基于YOLOv5l_CA的无人机目标检测算法

Target Detection Algorithm for Drones Based on YOLOv5l_CA

针对“黑飞”无人机目标小、尺度小、检测困难的问题,提出了一种基于YOLOv5l_CA的无人机目标检测方法。首先,建立无人机目标数据集,并对无人机进行标注;其次,运用YOLOv5l作为无人机训练时的基础网络,并设置好网络模型的超参数,用于模型的训练和评估;然后,对YOLOv5l进行改进,形成改进网络YOLOv5l_CA,它引入坐标注意力机制来增加天空下无人机的检测精度;最后,与其他网络模型进行对比实验。实验表明:改进的YOLOv5l_CA算法平均精度达到94.6%,分别高于YOLOv5s、YOLOv5m、YOLOv5l算法2.4%、1.9%、0.8%,有良好的表现,且满足了检测的实时性,验证了改进算法对无人机检测的可行性。

To address the issue of detecting unauthorized "black flight" drones,a drone target detection method based on YOLOv5l is proposed.Firstly,a drone target dataset is established,and the drones are annotated.Next,YOLOv5l is applied as the base network for drone training,and the network model's hyperparameters are set for model training and evaluation.Then,YOLOv5l was improved to form an improved network YOLOv5l_CA,which introduced a coordinate attention mechanism to increase the detection accuracy of drones under the sky.Finally,comparative experiments are conducted with other network models.Experimental results show that the improved algorithm YOLOv5l_CA achieves an average accuracy of 94.6%,which is 2.4%,1.9%,and 0.8% higher than YOLOv5s,YOLOv5m,and YOLOv5l algorithms,respectively.It performs well and meets the real-time detection requirements,verifying the feasibility of the improved algorithm for drone detection.