Research On Aircraft Detection In Large Scene SAR Image Based On Deep Learning And Similarity Measurement

With the rapid development of SAR system and imaging technology in recent years,the resolution of SAR images has gradually improved,the imaging quality has become better,and the available information in SAR images has increased.Aircraft is a very important target in both military and civilian fields,and aircraft detection is the main point of research in SAR image interpretation.However,due to the complexity of the background contained in the large scene SAR images,the sensitivity of the aircraft posture and the limited number of aircraft,it is still a difficult task to detect aircraft targets in large scene SAR images.Traditional aircraft detection methods have low detection accuracy and usually cannot meet practical application requirements.This paper studies the aircraft detection method in large scene SAR images Based on deep learning algorithm.The main research contents of the paper are as follows:First,the characteristics of airports and aircraft targets in large-scene SAR images are analyzed,and some typical characteristics are abstractly represented,which lays the foundation for the subsequent detection of airports and aircraft targets in large-scene SAR images.The relevant intelligent learning algorithms of airport location and aircraft detection are introduced,which provides a theoretical basis for the research of this topic.Then,in order to eliminate the interference outside the airport area and quickly and accurately locate the area where the airport is located in the large scene SAR image,this paper combines the characteristics of the airport data in the SAR image to construct an airport segmentation network model named GU-Net.The model is used to segment the airport runway and apron in the large scene SAR image.Morphological processing and regional connectivity are performed on the segmentation results.And the airport area is obtained according to the airport identification operator abstracted from the airport characteristics.Finally,YOLOv4-tiny is used to detect aircraft targets in the airport area of SAR images,and the YOLO model trained by a small number of samples has weak generalization ability and poor robustness.Although it can locate most aircraft targets in SAR images with high false alarm rate,to solve this problem,the aircraft similarity measurement algorithm is proposed to optimize the detection results of YOLO.Aircraft similarity measurement algorithm comprehensively considers information such as the spatial arrangement and scattering characteristics of aircraft,and uses the similarity between aircraft and the difference between aircraft false alarm targets to correct YOLO’s aircraft detection results,to a certain extent.Solved the problem of sensitive model confidence threshold and limited detection performance caused by insufficient training of the YOLO model.The experimental results show that combining the aircraft similarity measurement algorithm with YOLO can improve the performance of aircraft detection in SAR images.