摘要
针对输电线路无人机巡检实时性和准确性的要求,深入研究了YOLOv3目标检测算法在无人机巡检机载AI模块中的应用。利用将目标检测候选区选取和对象识别合二为一的YOLOv3算法,结合多尺度特征融合方式实现了目标检测的高准确性和实时优化,并采用残差块解决了模型退化问题。输电线路绝缘子检测结果表明:YOLOv3算法平均精度可达90%,相同条件下YOLOv3算法平均处理速度约为Faster RCNN算法的3.2倍,约为SSD算法的1.6倍。
According to the requirements for instantaneity and accuracy by UAV inspection of transmission line,the application of YOLOv3 target detection algorithm in UAV inspection airborne AI module was deeply studied.Through the YOLOv3 algorithm for both object detection candidate region selection and object recognition,in combination with the multi-scale feature fusion method,the high accuracy and instantaneity optimization of target detection were realized,and the residual block was used to solve the problem of model degradation.The results of transmission line insulator detection show that the average accuracy of YOLOv3 algorithm can reach 90%.Under the same conditions,the average processing speed of YOLOv3 algorithm is 3.2 times that of Faster RCNN algorithm and 1.6 times that of SSD algorithm.
作者
王昊
丁国斌
杨家慧
WANG Hao;DING Guobin;YANG Jiahui(Digital Grid Research Institute Co.,Ltd.of China Southern Power Grid,Guangzhou 510700,Guangdong,China;R&D Department,Guangzhou Zhixun Information Technology Co.,Ltd.,Guangzhou 510700,Guangdong,China)
出处
《沈阳工业大学学报》
CAS
北大核心
2024年第1期49-53,共5页
Journal of Shenyang University of Technology
基金
国家自然科学基金青年科学基金项目(61163113)。
关键词
无人机巡检
目标检测
图像识别
多尺度特征融合
残差块
输电线路
实时性
UAV inspection
target detection
image recognition
multi-scale feature fusion
residual block
transmission line
instantaneity