The Application Of Secondary Classification In SAR Image Change Detection

In the context of today’s big data era,high-performance computers such as high CPUs and GPUs are becoming more common.Graphics image processing with highperformance computers is no longer limited by storage space and computer processing power.At the same time,we should also see that the global economy is growing at a faster rate.The urbanization process is accelerating.Every day,new high-rise buildings are emerging and the infrastructure of urbanization is constantly improving.The land resources are expanded and the land use is constantly changing.Similarly,due to global warming and other factors,natural disasters such as floods,mudslides,mountain collapses and earthquakes frequently cause a region to undergo dramatic changes in a short period of time.Therefore,how to use satellite imagery has an important practical effect on the accurate detection of the changed regions.In order to use high-performance computer to make accurate detection of satellite remote sensing image in the changed area,eliminate the interference of noise to the detected area and solve the problems of unsatisfactory detection and low efficiency.This paper presents a method based on guided image filtering and reclassification strategy.The method mainly includes the following four steps.The original SAR remote sensing images are noisy,which has an important impact on the later detection results.Therefore,the original SAR images are firstly de-noised using Lee filter.Secondly,the difference image is generated by the neighborhood ratio operator,and enhanced by the guided image filter.Thirdly,the fuzzy c-means clustering algorithm is used for the initial classification.Finally,the extreme learning machine is used for the second classification to ensure the accuracy of the final classification results.The main outlines of the work and innovations made in this paper are as follows:1.The traditional SAR image change detection model pays insufficient attention to the edge detection of satellite remote sensing images and ultimately affects the accuracy of the results.This paper proposes to use a new filter,the guided image filter,to achieve differential image enhancement and edge retention.Compared with traditional filters,its time complexity is linear and its filtering process is fast.Firstly,a difference image is generated using a neighborhood-ratio operator.Secondly,the generated difference image is subjected to image enhancement processing by guided image filtering and subjected to edge retention.Through the above two steps,the accuracy of the classification results can be guaranteed.2.In order to further improve the accuracy of the classification.This paper proposes to use the extreme learning machine as a secondary classifier,which can realize autonomous learning and efficient classification in a short time.Firstly,uses the fuzzy C-means clustering algorithm to classify the generated difference images for the first time.The core of this method is to cluster the difference image according to the pixels.The difference image will eventually divided into three categories,namely changed category,unchanged category,and middle class.And then the changed category and the unchanged category pixel generated by the fuzzy C-means clustering algorithm are sent into the extreme learning machine.Finally,the trained extreme learning machine classifier is obtained.The intermediate categories generated by the fuzzy c-means clustering algorithm are sent to the trained ELM classifier,and finally subdivides the intermediate category into the changed category and unchanged category.Thus,the final classification results are obtained and the accuracy of the final classification results is verified.3.A SAR image change detection model based on guided filter and reclassification strategy is proposed.It mainly includes a guide filter to achieve differential image enhancement and an extreme learning machine to achieve secondary classification.Through the above two modules,SAR image change detection can be efficiently implemented.