基于改进YOLO v5方法的电力设备红外图像识别方法

王小栋; 吕通发; 鲍明正; 何永春; 辛鹏; 吴涛

基于改进YOLO v5方法的电力设备红外图像识别方法

国网内蒙古东部电力有限公司兴安供电公司, 内蒙古兴安盟 137400

基金项目:

国网内蒙古东部电力有限公司科技项目 52664020001S

详细信息

作者简介:
王小栋（1975-），男，本科，高级工程师（副教授级），主要从事输变电运维检修技术方面的研究工作，E-mail：13951423028@139.com

中图分类号: TM85
计量
- 文章访问数: 112
- HTML全文浏览量: 17
- PDF下载量: 41
出版历程
- 收稿日期: 2022-03-29
- 修回日期: 2022-04-25
- 网络出版日期: 2024-06-23
- 刊出日期: 2024-06-19

Infrared Image Recognition Method for Power Equipment Based on Improved YOLO v5

State Grid Inner Mongolia Eastern Electric Power Co. Ltd., Xing'an Power Supply Company, Hinggan League, 137400, China

摘要

摘要:
为解决电力设备红外图像有遮挡、分类不准确和特征提取不充分等问题，本文提出一种改进的YOLO v5识别方法。首先通过迁移学习的方法，将电力设备可见光图像和红外图像相融合，接着将Triplet注意力机制嵌入到特征提取网络中，对关键特征信息进行加权强化，最后通过多尺度融合的方法实现不同目标的识别。研究结果表明：相对于Faster R-CNN和SSD，本文方法的识别精度和识别效率最高，且适应于复杂背景下的多类型电力设备识别；本文方法的模型仅4.1 MB，相较于SSD缩减了80.8%，实现了网络模型的轻量化。本文方法为电力设备红外图像智能检测提供了新颖可行的方案。
- 电力设备 /
- 红外图像 /
- 迁移学习 /
- YOLO v5s /
- 注意力机制 /
- 轻量化模型
Abstract:
This study proposes an improved YOLO v5 method to solve the problems of inaccurate classification and insufficient feature extraction from power equipment infrared images. First, the visible light data and infrared images of the power equipment were fused using the transfer learning method. The triplet attention mechanism was then embedded into the feature extraction network for weighted intensification of key feature information. Finally, different targets were identified using multiscale fusion. The results show that compared with faster R-CNN and SSD, the proposed method has higher recognition accuracy and efficiency and is suitable for image recognition of multi-type power equipment in complex backgrounds. This method realizes a lightweight network model with a size of only 4.1 MB, which is a reduction of 80.8% compared to that of SSD, providing a novel and feasible scheme for intelligent infrared image detection of power equipment.
- power equipment /
- infrared image /
- ttransfer learning /
- YOLO v5s /
- attention mechanism /
- light-weight model

HTML全文

图 1 多尺度融合处理结构

Figure 1. Processing structure of multi-scale fusion

下载: 全尺寸图片幻灯片

图 2 TA机制原理

Figure 2. Principle of TA mechanism

下载: 全尺寸图片幻灯片

图 3 DIOU_NMS损失函数

Figure 3. Loss function of DIOU_NMS

下载: 全尺寸图片幻灯片

图 4 多种电力设备的可见光图像

Figure 4. Visible light images of various electrical equipment

下载: 全尺寸图片幻灯片

图 5 不同算法的识别结果

Figure 5. Recognition results of different algorithms

下载: 全尺寸图片幻灯片

表 1 环境要求

Table 1 Environment configuration

Item	Configuration
OS	Windows 11
CPU	Intel Xeon Silver 4114T 12C
GPU	NVIDIA GTX1080Ti
RAM	64 GB
Deep learning framework	PyTorch 1.8
Hard disk	1T

下载: 导出CSV

表 2 不同改进方法的识别效果

Table 2 Recognition effect of different improvement methods

Power equipment	Method 1	Method 2	Ours
Insulator AP	0.85	0.87	0.95
Arrester AP	0.86	0.89	0.98
CT AP	0.89	0.97	0.99
PT AP	0.87	0.96	0.98
Transformer bushing AP	0.89	0.92	0.97
mAP	0.87	0.92	0.97
recognition time /ms	27	7.8	5.6

下载: 导出CSV

表 3 不同识别算法的效果

Table 3 Effects of different recognition algorithms

Methods	mAP	mRC	Recognition time/ms
Ours	0.95	0.97	5.8
SSD	0.85	0.88	32
Faster R-CNN	0.91	0.93	78

下载: 导出CSV

表 4 不同方法的参数量与模型大小

Table 4 Number of parameters and model size of different methods

Methods	mAP@0.5	mAP@0.8	Parameter quantity /M	FLOPs/G	size/MB
Ours	0.97	0.92	0.18	4.2	4.1
YOLO v5s	0.92	0.87	1.32	30.1	29
YOLO v4	0.94	0.83	9.83	323.9	257
Faster R-CNN	0.92	0.81	5.89	145.9	113
SSD	0.86	0.82	0.99	23.8	21.4

下载: 导出CSV

参考文献(20)

[1]	陈俊佑, 金立军, 段绍辉, 等. 基于Hu不变矩的红外图像电力设备识别[J]. 机电工程, 2013, 30(1): 5-8. https://www.cnki.com.cn/Article/CJFDTOTAL-JDGC201301003.htm CHEN Junyou, JIN Lijun, DUAN Shaohui, et al. Power equipment identification in infrared image based on Hu invariant moments[J]. Journal of Mechanical & Electrical Engineering, 2013, 30(1): 5-8. https://www.cnki.com.cn/Article/CJFDTOTAL-JDGC201301003.htm
[2]	邹辉, 黄福珍. 基于改进Fast-Match算法的电力设备红外图像多目标定位[J]. 中国电机工程学报, 2017, 37(2): 591-598. https://www.cnki.com.cn/Article/CJFDTOTAL-ZGDC201702027.htm ZOU Hui, HUANG Fuzhen. Multi-target localization for infrared images of electrical equipment based on improved fast-match algorithm[J]. Proceedings of the CSEE, 2017, 37(2): 591-598. https://www.cnki.com.cn/Article/CJFDTOTAL-ZGDC201702027.htm
[3]	郑含博, 李金恒, 刘洋, 等. 基于改进YOLOv3的电力设备红外目标检测模型[J]. 电工技术学报, 2021, 36(7): 1389-1398. https://www.cnki.com.cn/Article/CJFDTOTAL-DGJS202107009.htm ZHENG Hanbo, LI Jinheng, LIU Yang, et al. Infrared target detection model for power equipment based on improved YOLO v3[J]. Electrotechnical Technology, 2021, 36(7): 1389-1398. https://www.cnki.com.cn/Article/CJFDTOTAL-DGJS202107009.htm
[4]	徐奇伟, 黄宏, 张雪锋, 等. 基于改进区域全卷积网络的高压引线接头红外图像特征分析的在线故障诊断方法[J]. 电工技术学报, 2021, 36(7): 1380-1388. https://www.cnki.com.cn/Article/CJFDTOTAL-DGJS202107008.htm XU Qiwei, HUANG Hong, ZHANG Xuefeng, et al. On-line fault diagnosis method based on infrared image feature analysis of high-voltage lead connector based on improved area full convolutional network[J]. Transactions of the Chinese Society of Electrical Engineering, 2021, 36(7): 1380-1388. https://www.cnki.com.cn/Article/CJFDTOTAL-DGJS202107008.htm
[5]	罗舜. 电力变压器套管将军帽发热故障的红外诊断分析[J]. 变压器, 2018, 55(1): 50-53. https://www.cnki.com.cn/Article/CJFDTOTAL-BYQZ201801018.htm LUO Shun. Infrared diagnosis analysis of power transformer bushing coupler heating[J]. Transformer, 2018, 55(1): 50-53. https://www.cnki.com.cn/Article/CJFDTOTAL-BYQZ201801018.htm
[6]	张杰, 付泉泳, 袁野. 变压器局部放电带电检测技术应用研究[J]. 变压器, 2018, 55(8): 66-71. https://www.cnki.com.cn/Article/CJFDTOTAL-BYQZ201905024.htm ZHANG Jie, FU Quanyong, YUAN Ye. Application research of electric detection technology of partial discharge for transformer[J]. Transformer, 2018, 55(8): 66-71. https://www.cnki.com.cn/Article/CJFDTOTAL-BYQZ201905024.htm
[7]	吕俊, 王福田, 汤进, 等. 基于全景温度场的电力设备在线自动识别与诊断[J]. 计算机与现代化, 2015(8): 19-23. https://www.cnki.com.cn/Article/CJFDTOTAL-JYXH201508004.htm LV Jun, WANG Futian, TANG Jin, et al. Online automatic recognition and diagnosis of electrical devices via thermal panorama[J]. Computer and Modernization, 2015(8): 19-23. https://www.cnki.com.cn/Article/CJFDTOTAL-JYXH201508004.htm
[8]	商俊平, 李储欣, 陈亮. 基于视觉的绝缘子定位与自爆缺陷检测[J]. 电子测量与仪器学报, 2017, 31(6): 844-849. https://www.cnki.com.cn/Article/CJFDTOTAL-DZIY201706007.htm SHANG Junping, LI Chuxin, CHEN Liang. Location and detection for self-explode insulator based on vision[J]. Journal of Electronic Measurement and Instrumentation, 2017, 31(6): 844-849. https://www.cnki.com.cn/Article/CJFDTOTAL-DZIY201706007.htm
[9]	周可慧, 廖志伟, 肖异瑶, 等. 基于改进CNN的电力设备红外图像分类模型构建研究[J]. 红外技术, 2019, 41(11): 1033-1038. https://www.cnki.com.cn/Article/CJFDTOTAL-HWJS201911007.htm ZHOU Kehui, LIAO Zhiwei, XIAO Yiyao, et al. Construction of infrared image classification model for power equipments based on improved CNN[J]. Infrared Technology, 2019, 41(11): 1033-1038. https://www.cnki.com.cn/Article/CJFDTOTAL-HWJS201911007.htm
[10]	赵乐, 王先培, 姚鸿泰, 等. 基于可见光航拍图像的电力线提取算法综述[J]. 电网技术: 2021, 45(4): 1536-1546. https://www.cnki.com.cn/Article/CJFDTOTAL-DWJS202104035.htm ZHAO Le, WANG Xianpei, YAO Hongtai, et al. Summary of power line extraction algorithms based on visible light aerial images[J]. Power System Technology: 2021, 45(4): 1536-1546. https://www.cnki.com.cn/Article/CJFDTOTAL-DWJS202104035.htm
[11]	史晋涛, 李喆, 顾超越, 等. 基于样本扩充的Faster R-CNN电网异物监测技术[J]. 电网技术, 2020, 44(1): 44-51. https://www.cnki.com.cn/Article/CJFDTOTAL-DWJS202001005.htm SHI Jintao, LI Zhe, GU Chaoyue, et al. Faster R-CNN power grid foreign body monitoring technology based on sample expansion[J]. Power System Technology, 2020, 44(1): 44-51. https://www.cnki.com.cn/Article/CJFDTOTAL-DWJS202001005.htm
[12]	乔林, 孙宝华, 徐立波, 等. 多特征联合稀疏表示的电力设备图像识别方法[J]. 自动化技术与应用, 2020, 39(11): 120-123. https://www.cnki.com.cn/Article/CJFDTOTAL-ZDHJ202011026.htm QIAO Lin, SUN Baohua, XU Libo, et al. Power equipment image recognition method based on multi-feature joint sparse representation[J]. Automation Technology and Application, 2020, 39(11): 120-123. https://www.cnki.com.cn/Article/CJFDTOTAL-ZDHJ202011026.htm
[13]	李文璞, 谢可, 廖逍, 等. 基于Faster RCNN变电设备红外图像缺陷识别方法[J]. 南方电网技术, 2019, 13(12): 79-84. https://www.cnki.com.cn/Article/CJFDTOTAL-NFDW202110012.htm LI Wenpu, XIE Ke, LIAO Xiao, et al. Based on faster RCNN substation equipment infrared image defect recognition method[J]. Southern Power Grid Technology, 2019, 13(12): 79-84. https://www.cnki.com.cn/Article/CJFDTOTAL-NFDW202110012.htm
[14]	杨光俊. 卷积神经网络在电力设备红外图像识别中的应用研究[D]. 广州: 华南理工大学, 2019. YANG Guangjun. Research on the Application of Convolutional Neural Network in Infrared Image Recognition of Power Equipment[D]. Guangzhou: South China University of Technology, 2019.
[15]	付强, 姚建刚, 李唐兵, 等. 基于红外图像的绝缘子串钢帽和盘面区域自动提取方法[J]. 红外技术, 2016, 38(11): 969-974. http://hwjs.nvir.cn/cn/article/id/hwjs201611011 FU Qiang, YAO Jiangang, LI Tangbing, et al. Automatic extraction method of steel cap and disk area of insulator string based on infrared image[J]. Infrared Technology, 2016, 38(11): 969-974. http://hwjs.nvir.cn/cn/article/id/hwjs201611011
[16]	马鹏, 樊艳芳. 基于深度迁移学习的小样本智能变电站电力设备部件检测[J]. 电网技术, 2020, 44(3): 1148-1159. https://www.cnki.com.cn/Article/CJFDTOTAL-DWJS202003041.htm MA Peng, FAN Yanfang. Small sample smart substation power equipment component detection based on deep transfer learning[J]. Power System Technology, 2020, 44(3): 1148-1159. https://www.cnki.com.cn/Article/CJFDTOTAL-DWJS202003041.htm
[17]	Tianchi J, LI Qiang, MAOSONG L, et al. Target detection method combining inverted residual block and YOLO v3[J]. Transducer and Microsystem Technologies, 2019, 36(11): 56-61.
[18]	王昕, 赵飞, 蒋佐富, 等. 迁移学习和卷积神经网络电力设备图像识别方法[J]. 中国测试, 2020, 46(5): 108-113. https://www.cnki.com.cn/Article/CJFDTOTAL-SYCS202005018.htm WANG Xin, ZHAO Fei, JIANG Zuofu, et al. Transfer learning and convolutional neural network power equipment image recognition method[J]. China Test, 2020, 46(5): 108-113. https://www.cnki.com.cn/Article/CJFDTOTAL-SYCS202005018.htm
[19]	WU D, LV S, JIANG M, et al. Using channel pruning-based YOLOv4 deep learning algorithm for the real-time and accurate detection of apple flowers in natural environments[J]. Computers and Electronics in Agriculture, 2020, 178(5): 174-178.
[20]	Selvaraju R R, Cogswell M, Das A, et al. Grad-CAM: visual explanations from deep networks via gradient based localization[C]//2017 IEEE International Conference on Computer Vision, 2017: 102-105.