Building Change Detection in Multitemporal Very High Resolution SAR Images L. Bruzzone, et al. Juanping Zhao 2015.07.17 [1] Marin C, Bovolo F, Bruzzone L. Building Change Detection in Multitemporal Very High Resolution SAR Images[J].IEEE Transactions on Geoscience and Remote Sensing, , 2015, 53(5): 2664-2682. 坦克的仿真
Problems Urban evolution
Problems Earthquake damages
Outline Background Related Work Methodology Experimental Results Discussions
Outline Background Related Work Methodology Experimental Results Discussions
Background Challenges More heterogeneous Speckle
Background Challenges Geometric distortions ground layover Double bounce roof shadow
Background Challenges Semantic Confusion Intra-Class Variations (operation conditions)
Outline Background Related Work Methodology Experimental Results Discussions
Urban areas Change detection Related works Urban areas Change detection Earthquake damages Building database updating Urban evolution… Methods Supervised analysis Fusion Large geometric scale
Building CD using Double Bounce Line Related works Building CD using Double Bounce Line Partial Information Double bounce line is not reliable
Outline Background Related Work Methodology Experimental Results Discussions
Methodology Two basic ideas Extract information of change at optimal building scale Exploit backscattering properties of bi-temporal images
Methodology Backscattering properties of single images Single detected VHR SAR images
Methodology Backscattering properties of single images Single detected VHR SAR images Backscattering properties of single images
Methodology Backscattering properties of single images Single detected VHR SAR images
Bi-temporal VHR SAR images Methodology Backscattering properties of bi-temporal images Bi-temporal VHR SAR images
Multiscale Decomposition Methodology Step 1 Log-ratio Transform Step 2 Multiscale Decomposition Step 3 Split-based Analysis Step 4 Candidate Detection Step 5 Fuzzy Classification Architecture of the proposed approach to building CD
Architecture of the proposed approach to building CD Methodology Log-ratio transform Architecture of the proposed approach to building CD
Methodology Log-ratio transform
Architecture of the proposed approach to building CD Methodology Detection of backscattering changes at optimal building scale Architecture of the proposed approach to building CD
2D stationary wavelet transform Methodology Detection of backscattering changes at optimal building scale 2D stationary wavelet transform Reducing the impact of small changes mitigation of the speckle effect
Architecture of the proposed approach to building CD Methodology Split-based threshold decision Architecture of the proposed approach to building CD
Methodology Split-based threshold decision
Architecture of the proposed approach to building CD Methodology Change building candidates Architecture of the proposed approach to building CD
Methodology Change building candidates
Architecture of the proposed approach to building CD Methodology Fuzzy rules decision Architecture of the proposed approach to building CD
Methodology Four Classes Fuzzy rules decision fully destroyed buildings new buildings changes that have a size comparable to a building all other changes that do not show a size comparable to a building Fuzzy rules decision Completeness Proportionality of areas Equivalence of length Alignment
Methodology Fuzzy rules decision
Methodology Fuzzy rules decision Equivalence of length Alignment
Outline Background Related Work Methodology Experimental Results Discussions
2009 L’Aquila Earthquake: Detection of Destroyed buildings Experimental Results 2009 L’Aquila Earthquake: Detection of Destroyed buildings sensor COSMO-SkyMed mode spotlight band X looks 1 resolution 1m Pixel spacing 0.5m*0.5m Pol-mode HH orbit ascending Incidence angle 57°- 58°
Experimental Results 2009 L’Aquila Earthquake: Detection of Destroyed buildings
Experimental Results 2009 L’Aquila Earthquake: Detection of Destroyed buildings
Trento Data Set: Detection of New buildings Experimental Results Trento Data Set: Detection of New buildings sensor TerraSAR-X TanDEM-X mode spotlight band X波段 looks 1 resolution 0.58m*1.1m Pixel spacing 0.454m*0.855m Pol-mode HH orbit ascending Incidence angle 53°
Experimental Results Trento Data Set: Detection of New buildings
Experimental Results Trento Data Set: Detection of New buildings
Outline Background Related Work Methodology Experimental Results Discussions
Discussions Taking home messages Unsupervised Scattering properties of bi-temporal images Multitemporal correlation between images Intrinsic multiscale nature of objects present in VHR images More flexible
Possibilities to improve Discussions Possibilities to improve Improve the building detector by better modeling the geometrical behaviors of building primitives; Investigate the possibility to discriminate among several building construction stages and/or building damage levels.