Multi-objective Evolutionary Fuzzy Clustering For PolSAR Image Superpixel Segmentation

Synthetic Aperture Radar(SAR)has the advantages that it can work uninterruptedly throughout the day,and the observation process is not affected by climate and light.So it plays an important role in national defense,military reconnaissance,forestry protection and other fields.Polarimetric synthetic aperture radar(Pol SAR)enhances the performance of SAR in acquiring targets.It uses multiple polarization methods to obtain various scattering parameters of targets from multiple angles.By interpreting and analyzing these parameters,more specific information in the target can be obtained.Superpixel is a set of regions composed of adjacent pixels with similar characteristics in the image,so superpixel can well retain the local structure of the image.If the image is interpreted based on the superpixel,the influence of speckle noise can be greatly reduced,and using superpixel instead of pixel for Pol SAR image processing can reduce the computational cost.At present,most of the superpixel segmentation algorithms of Pol SAR images belong to the single-objective optimization.However,in practical application,due to the interference and influence of specific complex scene factors and image noise,Pol SAR data becomes more and more complex,and these single-objective optimization algorithms are difficult to meet the requirements of superpixel segmentation accuracy of Pol SAR images.The main work of this paper is summarized as follows:1.Superpixel segmentation of Pol SAR images based on multi-objective evolution is studied.The superpixel segmentation of Pol SAR image is regarded as a multi-objective optimization problem,and the intra-class distance and inter-class distance are taken as the objective function at the same time.In the optimization strategy,a boundary function is introduced to enhance the fit between the superpixel edge and the real object edge,and a targeted evolution operator and population initialization method are designed to speed up the optimization of the algorithm.Through the experiments of superpixel segmentation on different types of Pol SAR images,it is verified that this work can effectively balance the relationship between intra-block similarity maximization and inter-block similarity minimization.2.This work studies the adaptive Pol SAR image superpixel segmentation based on multiobjective evolutionary fuzzy clustering in order to reduce the influence of manually setting the number of superpixels on the results of superpixel segmentation.In this work,the fuzzy clustering performance index and fuzzy clustering energy function are optimized at the same time.In order to automatically determine the number of superpixels that can achieve the best result of superpixel segmentation,an adaptive method to determine the number of superpixels is designed.In the optimization strategy,the mutation operator of multi-objective evolution is designed to improve the search ability of the algorithm.Experiments show that this work can automatically determine the optimal number of superpixels suitable for Pol SAR images,so as to achieve a good effect of superpixel segmentation of Pol SAR images.3.Pol SAR image superpixel segmentation based on image local information and multiobjective evolutionary fuzzy clustering is studied.The inter-class distance and intra-class distance based on image local information are taken as the objective functions to optimize the performance of Pol SAR image superpixel segmentation.In this work,the number of superpixels is determined in an adaptive way.When updating the superpixel center,in addition to using the evolutionary algorithm,the FCM fuzzy clustering center update method is also introduced to speed up the optimization of the algorithm.Through superpixel segmentation of different types of Pol SAR images,it is verified that this work has a good performance of superpixel segmentation of Pol SAR images.