Research On Ship Detection Method For Complex Scenes In Microwave Remote Sensing Data

Synthetic aperture radar(SAR)is a typical microwave remote sensing technology,which has the advantages of all-day and all-weather work.SAR enables wide-range imaging and plays an important role in earth monitoring.Marine monitoring is an essential part of earth monitoring,which has a profound impact on the development and progress of human beings.Ship is the main carrier for maritime activities,it has naturally become an indispensable part of marine monitoring.Ship detection is a primary task in ship monitoring,it has great significance in the organization and management of maritime activities,as well as the maintenance of national maritime rights and interests.SAR-based ship detection has proven to be feasible,effective,and critical.It occupies an important position in the work of ship detection.In recent years,SAR ship detection method based on convolutional neural network(CNN)has been widely studied and progressed significantly,but there is still room for improvement in complex scenes.Inshore and dense scenes are more common in practical tasks.Therefore,to improve the performance of ship detection in inshore and dense scenes,this paper applies CNN object detection method to conduct method research for the characteristics of ships in the two scenes.The followings are the main research content of this paper.A large number of CNN-based SAR ship detection methods have been proposed.But a systematic summary is lacking,which is not conducive to understanding the current research situation and technologies.This paper collects and organizes related works,then analyzes the current situation from multiple perspectives,including the design of detection methods,strategies to optimize detection performance,and solutions to detection difficulties.Based on the progress made by existing methods,the future work direction in this field is predicted to promote the solution of corresponding problems.For ship detection in inshore scenes,this paper considers feature optimization.The main problem faced in inshore scenes is that the land buildings have similar characteristics to ship targets.Therefore,this paper designs the mixed convolution channel attention(MCCA),which reweights each channel of the feature map by considering the connection between each channel of the feature map and all channels,and neighbor channels.MCCA aims to highlight important feature map channels and suppress useless feature map channels.For enhancing the feature map extraction,MCCA is embedded in the backbone network to highlight the differences between ships and backgrounds in feature maps,so that the network can identify ship targets more accurately.For ship detection in dense scenes,this paper chooses to optimize from the post-processing stage.It is necessary to preserve good detection result for each ship target and remove redundancy.By considering the geometric position relationship of ships in dense scenes,this paper designs geometric nonmaximum suppression(G-NMS),which performs corresponding processing operations according to different conditions,effectively removing redundancy while reducing missed ships.Experimentation on two datasets and performing detailed,in-depth analysis of the results.Experiments have proved that the MCCA designed in this paper is conducive to optimizing feature extraction and improving the performance of ship detection in inshore scenes,G-NMS can help improve the problem of missed ships in dense scenes.In addition,MCCA and G-NMS can be combined and complement each other to achieve better SAR ship detection performance.