摘要
基于深度学习的输电线路缺陷检测方法往往需要昂贵的硬件支持来实现实时和高精度的技术指标要求,难以大规模推广和普及。针对这一问题,提出一种改进的轻量级YOLOv4神经网络模型方法。具体的,为提高检测精度,在特征提取网络中增加了轻量级注意力机制;为提高检测速度,选用轻量级Mobilenet-v3网络作为YOLOv4的骨干网,同时采用了深度可分离卷积代替了传统YOLOv4模型中的多尺度特征金字塔结构普通卷积。实验结果表明,改进的轻量级YOLOv4模型能有效检测出输电线路中的鸟巢,平均检测精度(AP)值达到97.56%,相较于YOLOv4模型提高了1.80%,检测速度达到YOLOv3的45.8倍,是YOLOv4的56.6倍,具有较强的实际应用价值。
Transmission line defect detection methods based on deep learning often need expensive hardware support to achieve real time and high precision technical requirements,so it is difficult to be popularized on a large scale.To solve this problem,an improved lightweight YOLOv4 neural network model is proposed.Specifically,in order to improve the detection accuracy,a lightweight attention mechanism is added to the feature extraction network.In order to improve the detection speed,lightweight Mobilenet v3 network is selected as the backbone network of YOLOv4,and deep deprivable convolution is adopted to replace the traditional multi scale characteristic pyramid convolution in YOLOv4 model.The experimental results show that the improved lightweight YOLOv4 model can effectively detect nests in transmission lines,and the average detection accuracy(AP)value reaches 97.56%,which is 1.80%higher than that of YOLOv4 model.The detection speed is 45.8 times that of YOLOv3 and 56.6 times that of YOLOv4,which has strong practical application value.
作者
武明虎
雷常鼎
刘聪
WU Minghu;LEI Changding;LIU Cong(School of Electrical and Electronic Engineering,Hubei Univ.of Tech.,Wuhan 430068,China)
出处
《湖北工业大学学报》
2023年第2期11-16,共6页
Journal of Hubei University of Technology
