K} Spectral-spatial classification map *C. Chang and C. Lin, "LIBSVM: A library for Support Vector Machines," Software available at http://www.csie.ntu.edu.tw/∼cjlin/libsvm, 2011. Yuliya Tarabalka and James C. Tilton (yuliya.tarabalka@nasa.gov) Best merge region growing with integrated classification for HS data 10
  14. 14. Introduction Proposed spectral-spatial classification scheme Conclusions and perspectivesHierarchical step-wise optimization with classification 1 Each pixel xi = one region Ri Hyperspectral image preliminary class label L(Ri ) Preliminary class probabilities probabilistic classification {Pk (Ri ) = P (L(Ri ) = k|Ri ), k = 1, ..., K} Each pixel = one region While not converged  1  2  3 Find min(DC) between all pairs of spatially adjacent regions (SAR)  4  5  6 Merge all pairs of SAR with DC = min(DC). Classify new regions  7  8  9 Spectral-spatial  10  11  12 classification map   Yuliya Tarabalka and James C. Tilton (yuliya.tarabalka@nasa.gov) Best merge region growing with integrated classification for HS data 11
  15. 15. Introduction Proposed spectral-spatial classification scheme Conclusions and perspectivesHierarchical step-wise optimization with classification 2 Calculate Dissimilarity Criterion (DC) between Hyperspectral image spatially adjacent regions Preliminary DC = function of region statistical, geometrical probabilistic classification and classification features Each pixel = one region  1  ?  2  3 While not converged Find min(DC) between all pairs of spatially  4  5  6 adjacent regions (SAR) Merge all pairs of SAR with DC = min(DC). Classify new regions  7  8  9   Spectral-spatial classification map  10  11  ?  12 Yuliya Tarabalka and James C. Tilton (yuliya.tarabalka@nasa.gov) Best merge region growing with integrated classification for HS data 12
  16. 16. Introduction Proposed spectral-spatial classification scheme Conclusions and perspectivesHierarchical step-wise optimization with classification 2 Calculate Dissimilarity Criterion between adjacent regions: Compute Spectral Angle Mapper Compute SAM between region mean vectors SAM(ui, uj) between the region mean vectors ui and uj no yes L(Ri) = L(Rj) = k’ ui · uj SAM(ui , uj ) = arccos( ) ui 2 uj 2 no (card(Ri) > M) & yes DC = (2 – max(Pk’(Ri), (card(Rj) > M) Pk’(Rj)))SAM(ui, uj) If adjacent regions have equal class DC = ∞ labels → they belong more likely to DC = (2 – min(PL(Rj)(Ri), the same region: PL(Ri)(Rj)))SAM(ui, uj) DC = (2−max(Pk (Ri ), Pk (Rj ))) · SAM(ui , uj ) DC If two large regions are assigned to different classes → they cannot be merged together Yuliya Tarabalka and James C. Tilton (yuliya.tarabalka@nasa.gov) Best merge region growing with integrated classification for HS data 13
  17. 17. Introduction Proposed spectral-spatial classification scheme Conclusions and perspectivesHierarchical step-wise optimization with classification 2 Calculate Dissimilarity Criterion between adjacent regions: Compute Spectral Angle Mapper Compute SAM between region mean vectors SAM(ui, uj) between the region mean vectors ui and uj no yes L(Ri) = L(Rj) = k’ ui · uj SAM(ui , uj ) = arccos( ) ui 2 uj 2 no (card(Ri) > M) & yes DC = (2 – max(Pk’(Ri), (card(Rj) > M) Pk’(Rj)))SAM(ui, uj) If adjacent regions have equal class DC = ∞ labels → they belong more likely to DC = (2 – min(PL(Rj)(Ri), the same region: PL(Ri)(Rj)))SAM(ui, uj) DC = (2−max(Pk (Ri ), Pk (Rj ))) · SAM(ui , uj ) DC If two large regions are assigned to different classes → they cannot be merged together Yuliya Tarabalka and James C. Tilton (yuliya.tarabalka@nasa.gov) Best merge region growing with integrated classification for HS data 13
  18. 18. Introduction Proposed spectral-spatial classification scheme Conclusions and perspectivesHierarchical step-wise optimization with classification 2 Calculate Dissimilarity Criterion between adjacent regions: Compute Spectral Angle Mapper Compute SAM between region mean vectors SAM(ui, uj) between the region mean vectors ui and uj no yes L(Ri) = L(Rj) = k’ ui · uj SAM(ui , uj ) = arccos( ) ui 2 uj 2 no (card(Ri) > M) & yes DC = (2 – max(Pk’(Ri), (card(Rj) > M) Pk’(Rj)))SAM(ui, uj) If adjacent regions have equal class DC = ∞ labels → they belong more likely to DC = (2 – min(PL(Rj)(Ri), the same region: PL(Ri)(Rj)))SAM(ui, uj) DC = (2−max(Pk (Ri ), Pk (Rj ))) · SAM(ui , uj ) DC If two large regions are assigned to different classes → they cannot be merged together Yuliya Tarabalka and James C. Tilton (yuliya.tarabalka@nasa.gov) Best merge region growing with integrated classification for HS data 13
  19. 19. Introduction Proposed spectral-spatial classification scheme Conclusions and perspectivesHierarchical step-wise optimization with classification 2 Calculate Dissimilarity Criterion between adjacent regions: Compute Spectral Angle Mapper Compute SAM between region mean vectors SAM(ui, uj) between the region mean vectors ui and uj no yes L(Ri) = L(Rj) = k’ ui · uj SAM(ui , uj ) = arccos( ) ui 2 uj 2 no (card(Ri) > M) & yes DC = (2 – max(Pk’(Ri), (card(Rj) > M) Pk’(Rj)))SAM(ui, uj) If adjacent regions have equal class DC = ∞ labels → they belong more likely to DC = (2 – min(PL(Rj)(Ri), the same region: PL(Ri)(Rj)))SAM(ui, uj) DC = (2−max(Pk (Ri ), Pk (Rj ))) · SAM(ui , uj ) DC If two large regions are assigned to different classes → they cannot be merged together Yuliya Tarabalka and James C. Tilton (yuliya.tarabalka@nasa.gov) Best merge region growing with integrated classification for HS data 13
  20. 20. Introduction Proposed spectral-spatial classification scheme Conclusions and perspectivesHierarchical step-wise optimization with classification 2 Calculate Dissimilarity Criterion between adjacent regions: Compute Spectral Angle Mapper Compute SAM between region mean vectors SAM(ui, uj) between the region mean vectors ui and uj no yes L(Ri) = L(Rj) = k’ ui · uj SAM(ui , uj ) = arccos( ) ui 2 uj 2 no (card(Ri) > M) & yes DC = (2 – max(Pk’(Ri), (card(Rj) > M) Pk’(Rj)))SAM(ui, uj) If adjacent regions have equal class labels → they belong more likely to DC = ∞ the same region DC = (2 – min(PL(Rj)(Ri), If two large regions are assigned to PL(Ri)(Rj)))SAM(ui, uj) different classes → they cannot be merged together DC If two regions have different class labels → DC between them is penalized by (2 − min(PL(Rj ) (Ri ), PL(Ri ) (Rj ))) Yuliya Tarabalka and James C. Tilton (yuliya.tarabalka@nasa.gov) Best merge region growing with integrated classification for HS data 14
  21. 21. Introduction Proposed spectral-spatial classification scheme Conclusions and perspectivesHierarchical step-wise optimization with classification Hyperspectral image 3 Find the smallest dissimilarity criterion DCmin Preliminary probabilistic classification Each pixel = one region  1  2  3 While not converged Find min(DC) between  4  5  6 all pairs of spatially adjacent regions (SAR) Merge all pairs of SAR with DC = min(DC).  7  8  9   Classify new regions  DCmin Spectral-spatial  10  11  12 classification map Yuliya Tarabalka and James C. Tilton (yuliya.tarabalka@nasa.gov) Best merge region growing with integrated classification for HS data 15
  22. 22. Introduction Proposed spectral-spatial classification scheme Conclusions and perspectivesHierarchical step-wise optimization with classification 4 Merge all pairs of spatially adjacent regions Hyperspectral image with DC = DCmin . Preliminary probabilistic classification Each pixel = For each new region Rnew = Ri + Rj : one region While not converged Pk (Ri )car d(Ri ) + Pk (Rj )car d(Rj ) Find min(DC) between Pk (Rnew ) = all pairs of spatially adjacent regions (SAR) car d(Rnew ) Merge all pairs of SAR with DC = min(DC). Classify new regions L(Rnew ) = arg max{Pk (Rnew )} k Spectral-spatial classification map All the pixels in Rnew get a definite class label. Yuliya Tarabalka and James C. Tilton (yuliya.tarabalka@nasa.gov) Best merge region growing with integrated classification for HS data 16
  23. 23. Introduction Proposed spectral-spatial classification scheme Conclusions and perspectivesHierarchical step-wise optimization with classification 4 Merge all pairs of spatially adjacent regions with DC = DCmin .  1  2  3 For each new region Rnew = Ri + Rj :  4  5  6 Pk (Ri )car d(Ri ) + Pk (Rj )car d(Rj )  Rnew  8  9   Pk (Rnew ) = car d(Rnew )  11  12 L(Rnew ) = arg max{Pk (Rnew )} k All the pixels in Rnew get a definite class label. Yuliya Tarabalka and James C. Tilton (yuliya.tarabalka@nasa.gov) Best merge region growing with integrated classification for HS data 16
  24. 24. Introduction Proposed spectral-spatial classification scheme Conclusions and perspectivesHierarchical step-wise optimization with classification Hyperspectral image Preliminary 2 Calculate Dissimilarity Criterion between probabilistic classification adjacent regions Each pixel = one region 3 Find the smallest dissimilarity criterion DCmin While not converged Find min(DC) between all pairs of spatially 4 Merge all pairs of spatially adjacent regions adjacent regions (SAR) with DC = DCmin Merge all pairs of SAR with DC = min(DC). Classify new regions 5 Stop if all n pixels get a definite class label. Spectral-spatial classification map If not converged, go to step 2 Yuliya Tarabalka and James C. Tilton (yuliya.tarabalka@nasa.gov) Best merge region growing with integrated classification for HS data 17
  25. 25. Introduction Proposed spectral-spatial classification scheme Conclusions and perspectivesClassification maps SVM Proposed HSwC method Yuliya Tarabalka and James C. Tilton (yuliya.tarabalka@nasa.gov) Best merge region growing with integrated classification for HS data 18
  26. 26. Introduction Proposed spectral-spatial classification scheme Conclusions and perspectivesClassification accuracies (%) No. of Samp. SVM HSeg SVM ECHO HSwC Train Test MSF +MV Overall Accuracy - - 78.17 82.64 88.41 90.86 89.24 Average Accuracy - - 85.97 83.75 91.57 93.96 94.18 Kappa Coefficient κ - - 75.33 80.38 86.71 89.56 87.76 Corn-no till 50 1384 78.18 83.45 90.97 90.46 93.06 Corn-min till 50 784 69.64 75.13 69.52 83.04 82.53 Corn 50 184 91.85 92.39 95.65 95.65 97.28 Soybeans-no till 50 918 82.03 90.10 98.04 92.06 95.10 Soybeans-min till 50 2418 58.95 64.14 81.97 84.04 74.36 Soybeans-clean till 50 564 87.94 89.89 85.99 95.39 96.10 Alfalfa 15 39 74.36 48.72 94.87 92.31 97.44 Grass/pasture 50 447 92.17 94.18 94.63 94.41 93.96 Grass/trees 50 697 91.68 96.27 92.40 97.56 97.85 Grass/pasture-mowed 15 11 100 36.36 100 100 100 Hay-windrowed 50 439 97.72 97.72 99.77 99.54 98.86 Oats 15 5 100 100 100 100 100 Wheat 50 162 98.77 98.15 99.38 98.15 99.38 Woods 50 1244 93.01 94.21 97.59 98.63 99.52 Bldg-Grass-Tree-Drives 50 330 61.52 81.52 68.79 82.12 81.52 Stone-steel towers 50 45 97.78 97.78 95.56 100 100 Yuliya Tarabalka and James C. Tilton (yuliya.tarabalka@nasa.gov) Best merge region growing with integrated classification for HS data 19
  27. 27. Introduction Proposed spectral-spatial classification scheme Conclusions and perspectivesConclusions and perspectives Conclusions 1 New spectral-spatial classification method for hyperspectral images was proposed 2 New dissimilarity criterion between image regions was defined 3 The proposed method: improves classification accuracies provides classification maps with homogeneous regions Perspectives Explore further the choice of: optimal representative features for segmentation regions dissimilarity measures between regions Yuliya Tarabalka and James C. Tilton (yuliya.tarabalka@nasa.gov) Best merge region growing with integrated classification for HS data 20
  28. 28. Introduction Proposed spectral-spatial classification scheme Conclusions and perspectives Thank you for your attention!Yuliya Tarabalka and James C. Tilton (yuliya.tarabalka@nasa.gov) Best merge region growing with integrated classification for HS data 21
  29. 29. Best Merge Region Growingwith Integrated Probabilistic Classification for Hyperspectral Imagery Yuliya Tarabalka and James C. Tilton NASA Goddard Space Flight Center, Mail Code 606.3, Greenbelt, MD 20771, USA e-mail: yuliya.tarabalka@nasa.gov July 28, 2011

×