以YOLOv3的架构为基础,提出了一种基于双通道的快速低空无人机检测识别方法(Dual-YOLOv3)。该方法将红外与可见光的无人机图像同时输入到深度残差网络中进行特征提取,对所提取的特征图进行融合以增强特征的表达能力,利用多尺度预测网络对无人机目标进行类别判断和位置回归,得到检测识别结果。在真实采集的双波段无人机数据集上进行对比实验,结果表明,采用平均融合的Dual-YOLOv3-D在mAP(mean of average precision)上较单一数据源的YOLOv3提升了约6.1%,检测速度约为27 s -1。

Using the YOLOv3 architecture, we propose a recognition method for fast low-altitude unmanned aerial vehicle (UAV) detection based on the dual channel (Dual-YOLOv3). In this method, the infrared and visible UAV images are simultaneously input into the deep residual network for feature extraction, and the extracted feature maps are fused to enhance the expression ability of the features. Then, the multi-scale prediction network is used to determine the classification and the position regression of the UAV targets. Finally, we obtain the detection and recognition results. Comparison experiments are conducted on the real collected dataset of dual-band UAVs. The results show that the mAP (mean of average precision) of Dual-YOLOv3-D with average fusion is improved by 6.1% as compared with that of YOLOv3 with the single data source; the detection speed is approximately 27 s ^-1.