Algorithm Research On Evolutionary Multi–Objective SAR Image Change Detection Based On New Difference Image

The purpose of SAR image change detection is to detect changes in two SAR images at different times in the same scene.It is essentially a classification problem.The same pixels in the SAR image are divided into a class called the unchanged class;the different pixels are divided into another class,which is the changed class.At present,the process of SAR image change detection can be divided into three stages.The first stage is the preprocessing process for SAR images,including image registration,geometric correction and denoising.The main task of the second stage is to generate the difference image.This stage is the key of SAR image change detection.At present,the main methods are difference method,ratio method,logarithmic ratio method,neighborhood ratio method and so on.After the difference image is generated,the third stage is the analysis of the difference image,which is essentially the segmentation of the image.Difference image analysis is divided into two methods.One is the threshold segmentation method,another is the clustering method.Since the SAR image change detection is widely used in practical problems in recent years,it attracts more and more researchers.In this paper,about the generation of difference image and how to introduce ensemble learning and deep learning into the evolutionary multi-objective optimization algorithm,do the following three aspects:1.A SAR image change detection algorithm based on fluctuation parameter partition(Fluctuation Parameter Fuzzy C-Means,FPFCM)is proposed.In the algorithm,we design a new difference image method based on the fluctuation parameter,which is based on the difference image of the neighborhood ratio method,and overcomes the shortcomings of the spatial information utilization of the neighborhood ratio method.Secondly,in order to improve the accuracy of the algorithm,the method of neighborhood error correction is used to deal with the binary image obtained by FCM algorithm.The experimental results show that the accuracy of the proposed algorithm is higher than that of other algorithms.2.A new multi-objective optimization SAR image change detection algorithm based on ensemble learning is proposed(Ensemble Learning Multi-Objective Optimization Fuzzy C-Means,ELMOFCM).Based on the new difference image,the multi-objective optimization fuzzy c-Means(MOFCM)algorithm is used as the main framework.At the same time,in the multi-objective Pareto optimal solution set,The binary image is obtained by the selective ensemble method in ensemble learning.Experiments show that the performance of the algorithm is superior to other comparative algorithms.3.A Multi-Objective Optimization Fuzzy C-Means algorithm based on Deep Belief Network is proposed(DBNMOFCM).The algorithm is based on fluctuation parameter partition difference image,adopts MOFCM algorithm as the main framework,and introduces the Deep Belief Network learning method.Firstly,the reference image is used as the training data to train the image Feature model.Secondly,the obtained non-dominated solutions are the input of the model,the output are the image pixel classification labels.Thirdly,we get the change detection results according to the classification labels.Experiments indicate that PCC of the algorithm is higher than other algorithms.