摘要
为解决中小型绝缘子在检测目标时出现的漏检、错检问题,提出一种改进YOLOv4算法的中小型绝缘子检测方法。首先,通过增加特征融合层、引入SE注意力机制的方式改变YOLOv4网络结构,以发掘更有用的中小型绝缘子特征信息。其次,在模型结尾增加负例挖掘模块,抑制复杂背景干扰,提高绝缘子检测准确度。最后,使用改进k-means++算法重新聚类符合中小型绝缘子特征的先验框以加快模型收敛速度。实验结果表明,使用含负例挖掘模块的改进算法进行绝缘子检测,中目标和小目标的平均精度(Average precision,AP)分别达到88.91%和73.09%,对中小型绝缘子检测具有一定的参考价值。
In order to solve the problem of missed and wrong detection of small and medium-sized insulators in target detection,an improved YOLOv4 detection method for small and medium-sized insulators was proposed. By adding a feature fusion layer and introducing SE attention mechanism, the YOLOv4 network structure is changed to explore more useful feature information of small and medium insulators. Second, add a negative example mining module at the end of the model to suppress complex background interference and improve the accuracy of insulator detection. Finally, using improved k-means++ to re-cluster the prior boxes that fit the characteristics of small and medium insulators to speed up model convergence. The experimental results show that using the improved algorithm with negative example mining module for insulator detection, the Average precision(AP) of medium and small targets reach 88.91% and 73.09%, respectively, which has certain reference value for the detection of small and medium-sized insulators.
作者
李磊
李英娜
赵振刚
LI Lei;LI Yingna;ZHAO Zhengang(Faculty of Information Engineering and Automation,Kunming University of Science and Technology,Kunming 650500,China;Key Laboratory of Computer Technology Application of Yunnan Province,Kunming 650500,China)
出处
《电视技术》
2022年第3期74-82,共9页
Video Engineering
关键词
中小型绝缘子
改进YOLOv4
SE注意力
负例挖掘
small and medium-sized insulators
improved YOLOv4
SE attention
negative example mining
作者简介
李磊(1997-).男,硕士在读,研究方向为图像处理、信息物理融合;通信作者:李英娜(1973-),女,博士,教授,硕士生导师,研究方向为智能电力系统、计算机视觉、信息集成和智能分析等。E-mail:1044682866@qq.com。
