Ship Target Detection Technology In SAR Image Based On Depth Feature Enhancement

Synthetic Aperture Radar(SAR)can observe the earth all day without being affected by the weather,and it plays an important role in the detection of marine ship targets.In the military field,it can scout territorial sea areas,prevent military invasion,and safeguard sovereign interests;in the civilian field,it can supervise maritime transportation and prevent illegal activities at sea.With the advancement of satellite remote sensing technology and the launch of satellites such as Gaofen-3,the quality and quantity of SAR images have been improved,which has promoted the development of ship target detection technology in SAR images.However,the representation of SAR images is not intuitive,the sea and land scenes are complex and changeable,and there are many false alarms.In addition,there are many types of ship targets in SAR images,and the scale differences within the class are large.In this paper,deep learning technology is used to realize ship target detection in SAR image.Aiming at the above problems and challenges,a ship target detection technology in SAR image based on depth feature enhancement is proposed.The main research work and achievements are as follows:(1)Aiming at the problem that the SAR image scene is complex and susceptible to background interference such as ground objects and noise,this paper proposes to construct deep visual features to enhance the expression of target characteristics.Improvements are made to the YOLOv5 basic network,and a spatial coordination attention mechanism is introduced into the backbone feature extraction network to extract remote dependencies in different directions,thereby effectively enhancing the position information of ship targets and improving the detection ability of ship targets.(2)Aiming at the difficulties that there are many types of ship targets in SAR images,the scale differences within the class are large,and the proportion of targets in the large field of view is small,this paper proposes a weighted feature pyramid of twoway cooperative perception for multi-scale feature fusion.By fusing feature information of different scales through two-way channels,a variety of receptive fields are established.The jumping horizontal connection method is adopted to effectively alleviate the loss of shallow feature information.At the same time,learnable weights are added to constrain the feature maps of different scales participating in the fusion,and the algorithm realizes reliable prediction of multi-scale ships and small targets.(3)Aiming at the problem of false alarms in the detection results,this paper focuses on analyzing the characteristics of difficult false alarms,and proposes a false alarm elimination method based on the improved Rep VGG lightweight recognition network.By embedding an efficient channel attention module in the Rep VGG network,the feature correlation between channels is improved,and the structure reparameterization method is used to convert the model into a single-way structure in the inference stage to speed up the inference speed.The confidence threshold strategy is adopted to perform secondary discrimination on the detection and screening targets,and some difficult false alarms are eliminated.Through experimental verification,this method effectively improves the overall performance of the ship target detection algorithm.Based on the research of the above key technologies,this paper implements a set of algorithm software for ship target detection in SAR images.Through quantitative and qualitative experimental analysis on the large public SAR ship datasets AIRSARShip-1.0 and AIR-SARShip-2.0,the effectiveness of the algorithm proposed in this paper for solving related problems is systematically verified.