Research Of High Resulotuon SAR Image Change Detection Based On Semantic Analysis

Synthetic Aperture Radar (SAR) can supplyall-day, all-weather remotesensing informationof earth observation, which plays a very important rolein military reconnaissance, geographical mapping and marine surveillance.With the promotion of SAR image resolution, it can provide moreabundant information about the image content. While owing to someinevitable shortcomings such as speckle noise, deficiency of thestructural information, it is necessary to develop a accurate, fast,applicable interpretation technology.SAR image change detection can obtain the change information byanalyzing the difference between two different temporal images of thesame area, which is used widely in disaster area positioning and evaluation,urban expansion and environment monitoring. The thesis is facing to thechange detection in the high resolution SAR image interpretation.Referring to the unfavorable elements to change detection and thedomestic status quo, the thesis focus on resolve the semantic informationdeficiency problem during the SAR image change detection. Consideringthe high resolution SAR images characteristic, we proposed a methodnamed change detection base on semantic analysis. It contains two aspects:change detection based on unsupervised land cover type classification,change detection base on bag of word model. Those two methods canrespectively implement the change detection of land cover type in simplescene images and the change detection result with semantic information incomplex scene images. We expected to realize the extent of the changedetection results from statistical significance analysis to the semanticinformation change.Two group of different temporal SAR images are used to verify the feasibility of detection method based on land cover classification and bagof words model. Through analyzing the probability of the semantic wordsin two temporal images, combined with the statistical change detectionmethod based on pixel, we obtain the structure semantic changeinformation of change area. It is proved by the experiments that thepropose method can reach the goal of obtaining semantic information ofchange detection result.