基于深度可分离卷积的轻量级YOLOv3输电线路鸟巢检测方法  

Bird’s Nest Detection on Lightweight YOLOv3 Transmission Line Based on Deep Separable Convolution

作　　者：杨学存[1] 和沛栋 陈丽媛 李杰华 YANG Xuecun;HE Peidong;CHEN Liyuan;LI Jiehua(School of Electrical and Control Engineering,Xi’an University of Science and Technology,Xi’an 710000,China)

机构地区：[1]西安科技大学电气与控制工程学院,陕西西安710000

出　　处：《智慧电力》2021年第12期88-95,共8页Smart Power

基　　金：国家自然科学基金资助项目(51804250)。

摘　　要：针对输电线路无人机巡检图像鸟巢检测现有方法实时性差及小目标检测能力较弱的问题,提出一种基于深度可分离卷积的轻量级YOLOv3输电线路鸟巢检测方法。首先,使用Mosaic数据增强方法增强数据集并变相提升训练集中小目标的数量;然后,在主干特征提取网络使用深度可分离卷积代替部分标准卷积,提高检测网络的速度,并降低网络参数量从而降低权重文件内存,再使用PANet代替FPN,进一步提升特征融合的能力,增强对小目标的检测能力;最后,使用标签平滑进行训练,解决由于极少量标签错误导致的网络过度自信问题和网络过拟合问题。将某供电局无人机巡检视频剪切成图像制作数据集,使用本文算法与原始YOLOv3算法进行比较,并做消融实验。实验结果表明,本文的算法逐步提升了模型的速度和精度。Targeting the problems of poor real-time performance and weak small target detection ability existing in current methods for bird’s nest detection on transmission lines using UAV imagery,the paper proposes the nest detection method for lightweight YOLOv3 transmission line based on deep separable convolution.Firstly,Mosaic data augmentation method is used to enhance the dataset and improve the number of small targets in training set.Then deep separable convolution is used to replace part of standard convolution in main feature extraction network to improve the speed of a detection network and reduce the number of network parameters,reducing the weight file size.In addition,PANet is used instead of FPN to further improve the ability of feature fusion and enhance the detection ability of the small targets.Finally,label smoothing is used for training to solve the problems of the overconfidence and overfitting of the network caused by a small number of labeling errors.The UAV inspection video of a power supply bureau is cut into an image to make a data set.The proposed algorithm is compared with the original YOLOv3 algorithm,and the ablation experiment is carried out.The results show that the proposed algorithm improves gradually the speed and accuracy of the model.

关 键 词：无人机巡检 YOLOv3 深度可分离卷积 PANet 目标检测 

分 类 号：TM755[电气工程—电力系统及自动化]