摘要
为了提高巡检绝缘子质量检测的安全性,设计了一种利用无人机射线检测系统。选择YOLOv3算法来实现人机电网巡检绝缘子缺陷检测的功能,再根据准确率、误报率、召回率、漏报率指标来综合评价DCNN算法处理性能。研究结果表明,YOLOv3获得了比R-CNN与FasterR-CNN网络更优的性能,采用YOLOv3算法得到的缺陷绝缘子AP为99.2%。YOLOv3网络具备更高检测精度与定位精度,有效控制训练误差,对于巡检绝缘子具备更优检测性能。YOLOv3网络帧速率为47.63 f/s,达到了R-CNN的24倍以及FasterR-CNN的12倍,充分满足实时检测的性能;YOLOv3比FasterR-CNN的误报率增大了0.81%,不过漏报率减小了1.4%。该研究能取代人工检测,大大降低了危险系数,具有很好的实际应用价值。
In order to improve the safety of inspection insulator quality,a radiographic detection system using un⁃manned aerial vehicle(UAV)was designed.The YOLOv3 algorithm was selected to realize the function of insulator defect detection in man-machine power grid inspection,and then the processing performance of DCNN algorithm was comprehensively evaluated according to the accuracy rate,false positive rate,recall rate and false negative rate.The results showed that YOLOv3 had better performance than R-CNN and FasterR-CNN networks,and the defec⁃tive insulator AP obtained by YOLOv3 algorithm was 99.2%.YOLOv3 network had higher detection and positioning accuracy,effectively controled training errors,and had better detection performance for inspection insulators.The frame rate of YOLOv3 network was 47.63 f/s,which was 24 times that of R-CNN and 12 times that of Fast⁃erR-CNN,fully meeting the performance of real-time detection.Compared with FasterR-CNN,the false positive rate of YOLOv3 increased by 0.81%,but the false positive rate decreased by 1.4%.This research can replace manu⁃al detection,greatly reduce the risk factor,and has good practical application value.
作者
窦国贤
李小威
宋杰
黄杨翼
杨彬彬
DOU Guoxian;LI Xiaowei;SONG Jie;HUANG Yangyi;YANG Binbin(Anhui Jiyuan Software Co.,Ltd.,Hefei 230061,China)
出处
《粘接》
CAS
2024年第6期182-184,192,共4页
Adhesion
