Research On Ground Target Efficient Detection Method On Large Scene SAR Images

Synthetic Aperture Radar(S AR)is an active microwave sensor that can continuously and stably image targets,achieving all-weather and all-time remote sensing monitoring of the ground.Target detection,as the leading part of SAR image interpretation,has important research significance and application value.Traditional target detection methods are constrained by the detection mechanism and tend to produce a large number of false alarms and missed detections when facing large-scene SAR images.Moreover,the low detection efficiency making it difficult to meet the requirements of accurate and fast target detection.In order to address the aforementioned issues,this thesis focuses on SAR ground target detection in large scenes and conducts research on accurate detection and efficiency improvement.The main research work includes:1.The basic theory of efficient SAR target detection was studied.The characteristics of targets in large scene SAR images were analyzed,and the reasons for the limited performance of existing large scene SAR image detectors were examined.The mechanisms of traditional target detection methods and deep learning-based target detection methods were studied,and the efficiency improvement measures corresponding to these two types of methods were explained,laying the theoretical foundation for subsequent research.2.A ground target detection method for large scene SAR images was proposed.A detection in imaging framework was established for large scene SAR images.By gradually extracting potential target regions during the imaging process,the framework can achieve high-resolution imaging and detection of target areas.In addition,a scene discrimination and target detection network is designed,which adaptively changes the detection network structure by identifying the scene type,enabling accurate target detection in large scenes and reducing detection processing time.3.Efficiency improvement methods for SAR target detection networks were studied.The network structure was improved using depth separable convolution and inverse residual modules to achieve lightweight target detection network structures.Based on attention mechanisms,a network channel importance evaluation criterion was established,and L1 regularization was used for sparse training of the network to reduce the number of network channels.Finally,while ensuring detection performance,the computational complexity of the method was further reduced,achieving fast detection of SAR ground targets.The effectiveness of the proposed method has been verified through experiments.The experimental results show that the above method can achieve fast and accurate detection of SAR ground targets in large scenes.