改进实时目标检测算法的电力巡检鸟巢检测

Bird nest detection based on improved YOLO algorithm

本文提出一种基于深度学习算法的实时目标检测模型,用于电力无人机巡检中鸟巢的自动检测。通过基于距离的K-means聚类算法,对数据集的标记框重新聚类,获得了更适用于识别不同杆塔在多种所处环境下的鸟巢的锚点集合。检测结果表明,使用新集合的算法均值平均精度提高至0.896,同时召回率和平均交并比均有提高;且运用本文算法可对巡检视频进行实时化处理(单帧处理时间低于30ms),便于后续问题的实时分析及处理,满足电力巡检智能化、常态化应用需求。

This paper proposes an object detection model based on YOLO deep learning algorithm, which is used for automatic detection of nests in electric drone inspection. Through the distance-based K-means clustering algorithm, the marker sets of the dataset are re-clustered. And the anchor set which is more suitable for identifying the nests in different environments of different towers is obtained. The test results show that the mean average precision(MAP) of the algorithm using the new set is increased to 0.896 while the recall rate and the average cross-over ratio are improved. The algorithm processing is close to real-time(less than 30 ms), which can maintain the application requirements of intelligent and normalized power inspection.