Improved Salp Swarm Algorithm And Its Application In Remote Sensing Image Segmentation

With the rapid development of remote sensing technology,remote sensing images have laid a solid foundation for many research work.However,there are a large amount of data types in remote sensing images,which are complex,multispectral and multi-level,so remote sensing image segmentation has always been the key research direction of remote sensing image processing.Although the existing research on image segmentation has been extensive and mature,it is limited to varying degrees in practical application to remote sensing images.In order to accurately segment remote sensing images,based on the Density Peak Clustering(DPC)algorithm,this thesis proposes to optimize the clustering algorithm by using the characteristics of the Salp Swarm Algorithm(SSA),and focuses on improving the performance of the DPC algorithm in remote sensing image segmentation.The specific contents are as follows,(1)In view of the slow convergence speed and poor accuracy of the SSA,an improved SSA construction and search strategy is proposed in this paper.Firstly,the cosine function is introduced to propose a multi-leader strategy,which can adaptively adjust the number of leaders and followers in the salp swarm in the search process,improve the diversity of the salp population,and lay the foundation for accurately finding the global optimal value;Secondly,Chebyshev mapping function is introduced into the leader position update formula to enhance the ergodicity and global search ability in the process of leader’s movement;Finally,the linear decreasing inertia weight is introduced into the follower position update formula to reduce the follower’s dependence on the position of the previous salp and improve the local search ability of the algorithm.In the experiment,four unimodal functions and four multimodal functions are used as test functions,and compared with five other population optimization algorithms.The results show that the improved algorithm has higher accuracy,faster convergence speed and stronger stability.(2)In order to solve the difficulty of manually selecting parameters and clustering center of DPC,this thesis combines DPC algorithm with improved SSA.Firstly,the product of normalized local density and relative distance is calculated,and the clustering center is selected adaptively to avoid the subjectivity of manual selection;Secondly,the density measure is introduced into the information entropy to propose the density estimation information entropy,and then the fitness function is established;Finally,the optimal value of fitness function is solved by using the improved SSA to obtain the optimal truncation distance.In order to verify the effectiveness of the algorithm,four commonly used synthetic data sets are selected in the experiment.The results show that compared with the traditional density clustering algorithm,the proposed algorithm has better clustering effect and higher accuracy.(3)In order to solve the problem of data dimension in the application of DPC algorithm in remote sensing image segmentation,super-pixel segmentation algorithm is introduced.Firstly,the remote sensing image is segmented by super-pixel,and the 5-Dimensional features based on position and color are obtained.Then the 5-Dimensional features of all super-pixel blocks are clustered by DPC algorithm,and the homogeneous regions are combined to realize the remote sensing image segmentation.In the experiment,four different scene images are chosed from the remote sensing data set,and Worldview2 and Quickbird large-scale remote sensing images are selected as the research area.The experiment is compared with the other three algorithms.The results show that the proposed algorithm has more accurate segmentation results and better applicability.This paper has 34 figures,9 tables and 82 references.