| The popularity of unmanned aerial vehicles not only brings convenience to people,but also causes safety risks.Various "black flying" accidents happen from time to time.In order to avoid the loss caused by UAV,it is very meaningful to carry out the detection and tracking research on UAV.The detection and tracking field has been developed for many years,and radar and other means have been quite mature.However,when it is applied to UAV,such as low,slow and small targets,it is not acclimated to the environment.In this context,this paper starts from visual means and combines the current popular deep learning method to explore the methods used for UAV detection and tracking.The main work is as follows:1.Based on the spatial pyramid module and lightweight feature extraction module,an anchor-free frame detection network MS-Center Net is designed.Starting from practical application,the network is optimized for UAV target detection in this paper.The network model obtained in the end is not only of small number of parameters and computation,but also of higher precision than before optimization.2.Based on GFL loss,a twin neural network for tracking-G-SIAMCPP is designed which solves the problem of asynchronization between classification and quality assessment in the original network structure,and uses a more general distribution to replace the special Dirac distribution in the prediction of border position.Therefore,the network positioning accuracy is higher.3.Based on the idea of redetection,the LG-Si AMCPP network is designed by combining MS-CENNET and G-Si AMCPP,which can alleviate the offset problem caused by error accumulation in the tracking process or the tracking failure problem caused by the non-updating of the tracking template.In the tracking process,the new network can update the changed template in time and correct the position error of the template in time,showing better robustness.Finally,the experiment shows that the method designed in this paper for the detection and tracking of UAV can maintain a good speed with high accuracy.Compared with other methods,it also has certain advantages.On the whole,this method has a higher practical value,and it is a step forward on the road of realizing intelligent UAV countermeasures. |