Research On SAR Ship Target Detection Method Based On Deep Learning

Ship detection in synthetic aperture radar(Synthetic Aperture Radar,SAR)images is currently the most important part of maritime applications,and is of great significance in the civil and military fields.The large span of different target scales in SAR ship images and the insensitivity of traditional detection methods to small targets are two reasons why multi-scale detection methods have become the mainstream research direction.In recent years,convolutional neural network(Convolutional Neural Network,CNN)-based target detection has achieved remarkable success in the field of optics.CNN-based ship detection methods have been applied in Vietnam due to their automatic feature extraction and high detection accuracy.In the increasingly complex sea environment,it has gradually become an inevitable trend in SAR image ship detection research.In view of the ability of CNN to automatically learn features,this paper proposes a ship target detection algorithm based on deep learning.The related research content and progress are as follows:Firstly,This paper summarizes the research status of ship target detection in SAR images,analyzes the advanta ges and disadvantages of existing ship detection methods,and discusses in detail the key factors affecting the quality of SAR ship images.Secondly,Aiming at the missed detection problem caused by micro-scale ship targets in SAR images,this paper proposes an improved convolutional neural network SAR image ship detection method based on region suggestions.The algorithm framework of Faster R-CNN in image detection is analyzed in detail.In order to realize ship detection and recognition on SAR images,it combines the characteristics of SAR ship images and proposes improvements for the existing micro-scale problems.Finally,the improved method is used to realize the SAR ship detection and the effectiveness of the improved method in the SAR image ship target detection is verified through experiments.Finally,Aiming at the problem of detection performance degradation caused by massive deep fusion features,an improved convolutional neural network SAR image ship detection method based on frame regression is proposed.This paper analyzes the algorithm framework of object detection algorithms based on border regression: YO LOv3(You Only Look O nce version 3)and YOLO v4(You Only Look O nce version 4)in image detection,and focuses on two depth feature selection methods: SENet(Squeeze and Exception Networks)and CBAM(Convolutional Block Attention Module),The feasibility of the improved method for ship detection in SAR image is verified by experiments;At the same time,compared with the exis ting ship detection methods.The experimental results show that the proposed method has obvious advantages over the existing methods in different SAR image scenes such as inshore,islands and reefs,and has higher detection accuracy.