Adaptive Feature Weight-based Multi-Objective SAR Image Segmentation Algorithm

In recent years,synthetic aperture radar(SAR)gradually has been applied to many realworld applications because of its advantages,which involves the process of extraction,recognition,understanding and analysis of targets in the image.Among these tasks,SAR image segmentation is indispensable to image interpretation in the process steps,and it is the basis of image processing.Therefore,image segmentation has always been a hot topic in SAR image processing research.Aiming at problems in computing similarity of multi-feature by using Euclidean distance in exited SAR segmentation methods,and several SAR image segmentation methods based on the improved similarity of combination feature are proposed.The main work can be outlined as follows:1.A multi-objective SAR segmentation algorithm based on the fair integration of feature is proposed.Firstly,after analyzing problems in computing similarity of the combination feature by using Euclidean distance,we propose a similarity calculation method for different kinds of feature.Secondly,this improved similarity calculation method is introduced into the objective functions in multi-objective clustering to improve the clustering performance of the algorithm.Experimental results show that the new algorithm has great advantages in keeping image edge and consistency,and the segmentation accuracy is higher than other comparative algorithms in this study.2.An adaptive feature weight-based double-layer multi-objective SAR image segmentation algorithm is developed.The proposed algorithm is divided into two layers,in the first layer,the algorithm adaptively identifies the image's dominant feature and optimal feature weight by using differential evolution.In the second layer,multi-objective clustering functions are established and optimized by using the feature weight obtained by the upper layer.Finally,a solution set with high segmentation accuracy is obtained.In the experimental part,six synthetic texture images and two real SAR images are segmented,and the experimental results show that the new algorithm has great advantage in each evaluation index.Moreover,the weight obtained by the new algorithm is compared with four fixed weights on three synthetic texture images,and the results show that the new algorithm can be adaptive to identify the dominant feature of different image,and can get a better feature weight.3.An adaptive feature weight-based multi-objective SAR automatic clustering algorithm is put forward.The algorithm is also divided into two layers,the first layer is mainly used to optimize the number of cluster and feature weight,and then pass the optimized two parameters to the next layer.In the second layer,after receiving the first layer parameters,automatic clustering problem is converted to non-automatic clustering problem.Then multiobjective algorithm is used to optimize the multi-objective clustering function with obtained feature weight and the final segmentation result is got.In the experimental part,several synthetic texture images and real SAR images are segmented.The experimental results show that the new algorithm has great advantage in each evaluation index,and can adjust the number of cluster categories and get better feature weight.