基于深度学习的无人机入侵检测研究

Research on UAV Instrusion Dectection Based on Deep Learning

针对现有探测系统探测低空、慢速、小型、低RCS、弱红外辐射特性的无人机类目标的能力不足,无法有效地通过图像侦测到无人机入侵的问题,采用基于深度学习的方法来对无人机图像进行识别和检测,提出了以YOLOv5目标检测模型为基础同时结合ResNet50分类模型的无人机入侵检测方法。经过实验对比结果表明:该方法识别Drone-detection-dataset数据集中4种空中目标的准确率、召回率和均值平均精度值均高于原始YOLOv5模型;且相比于Faster-RCNN和SSD检测模型,YOLOv5+ResNet50检测和识别无人机图像的效果更优,特别是针对于远距离的无人机图像。因此YOLOv5+ResNet50可以有效地对入侵无人机进行检测和预警。

In view of the insufficient capability of existing detection system to detect UAV targets with low altitude,slow speed,low RCS and weak infrared radiation,and difficulties of detecting drones effectively through images,the method based on YOLOv5 and ResNet50 models was proposed,and the deep learning method was used to detect and identify UAV images.Comparing with original YOLOv5model,accuracy,recall and mAP of identifying aerial targets in Drone-detection-dataset are higher based on the proposed model.Besides,comparing with Faster-RCNN and SSD,YOLOv5+ResNet50 could reach better results of detecting UAV images,especially for distant UAV targets.As a result,YOLOv5+ResNet50 can be used to detect UAV efficiently.