Research On Ship Target Detection And Recognition Methods In Spaceborne SAR Images

Spaceborne Synthetic Aperture Radar(SAR)is widely used in ship target monitoring as a microwave remote sensor capable of observing targets all day long,however,the noise,ocean clutter and other interferences in the spaceborne SAR image,as well as the scale transformation of ship targets all increase the difficulty of target detection and recognition.These problems all put forward a stronger demand for the interpretation technology of spaceborne SAR images.Therefore,this article focuses on spaceborne SAR images and conducts in-depth research on ship target detection and recognition methods.In terms of ship target detection in spaceborne SAR images,aiming at the detection problem of Constant False Alarm Rate(CFAR)algorithm in a densely distributed environment of multiple ships,a fast iterative CFAR-based ship target detection method is proposed.The detection threshold is continuously updated by iterative outliers to improve detection accuracy,and the calculation of parameter estimation and initial detection threshold is improved to improve detection efficiency.Using experiments to compare with the two-parameter CFAR algorithm and iterative detection method,the detection method based on fast iterative CFAR has ideal detection effects in multi-target areas.Aiming at the detection problems of deep learning algorithms for small-scale ship targets and complex background environments,a ship target detection method based on Receptive Attention Pyramid Network(RAPN)is proposed.Using Feature Pyramid Networks(FPN)for deep fusion of different levels of feature maps to enhance the detection capabilities of small-scale ship targets,and improve the horizontal connection method in FPN,and enhance the e xtraction and refinement of features by combining with Receptive Feld Block(RFB)and Convolutional Block Attention Module(CBAM).Compared with Faster R-CNN and FPN on the SSDD dataset,the RAPN-based detection method has obvious detection advantages in both far-shore and near-shore scenes.The comparison and analysis of ship target detection methods based on fast iterative CFAR and RAPN verify that the two methods have excellent detection performance for ship targets in spaceborne SAR images.In terms of ship target recognition in spaceborne SAR images,a method of ship target recognition based on WGAN-GP and Convolutional Neural Network(CNN)is proposed in view of the situation that space-borne SAR images have less classification data.Use WGAN-GP to complete the pre-training of CNN to reduce the scale of the annotation data with ship target cat egories required for recognition network training.A variety of experiments on the FUSAR-ship data set verify the effectiveness and advancement of the recognition method based on WGAN-GP and CNN for ship target recognition in spaceborne SAR images.