Research On Infrared Dim Small Target Detection And Tracking Method Based On Deep Learning

The detection and tracking of infrared dim and small targets pose a profound challenge to the depth learning method because of the unique nature of their images.Background factors,mutual occlusion between multiple targets,the change of target motion attitude,the intrusion of non target objects and the uncertainty of ambient light are all factors that affect the detection and tracking of infrared dim and small targets.The core work of this paper is to study and improve the existing target detection and tracking algorithms to improve the performance in the task of infrared dim small target detection and tracking.Taking YOLO V5 and deepsort algorithms as the framework,aiming at the characteristics of infrared images and infrared weak and small targets,the above two deep learning frameworks are improved to better realize the detection and tracking of infrared weak and small targets in tasks.In the task of infrared dim and small target detection and tracking,aiming at the problem of the performance degradation of YOLO V5 algorithm,this paper introduces the fusion attention mechanism based on the YOLO V5 algorithm,and weights the image features in channel and space,so as to strengthen the effectiveness of the result of feature extraction and improve the ability of the network model to distinguish the target.In addition,aiming at the possible occlusion problem among multiple target objects,a regression box filtering method based on adaptive threshold non maximum suppression algorithm is proposed to solve the problem of mutual occlusion of target objects in multi-target detection task tracking.In the task of infrared dim small target tracking,in view of the fact that the infrared dim small target is easy to be affected by complex environment,multi-target object interaction and changeable target motion attitude,the key features of the target object are extracted by using deepsort as the framework and YOLO V5,and the position of the infrared dim small target in the next frame image is predicted by combining Kalman filter algorithm and Hungarian algorithm.Experiments show that the principal component analysis can significantly improve the tracking performance in the infrared small target tracking task.