基于旋转YOLOv5的电力作业车态势感知方法研究

Research on Situation Awareness Method of Electric Work Vehicle Based on Rotated YOLOv5

摘要:

针对电力作业场景下的电力作业车安全监测问题，提出了一种基于YOLOv5的旋转目标检测算法，通过检测电力作业车的机械臂旋转角度，判断电力作业车在当前作业场景下的作业姿态是否安全。文中以YOLOv5为主干网络，采用环形平滑标签的方法，将角度的边界回归问题转化为分类问题，解决了角度周期性变化带来的损失值突变问题。实验结果表明，CSPDarkNetYOLOv5l模型对电力作业车机械臂的平均检测精度达到了80.02%，均优于GhostNetYOLOv5l、MobieleNetV3YOLOv5l、ShuffleNetV2YOLOv5l，并且对机械臂的旋转角度预测也最接近真实值。

Abstract:

Aiming at the problems of safety monitoring of electric work vehicle in power operation scenes, a rotating target detection algorithm based on YOLOv5 is proposed. By detecting the rotation angle of mechanical arm of electric work vehicle, it can judge whether the working posture of electric work vehicle in the current operation scene is safe or not. The data annotation method of YOLOv5 is modified, and the circular smooth label method is adopted to transform the boundary regression problem of angle into a classification problem, so as to solve the sudden change of loss value caused by periodic change of angle. The experiment results show that the average detection accuracy of CSPDarkNetYOLOv5l model for the mechanical arm of electric work vehicle reaches 80.02%, which is better than GhostNetYOLOv5l, MobieleNetV3YOLOv5l and ShuffleNetV2YOLOv5l, and the prediction for the rotation angle of mechanical arm is also the closest to the real value. Key words：rotating target detection; angle prediction; circular smooth label; power operation