Research On Ground Military Target Recognition Tracking Algorithm Based On Yolov5

In recent years,deep learning and artificial intelligence technology have developed rapidly,and computer vision has played a key role in it.As a hot spot in the field of computer vision,target recognition and tracking technology has received widespread attention,and the technology has been widely used in communication engineering,aerospace,military weapons and other fields.Especially in the military field,target identification and tracking technology is a key technology for the current research and development of military equipment,and it is also an important research direction for future war changes.Nowadays,there are many target recognition and tracking algorithms,which have better performance than traditional methods in the past,and can achieve target recognition and tracking in the case of limited data samples and complex scenarios,with high accuracy.At present,in China,with the emergence of new weapons such as automated weapons,unmanned weapons,and patrol missiles,target recognition and tracking technology based on deep learning is more meaningful and valuable for research.However,in actual combat,the real-time requirements for the recognition algorithm are high,and factors such as scale change and target occlusion will have a greater impact on the tracking algorithm,which makes the target recognition tracking technology still have great challenges.Therefore,this project takes ground military equipment such as tanks and armored vehicles as the research object,and carries out the research of ground military target recognition and tracking algorithm based on yolov5,and studies the three problems of small target recognition,target occlusion and scale change faced by target recognition and tracking technology at this stage.In the target recognition stage,aiming at the problem of poor recognition of small targets by yolov5 algorithm,an improved yolov5 algorithm based on convolutional block attention module(CBAM)is proposed.Experiments show that compared with the yolov5 algorithm,the improved CBAM-yolov5 algorithm has an increase in accuracy of 2%,a recall increase of 0.5%,a m AP@0.5 increase of 1.3%,and a m AP@0.5:0.95 improvement of 1.2%.This shows that the addition of attention mechanism(CBAM)module improves the recognition effect of small targets significantly,and effectively solves the problem of poor recognition of small targets by yolov5 algorithm.In the target tracking stage,aiming at the influence of scale change and target occlusion on target tracking,an anti-occlusion KCF improved algorithm for scale change is proposed.In terms of target occlusion,this paper proposes an improved KCF algorithm that adds the occlusion criterion APCE and Kalman filter position prediction.Compared with the KCF algorithm,the accuracy of the KCF improved algorithm is increased by 6.2%,and the success rate is increased by 6.6%,indicating that the KCF improved algorithm can reduce the pollution and drift of the model to a certain extent,improve the anti-occlusion ability of the KCF algorithm,and prevent the target tracking failure.In terms of scale change,this project proposes to predict the optimal scale of the target by fusing the scale-dependent filter in the DSST target tracking algorithm on the basis of the KCF original algorithm.Compared with the KCF algorithm,the accuracy of the KCF improved algorithm is increased by 5.1%,and the success rate is increased by 5.7%,indicating that the KCF improved algorithm can perform scale changes and improve the defect of the fixed scale of the KCF algorithm.