UAV Detection Method Based On Computer Vision Technology

In recent years,UAV technology has developed rapidly and is widely used in various industries.However,with the continuous development of UAV technology,there are more and more terrorist attacks and armed conflicts caused by UAVs,as well as black flight incidents and privacy leaks at airports,and the need for UAV technology is becoming stronger and stronger.Therefore,it is important to study accurate UAV detection methods for national defense and social security.However,due to the characteristics of "low flight trajectory,slow movement speed and small size",it is difficult to detect UAVs.To address the shortcomings of current UAV detection methods,this thesis proposes an UAV detection method based on computer vision technology,which aims to solve the problems as the UAV is difficult to detect due to its small size,detection accuracy is limited by light,easy to confuse with birds flying in the air,detection speed is difficult to meet the needs of actual application scenarios and difficult to track,etc.The specific solutions are as follows:1.In order to solve the problem that detection is difficult at night or under dim light conditions,two data sets of visible light and thermal image are established in this thesis for the generation of two models of day and night.The Det-Fly public dataset is added to supplement the self-collected dataset,which in total contained data from 23,858 images.At the same time,two data enhancement methods are added.In order to make up the detection gap between the thermal image model and the visible light model,the gray visible light image data enhancement technology is added to the thermal image data set.And to further improve the background and location diversity of the dataset,a data enhancement method for small targets is added to both datasets.The experimental results can prove the effectiveness of the two data enhancement methods.2.According to the requirements of detection speed in practical application scenarios,this thesis selects the YOLOv5 algorithm with faster detection speed to implement the UAV target detection model.By testing different versions of YOLOv5 and input resolutions,the optimal parameters suitable for UAV detection are determined,and two detection models of visible light and thermal image were obtained.Meanwhile,the BN layer-based model pruning method is added to further improve the detection speed of the model.In addition,in order to solve the problem of confusing UAVs with birds,this thesis selects the Efficient Net algorithm to perform secondary classification of YOLOv5 detection results,which can effectively improve the detection accuracy.3.To solve the problems of possible occlusion,overlap,or unclear in UAV detection,this thesis introduces a target tracking algorithm to achieve real-time tracking of UAV targets.The Deep SORT algorithm is selected and improved by introducing DIo U matching to improve the accuracy of predicted trajectory and reduce the number of lost targets.The experimental results show that the proposed method has high accuracy and practicality in real-time UAV target tracking.To better demonstrate the tracking effect,this thesis visualizes the tracking system based on Python GUI,provides visible light and thermal image mode selection,and can precisely locate each UAV target that appears in the field of view and track it in real time.Overall,this thesis proposes an UAV detection method based on computer vision technology,which addresses the common detection difficulties in the field of UAV detection through data set optimization processing,model selection and pruning,and detection+tracking mode.The experimental results show that the proposed algorithm has certain advantages in terms of accuracy and speed,and the method has certain practical application value.It can provide some reference and guidance for the development of UAV technology.