Study On SAR Image Processing Methods Applied To Earthquake Damage Information Extraction

Remote sensing technology, with characteristics of fast access, large areacoverage, rich information and independence on ground conditions, has been provedas the most effective way of obtaining earthquake damage information. Syntheticaperture radar (SAR), operateingat microwave band (1cm1m), with all-time andall-weather capability of earth observation, is playing an important role in earthquakeemergency response. Many ofthe optical image processing methods areusually notapplicable to SAR image directly, due to the differnence of imaging mechanismsbetween SAR and optical remote sensing. With the development of computertechnology, automatic processingtechnology is requiredfor digital image processing.Therefore, carrying out research on key technology of SAR image processing, andexploring automatic damage information extraction methods from SAR, are importantsubjects in SAR image applications.Based on the analysis of SAR imaging principles and image characteristics,methodsof coherent speckle noise suppression, automatic image co-registration andchange detection frommulti-temporal SAR images for earthquake damage informationextraction was studied. In this paper, a image co-registration method based on Scale-invariant Feature Transform(SIFT) has been proposed and applied to earthquakedamage information extraction.In this paper, SAR imaging principles and image characteristics were introduced,and theoretics and methods of coherent speckle noise suppression, imageco-registration and change detection were summarized.SAR image automatic co-registration methodbased on improved SIFT has beenproposed:(1) ISEF (Infinite Symmetric Exponential Filter) was used to reduce specklenoise before the computation of the scale space pyramid;(2) SIFT algorithm was usedto detect the feature points and the first scale-space octave was skippedfor reducingprocessing time;(3) omiting the false matches in the Euclidean space. In thisexperiment, ALOS PALSAR images acquired before and after YuShu earthquakewereprocessed. Results showed that the co-registration method proposed in this paper hasincreased the number of the features detected and improved the robustness.Thematching accuracy could meet the sub-pixel level requirement and the processingtime was decreased by more than60%.Finally, correlation coefficient method based on the local featureswas used to co-registed SAR images for earthquake damage information detection. Correlationcoefficient image was segmented with thresholds to generate building damagelevelmap. The results showed that building damage distribution extracted from SARimages was consistent with the results from high spatial resolution aerial image.In summary,the conclusions are as follows:(1) The number and stability of SIFT feature extracted could be affected byspeckle noise of SAR image.This study has shown that ISEFalgorithm was helpful forimproving the stability of SIFT algorithm.(2)SIFT algorithm is characterised by computational complexity andtime-consuming, and the features extracted in the first layer of the pyramid are usuallynot stable.The processing efficency and matching accuracy of feature points detectedcould be improved by skipping the first layer of the pyramid, and the false matcheswould also be reduced.(3)The study of automatic co-registration method for SAR images based on theSIFT algorithm would play an important role in disaster information rapid extractionfor earthquake response.