Methods Of Extracting Building Damage From High Resolution Sar Imagery

Earthquake is one of the most devastating disasters. Rapid and accurate damageassessment is crucial for emergency rescue after a destructive earthquake. Because ofthe excellent characteristics such as fast access, large area coverage, beinginformative and independence on ground natural conditions, remote sensingtechniques have been playing an important role in earthquake emergency response,among which the synthetic aperture radar (SAR) remote sensing is especiallypromising as the SAR sensors can acquire imagery all-day and all-weather. In allkinds of damages caused by the catastrophic earthquakes, the building damage is mostrelevant to casualties and economic losses. Therefore, the study on the methods ofbuilding damage information extraction from SAR imagery is of important practicalsignificance. There has been a tremendous advancement in very high resolution (VHR)SAR remote sensing in recent years, while the researches on the methods of extractingbuilding damage from VHR SAR imagery are insufficient.In this thesis, the methods of building damage information extraction from VHRSAR imagery for earthquake emergency response are studied. The main contentincludes three aspects: the fundamental theories and the data sources of SAR remotesensing related to the building damage information extraction, the visualinterpretation method, and the automatic method based on the fractal analysis of thepost-earthquake SAR imagery.First, the physical basis of radar remote sensing, the geometric and radiationcharacteristics of SAR imagery was introduced， and the performances, such as thespatial resolution, temporal resolution, polarimetric capability, swath of the SARsystems COSMO-SkyMed、TerraSAR-X、RADARSAT-2and homemade airborne Xband SAR were summarized.Then the visual interpretation method of extracting building damage from SARimagery was investigated. In this investigation, a standard of the building damageclassifications and degrees was proposed. Based on the geometry model of buildings,the idealized characteristics of buildings in SAR imagery were analyzed. Using VHRairborne SAR imagery, the appearance of buildings in real SAR imagery wasdiscussed. Based on the work mentioned above, the interpreting indicators fordifferent damage grades were established and the building damage in Yushu city dueto the2010Yushu MS7.1earthquake was assessed quantitatively by visualinterpretation using VHR airborne SAR imagery acquired after the earthquake. Theaccuracy of the results obtained from the SAR imagery was evaluated by comparingwith the results obtained from the airborne optical imagery.Finally, an automatic method of building damage information extraction based onthe fractal analysis of the post-earthquake SAR imagery was developed. A movingwindow which moves pixel by pixel is used in this proposed method. When thewindow moves to a place, a fractal dimension which is called local fractal dimension‘can be computed with the pixels contained in the window and the local fractaldimension is then arranged to the pixel in the middle of the window. In this way, mostpixels of the image except border ones are characterized with corresponding fractaldimensions, which is different from that the entire image was characterize with onlyone fractal dimension. In each moving window, the fractal dimension is computedusing triangular prism method and differential box counting method. Thiseffectiveness of different methods was demonstrated in Yushu city with VHRairborne SAR imagery. Based on the average fractal characteristic of the city blocks, the average damage in Yushu city was assessed block by block.Following conclusions could be drawn from the study on the visual interpretationmethod:(1) The SAR image of buildings typically consisted of layover area, cornerreflector and shadow.(2) The most important interpreting indicators for buildingdamage information extraction from SAR imagery are tone, shape, texture andshadow.(3) It is relatively easy to identify the intact buildings and collapsed buildingsin VHR SAR imagery, while it‘s rather difficult to identify the partially collapsedbuildings.(4) It is almost impossible to detected building damage in steep mountainareas and dense urban areas.(5) A desirable damage result could be derived in Yushucity from VHR airborne SAR imagery by visual interpretation method.The local fractal analysis of Yushu city shows that the local fractal dimension is aneffective texture measure which can indicate the building damage degree to someextent. However, different computation methods or different moving window sizeswill achieve different results. Generally, the differential box counting method canobtain better texture analysis results than triangular prism method for VHR airborneSAR imagery. For both the methods, the smaller window size could get better results.However, there are fractal dimensions greater than3or less than2which are beyondthe theoretical range for both the methods. It suggests that these two methods sould beimproved in the future.