Research Of Ship Detection In SAR Images Based On Depth Features

Ship detection based on Synthetic Aperture Radar(SAR)images has become one of the key technical problems of SAR image interpretation.It is also the most important part of the current marine application,which is of great significance in civil and military fields.Due to the fact that different ship targets of SAR images vary in scales and traditional ship detection methods are insensitive to small-scale ships,the research on multi-scale ship detection methods has been the focus of ship detection in SAR images.In recent years,the object detection algorithms based on the Convolutional Neu-ral Network(CNN)have achieved great success in the optical field.The ship detection method of SAR images based on CNN is suitable for more and more complex sea condi-tions with its characteristics of automatic feature extraction and high detection accuracy.Therefore,it has gradually become an inevitable trend in the research of ship detection in SAR images.Based on the characteristics of automatic learning feature of CNN,a ship detection method with the depth features of SAR images is designed in this paper.The main work is as follows:Firstly,the development of ship detection in SAR images is summarized,and the advantages and disadvantages of the existing ship detection methods are analyzed in this paper.Then,the basic theory of CNN and the main methods of depth feature extraction are introduced.In addition,the object detection methods based on deep learning are studied,and the effectiveness of the extracted depth feature to ship detection in SAR images is explored through experiments.Secondly,in order to solve the missed detection problem of small-scale ships caused by only using features from deep layers of CNN,the ship detection method in SAR images based on depth feature fusion is studied in this paper.And then,more details are illustrated for the element-wise addition and its typical application,feature pyramid network.Then,Dense Feature Pyramid Network based on concatenation is proposed,and the effectiveness of dense connection based on different depth feature fusion methods for ship detection in SAR images is explored through experiments.Finally,as for the problem of detection performance degradation caused by massive fused depth features,this paper focuses on the Convolutional Block Attention Module(CBAM),a depth feature selection method,which is integrated into the Dense Feature Pyramid Network,and thus the ship detection based on Dense Attention Pyramid Network(DAPN)is proposed.The feasibility of DAPN for multi-scale ship detection in SAR im-age is explored through experiments.At the same time,comparative experiments with the existing ship detection methods are set up.The experimental results show that the pro-posed method has higher detection accuracy and outperforms other methods in different scenes such as offshore and inshore scenes of SAR images.