Study Of Ship Target Detection Algorithm In SAR Image Based On Deep Learning

At present,deep learning algorithm has been widely used in target detection of SAR(Synthetic Aperture Radar)images.However,the current mainstream deep learningbased detection methods have problems such as high model complexity,slow detection speed and low detection performance for targets of special size.In view of the above problems,this thesis will focus on the deep learning-based ship target detection algorithm in SAR image,and the main work completed is as follows:(1)Aiming at the problem that traditional deep learning networks have high model complexity and slow detection speed,this thesis proposes a lightweight target detection network based on improved Fast-R-CNN,which ensures high detection accuracy and greatly improves target detection speed.Secondly,aiming at the problem of missed detection of special size targets in SAR images,a feature extraction network with feature relay amplification and multi-scale feature connection was designed to extract features of targets of different sizes in wide area scenes and improve the detection performance of the network to targets of different sizes.In addition,the identification and positioning task network is also targeted to improve the accuracy of positioning and detection speed.(2)Based on the open source SSDD(SAR Ship Detection Dataset),the proposed detection network,the traditional Faster R-CNN network and the SSD network were compared and analyzed in terms of detection accuracy,detection speed and detection performance of targets of different sizes.Experimental results show that,on the premise of ensuring high detection accuracy,the detection speed of the model proposed in this paper is greatly improved compared with Faster R-CNN and SSD models.And the model proposed in this thesis has obvious advantages for the detection performance of targets with special size.(3)Aiming at the problem that the size of the actual SAR image is too large to be directly fed into the deep learning network,a preprocessing model based on the actual SAR image is proposed.By preprocessing the actual SAR images,such as segmentation,pre-recognition and stitching,a fast,accurate and lightweight target detection network model suitable for actual SAR images is designed in combination with the target detection network proposed in this thesis.