Research On Multi-temporal SAR Change Detection Algorithm Based On Multi-feature Fusion

In recent years,with the development of aeronautics and Astronautics and related remote sensing technology,more and more high-precision and high-resolution SAR data are applied in various fields.Change detection using SAR images has become one of the main applications of SAR data,and is also a hot research at home and abroad.The traditional change detection technology uses two images in the same area to compare the feature differences of the same location in different time phases to form feature difference images and cluster the difference images.In recent years,with the continuous deepening of research,more and more detection algorithms have appeared,but because each feature has certain limitations in the application of actual change detection,so far there is no one applicable to all situations Algorithm,in order to improve the detection accuracy and the application scenarios of the algorithm,the research of change detection has gradually developed in the direction of multi-feature fusion.Firstly,this paper systematically studies the detection flow of the classical two-phase direct comparison method,aiming at the problems in the classical comparison method,this paper mainly carries out the following research:1.Aiming at the disadvantage of low detection accuracy caused by not fully considering the relationship between neighborhood when detecting the change of pixel gray value in the traditional two-phase change detection algorithm,in this paper,a texture feature fusion method is proposed to construct the difference image.In this method,the entropy and mean statistical properties of the gray level co-occurrence matrix are calculated,and the weights are assigned to the image of the difference of entropy and mean according to the gray level changes in the region in order to reconstruct a differential image.The validity of the multi-feature fusion algorithm is verified.2.Aiming at the clustering of different images in dual-phase change detection algorithm,this article first studies the OTSU threshold segmentation algorithm.In view of the fact that the traditional OTSU threshold segmentation results are affected by the intensity and proportion of the changing regions in the difference images in actual projects,this paper has made certain improvements.By performing multiple iterations on the basis of traditional OTSU threshold segmentation,the final iteration value is determined according to the difference between adjacent iteration thresholds.In this paper,two threshold segmentation algorithms are tested on the same difference image and prove the rationality of the iterative threshold segmentation results.3.When the local objects occupy less pixels and are scattered in the image,the accuracy of the traditional dual-phase detection algorithm is lower,especially in low-resolution SAR images.Therefore,this paper proposes a change detection algorithm that adds SAR images to the time dimension and constructs a long time series.The algorithm mainly models the change rule of the image points on the time series and extracts the change detection region,the algorithm can effectively detect the building demolition and construction,and has been verified in the data of Raytheon Mountain Hospital.4.Based on the detection of building demolition and construction by using long time series gray features,a fusion detection algorithm of entropy and gray features is proposed.In this algorithm,entropy and gray feature are modeled on time series respectively,and the final result is obtained by fusion of detection results.The experimental data show that the feature fusion algorithm can make up for the shortcomings of the single gray-scale feature in the detection of long time series.5.In this paper,with the help of VS2015 platform and C++ language,dual-phase change detection is implemented in engineering,and the practicability of the software is tested through the measured data.