摘 要:为了解决复杂背景下,绝缘子准确快速识别的实时性问题,提出了一种基于 YOLOv5 改进的轻量型绝缘子检测算法模型。在网络结构中融入了 Shufflenet v2 网络和深度卷积模块,通过控制通道数和减少网络层数来减少参数量,采用 K-means算法调整 anchor 框,并提出了改进损失函数 DCIoU 加速了损失函数的收敛。实验结果表明,改进的 YOLOv5 算法在参数量上仅有原网络的 10%,准确率提高了 0.2%,推理速度提升了 2 帧。
关键词:深度学习;目标检测;绝缘子;YOLOv5
DOI:10.19850/j.cnki.2096-4706.2023.06.019
基金项目:湖北省教育厅科学技术研究计划项目(B2016092)
中图分类号:TP18;TP391.4 文献标识码:A 文章编号:2096-4706(2023)06-0073-04
Fast Detection of Insulators Based on Improved YOLOv5
HUANG Shiyi, DONG Xiaojie, YANG Longhuan, WANG Yifan
(College of Intelligent Systems Science and Engineering, Hubei Minzu University, Enshi 445000, China)
Abstract: In order to solve the real-time problem of accurate and rapid identification of insulators under the complex background, an improved lightweight insulator detection algorithm model based on YOLOv5 is proposed. Shufflenet v2 network and deep convolution modules are integrated into the network structure to reduce the number of parameters by controlling the number of channels and reducing the number of network layers. K-means algorithm is adopted to adjust the anchor box, and the improved loss function DCIoU is proposed to accelerate the convergence of the loss function. Experimental results show that the improved YOLOv5 algorithm is only 10% of the original network in terms of the number of parameters, the accuracy rate is increased by 0.2%, and the inference speed is increased by 2 frames.
Keywords: deep learning; target detection; insulator; YOLOv5
参考文献:
[1] 中国电力企业联合会 . 中国电力行业年度报告 2020 [R/OL].(2020-06-12).https://cec.org.cn/detail/index.html?3-284214.
[2] 邵瑰玮,刘壮,付晶,等 . 架空输电线路无人机巡检技术研究进展 [J]. 高电压技术,2020,46(1):14-22.
[3] 陈伟骏,周长胜,黄宏博,等 . 基于卷积神经网络的目标检测算法综述 [J]. 北京信息科技大学学报:自然科学版,2020,35(2):90-96.
[4] YANG Y L,WANG L J,WANG Y,et al. Insulator SelfShattering Detection:A Deep Convolutional Neural Network Approach [J]. Multimedia Tools and Applications,2019,78(8):10097-10112.
[5] 林志成,缪希仁,江灏,等 . 多模型融合的输电线路绝缘子自爆故障检测 [J]. 福州大学学报:自然科学版,2020,48(2):217-223.
[6] 唐小煜,熊浩良,黄锐珊,等 . 基于改进的 U-Net 和YOLOv5 的绝缘子掩模获取与缺陷检测 [J]. 数据采集与处理,2021,36(5):1041-1049.
[7] MA N N,ZHANG X Y,ZHENG H-T,et al. ShuffleNet V2: Practical Guidelines for Efficient CNN Architecture Design [J/OL].arXiv: 1807.11164 [cs.CV].[2022-01-03].https://arxiv.org/abs/1807.11164.
作者简介:黄施懿(1998—),男,汉族,江苏南通人,硕士研究生在读,研究方向:目标检测与人工智能;董效杰(1978—),男,汉族,上海人,讲师,博士,研究方向:红外图像与小目标特征;杨龙欢(1996—),男,汉族,四川成都人,硕士研究生在读,研究方向:目标检测与人工智能;王一帆(1998—),男,汉族,陕西西安人,硕士研究生在读,研究方向:目标检测与人工智能。