Research On Side-Scan Sonar Image Segmentation Under Long-Tailed Distribution And Application Of Virtual Simulation Platform

In recent years,with the development of the global economy and the emphasis on ocean resources,Autonomous Underwater Vehicles(AUVs)have played an important role in ocean resource development and protection.AUVs are widely used in many fields,such as ocean research,oil and gas exploration,mineral resource surveys,fisheries,and military operations.Valuable oceanic information can be obtained by segmenting,classifying,and recognizing sonar images acquired by AUVs.Therefore,the research on side-scan sonar image interpretation holds significant meaning.Underwater side-scan sonar image segmentation is an important technology in underwater surveying.However,in practical applications,the visual imaging of underwater robots faces many challenges,such as uneven underwater illumination,image blurring,and color distortion,which can affect the quality of the images and the accuracy of segmentation.On the other hand,due to the complexity of the marine environment,sound waves are influenced by multiple factors,leading to strong noise and blurring in side-scan sonar images.Furthermore,the reflected signal strengths of targets at different depths vary greatly,presenting a typical long-tailed distribution,which becomes the bottleneck that restricts further improvements in accuracy and efficiency.The contribution of this paper is to propose an input-output data balancing framework to address the long-tailed data distribution problem in underwater side-scan sonar image instance segmentation.To overcome the challenges of obtaining real underwater sonar data,expensive and difficult-to-practice underwater robot equipment,and environmental uncertainties,a virtual simulation platform has been developed.This platform can synthesize sonar images at different noise levels and seabed states,and simulate experimental scenarios and algorithm optimizations under different parameter settings to improve the robustness and accuracy of sonar image segmentation algorithms.The virtual simulation platform has significant advantages over real experiments,such as convenient data acquisition,controllable environment,and low cost.It provides a more convenient research condition for underwater side-scan sonar image segmentation under long-tailed distribution.Additionally,the virtual simulation platform has been applied in the teaching practices of the courses "Robotics Fundamentals" and "Program Design and Problem Solving" in the School of Artificial Intelligence,helping students better understand the basic principles of robotics technology and program design,and improving their practical operation abilities.Overall,the proposed method and virtual simulation platform have great potential for further research and practical applications in the field of underwater sonar image segmentation.This paper presents the development of a virtual simulation platform to address the challenges associated with acquiring real side-scan sonar data,high equipment costs,and difficulties in practical implementation,as well as environmental uncertainty.The platform enables the synthesis of sonar images under different noise levels and seabed conditions and allows for simulation experiments and algorithm optimization under various parameter settings,with the aim of improving the robustness and accuracy of sonar image segmentation algorithms.The virtual simulation platform offers several advantages over experiments conducted in actual environments,including convenient data acquisition,a controllable environment,and low cost.This platform provides a more convenient research environment for studying underwater side-scan sonar image segmentation under a long-tailed distribution.Moreover,the virtual simulation platform has been implemented in the teaching practices of the "Fundamentals of Robotics" and "Programming and Problem Solving" courses in the School of Artificial Intelligence,helping students gain a better understanding of the basic principles of robotics technology and programming and improving their practical operational abilities.