Research And Application Of UAV Intrusion Early Warning Based On Deep Learning In Entity Security System

In recent years,with the development of economy and the progress of science and technology,the application range of UAV has gradually expanded from military field to civil field However,the imperfect supervision system of UAV leads to the frequent occurrence of “ black flight ” and “ indiscriminate flight ”,which poses a serious threat to the safety of major physical systems.Therefore,the research on UAV detection is of great significance in physical security system and so on.Radar detection,radio detection,photoelectric detection and other forms of detection methods have been applied to the UAV detection system,but there are also detection defects in detection distance and applicable conditions.Visual detection uses the camera to obtain the video of the monitoring area,and uses the image features to detect and track the UAV.Due to the low cost,strong universality,long detection distance and visual display of detection results,visual monitoring has become a research hotspot in the field of anti UAV monitoring.The system adopts different detection algorithms for different airspace characteristics,one is small-scale UAV detection in long-distance airspace,the other is UAV detection in complex background in short-distance airspace.The system realizes UAV monitoring in full airspace coverage.The main work completed is as follows:In order to better carry out the research of this subject,this paper collects and makes pictures of various types of UAVs.The UAV is photographed from multiple angles,multiple scenes and multiple scales,and then the video is processed into an image containing the UAV through software,and the data set is expanded by image preprocessing.Finally,all the image data are manually labeled to build a small UAV data set.A multi-scale detection network based on improved feature pyramid structure is studied.After analyzing the data set,it is found that UAV has the characteristics of multiscale change.When the scale of UAV is small,YOLOv3 detection network will have the phenomenon of false detection and missing detection,and the robustness is not strong enough.To solve the above problems,this paper adds a new detection layer to form a feature pyramid structure with a larger scale range and form a better multi-scale detection network.It not only improves the detection effect of small-scale UAV,but also improves the robustness of the detection model to multi-scale UAV detection.In order to solve the interference problem of UAV detection caused by complex background,on the basis of improving the multi-scale detection network of feature pyramid,the channel and spatial attention module are added to the feature extraction network,so that the network will pay more attention to the characteristics of foreground targets,so as to solve the interference problem of background.At the same time,the residual connection based on the second-order term is used to strengthen the nonlinearity of the feature extraction network and enhance the generalization performance of the model.The small target of UAV is tracked by multi-scale kernel correlation filters(KCF)algorithm.Aiming at the problem that the KCF algorithm is difficult to track the fastmoving UAV,an improved KCF algorithm based on Kalman filter is proposed to track the fast-moving target.The algorithm predicts the displacement of the target between the current frame and the next frame through Kalman filter,so as to estimate the position of the UAV in the next frame,and finally track the KCF at the estimated position.Experimental results show that the improved KCF algorithm can track fast moving targets,and the tracking accuracy is basically unchanged,and meets the real-time requirements of the system.Design and implement the overall scheme of the system.This paper combines the improved YOLOv3 target detection algorithm with KCF tracking algorithm to design a set of UAV intrusion early warning system.The system uses the moving target detection algorithm to detect the suspected target,and uses dual cameras to capture video and images.When the moving target is found,the deep learning detection algorithm with high robustness and strong generalization performance is used to identify,detect and track the UAV target.When the target is found to be UAV,an early warning signal will be sent to the security system.