| Target segmentation of sonar image is a challenging work,which is not only of great significance to the direct realization of the exploration of submarine resources and target extraction,but also has been paid attention to as the important basis for detection correctly and tracking of targets.Due to the particularity of the underwater imaging environment,sonar image segmentation can not be realized directly by other classical image segmentation methods.Based on this problem,we establish a sonar image target segmentation framework and divides the sonar image segmentation task into the region and edge segmentation.Two unsupervised region segmentation methods for sonar image are proposed;An improved level set method(LSM)of sonar image contour extraction is proposed;Lattice Boltzmann method(LBM)is introduced and improved to realize the global segmentation of sonar image;The research results are integrated into a comprehensive GUI and extended to other image processing.The main contents and innovations of our research are as follows:(1)Two novel unsupervised methods of the sonar image are proposed to realize the target region segmentation of sonar images.The two methods are named cluster connectivity analysis(KCA)and cluster morphological analysis(KMA).LBM is introduced and improved as the global segmentation result of sonar images.The experimental results of real sonar images show that the proposed methods are better than the other four region segmentation algorithms.(2)Aiming at the problem of sonar image edge segmentation,an improved level set method is proposed.The improved level set algorithm uses a new initialization model to replace the random initialization model,and finally realizes the successful segmentation of sonar image edge contour.Based on a distance regularized level set(DRLSE)method,the minimum moments generated by unsupervised segmentation are embedded in the energy function of level set evolution.Combined with the global region segmentation of LBM,the task of the region and edge segmentation of sonar image is realized at the same time.In order to verify the superiority of the improved level set algorithm,we not only compare four widely used edge detection algorithms horizontally but also compare five classic level set segmentation algorithms with high frequency.Finally,four evaluation indexes are used to objectively evaluate the accuracy of the algorithm.The data results show that the average result of our methods is better than others,and the segmentation accuracy is improved significantly.(3)A full-automatic Matlab GUI is designed and implemented for sonar image segmentation.The segmentation methods proposed and improved in this paper are successfully embedded into the platform interface to realize one-click segmentation of sonar images.The platform is also extended to general image preprocessing and advanced processing.The platform consists of three parts: interface guidance,menu function,and interface display module.The menu function module includes 17 common algorithms,such as image geometric operation,filtering,morphological operation,edge detection,and threshold segmentation.The design and use of the GUI not only provide a great convenience for users,but also realizes the application transformation of our research,and the practical application significance of this paper is enhanced. |
