Research On Target Recognition And Tracking Of SUAV By Swarm

With its small size and ultra-low altitude flight,SUAV by swarm can avoid most traditional detection and tracking methods,and is widely used in military operations.The Russian-Ukrainian battlefield shows that UAV technology is developing in the direction of autonomy,intelligence and clustering,so it is of great significance to carry out the research of vision-based SUAV by swarm target recognition and tracking algorithm.This paper proposes an algorithm to optimize the target recognition model and fuse the target tracking algorithm for ground-to-air recognition of fixed-wing SUAV by swarm,the inaccuracy of tracking UAVs,and the failure of reproduction tracking after UAVs fly out of the field of view.The main research contents are as follows:Aiming at the problem that the YOLOv5 s network model has poor recognition effect on small targets,the method of network layer merging is used firstly,and the problem of spatial insensitivity of nonlinear activation function and insufficient extraction of network features is solved by fusing convolutional layers and batch standardized layers to reduce the number of parameters and accelerate network reasoning.Secondly,the loss function is optimized by redefining the angle penalty cost and considering the angle between the expected regression,so as to improve the network convergence speed and the accuracy of network inference.The adaptive moment estimation method is used to solve the problem of network convergence falling into local optimality,and then improve the network structure,and fuse the shallow feature map with the deep feature map to further improve the defect of insufficient extraction of small target features by the model.The test results show that the improved model has a significant improvement in small target recognition.Aiming at the problems of target ID conversion and target loss in the tracking process,the cosine distance and Mahalanobis distance are used for feature matching to avoid the target ID conversion in the tracking process,and at the same time,DIOU is used to improve the coincidence accuracy and data integrity of the unmatched target recognition frame and the predicted tracking frame,and on this basis,the improved target recognition model and tracking algorithm are fused to make it have the ability to automatically identify and track multiple targets.The experimental results of SUAV by swarm target recognition and tracking show that the fusion algorithm can accurately and automatically identify and track SUAV by swarm under visible light conditions.