改进YOLOv7的电力设备红外图像多目标检测

杨达伟; 杨明圣; 付博

改进YOLOv7的电力设备红外图像多目标检测

1.
广东电网有限责任公司珠海供电局, 广东珠海 519075
2.
广东电网电力科学研究院, 广东清远 511538

详细信息

作者简介:
杨达伟（1985-），男，本科，高级工程师。研究方向为电气工程及其自动化、机器视觉。E-mail：liughua_Swit@163.com

中图分类号: TP391
计量
- 文章访问数: 109
- HTML全文浏览量: 11
- PDF下载量: 45
出版历程
- 收稿日期: 2023-10-29
- 修回日期: 2024-03-06
- 刊出日期: 2025-03-19

Improved YOLOv7 for Multi-Target Detection of Infrared Images of Power Equipment

1.
Zhuhai Power Supply Bureau, Guangdong Power Grid Co., Ltd, Zhuhai 519075, China
2.
Guangdong Power Grid Electric Power Research Institute, Qingyuan 511538, China

摘要

摘要:
针对变电站红外图像中目标数量多、外观相似、背景与目标颜色相近时容易出现漏检误检等问题，提出了一种改进YOLOv7的电力设备红外图像多目标检测方法。首先，为了更好地保留红外图像中的浅层信息，在SPPCSPC模块中引入空洞卷积与均值池化以扩大感受野的同时防止红外小目标淹没在背景中；其次，针对多目标检测中的误检漏检等问题，在头部网络中引入了轻量型SimAM注意力机制以重点关注感兴趣区域。最后，选择一个更适用于小目标检测的NWDloss损失与CIOU损失相结合的混合边框回归损失函数，它可以有效提高红外图像中不同尺度目标检测的准确性。我们在自建电力设备红外图像数据集上与其他7种具有代表性的检测方法进行了对比实验。实验结果表明，改进后的YOLO v7网络模型漏检误检等情况得到明显改善，mAP达到88.9%，相比其他有代表性的目标检测算法在电力设备红外多目标检测上的效果有明显提升。
- 电力设备检测 /
- YOLOv7 /
- 红外图像 /
- 空洞卷积 /
- 注意力机制
Abstract:
To address issues such as a large number of targets, similar appearance, and easy to miss detection and misdetection when the background and target are of similar color in substation infrared images, a multi-target detection method is proposed for electric power equipment by improving YOLOv7. First, to better retain the shallow information in the infrared image, cavity convolution and mean pooling were introduced into a spatial pyramid pooling cross-stage partial convolution (SPPCSPC) module, to expand the receptive field while preventing small infrared targets from being submerged in the background. Second, to deal with misdetection and detection omission in multi-target detection, a lightweight simple attention module (SimAM) was introduced into the head network to focus on the region of interest. Finally, a hybrid edge regression loss function suitable for small-target detection, combining the normalized Gaussian Wasserstein distance (NWD) and complete intersection over union (CIOU) losses, was chosen to effectively improve the accuracy of target detection at different scales in infrared images. We conducted comparison experiments with seven other representative detection methods using a self-constructed infrared image dataset of power equipment. The experimental results showed that the improved YOLOv7 network model significantly improves leakage detection and reduced false detection. Its mean average precision (mAP) reached 88.9%, which is a significant improvement compared with those of other representative target detection algorithms for infrared multi-target detection of power equipment.
- power equipment detection /
- YOLOv7 /
- infrared images /
- dilation convolution /
- attention mechanism

HTML全文

图 1 YOLO v7网络结构

Figure 1. YOLO v7 network structure

下载: 全尺寸图片幻灯片

图 2 SPPCSPC-DAVG模块

Figure 2. SPPCSPC-DAVG module

下载: 全尺寸图片幻灯片

图 3 SimAM三维注意力机制

Figure 3. SimAM 3D attention mechanism

下载: 全尺寸图片幻灯片

图 4 MPlayer-SAM模块

Figure 4. MPlayer-SAM module

下载: 全尺寸图片幻灯片

图 5 数据标注示例

Figure 5. Example of data labeling

下载: 全尺寸图片幻灯片

图 6 损失函数迭代对比

Figure 6. Loss function iteration contrast

下载: 全尺寸图片幻灯片

图 7 P-R曲线对比

Figure 7. P-R curve comparison

下载: 全尺寸图片幻灯片

图 8 检测实验结果对比

Figure 8. Comparison of detection experimental results

下载: 全尺寸图片幻灯片

表 1 消融实验结果

Table 1 Results of ablation experiments

Methods	Precision/%	Recall/%	mAP@ 0.5	mAP@ 0.5-0.95	FPS
YOLO v7	82.5	90.4	0.858	0.544	24.96
YOLO v7+SPPCSPC-DAVG	84.4	91.0	0.874	0.549	24.52
YOLOv7+SPPCSPC-DAVG+MPlayer-SAM	85.7	94.2	0.880	0.552	22.79
YOLOv7+SPPCSPC-DAVG+MPlayer-SAM+NWDloss	86.5	93.6	0.889	0.558	22.27
Note: The bold fonts represent the best in each category.

下载: 导出CSV

表 2 不同算法的指标对比

Table 2 Comparison of results of different algorithms %

Method	mAP	Reactor	Joint_1	Measure	Knife	OCT	Bushing	Resister	LA
Faster-RCNN	72.5	73.2	48.6	72.4	81.4	88.5	52.1	83.4	80.2
Cascade-RCNN	85.2	86.9	59.1	92.6	89.3	97.8	64.4	97.6	94.2
RetinaNet	83.8	93.7	57.5	93.2	85.4	95.6	56.0	94.1	94.8
SSD	78.8	84.9	55.4	87.1	79.2	92.6	51.3	89.5	90.7
YOLOX-S	83.1	96.0	58.7	91.3	89.1	90.3	45.9	97.9	96.1
YOLOv5-S	79.8	86.1	57.9	83.7	85.6	93.3	55.6	87.2	89.4
YOLOv7	85.8	94.1	58.8	93.7	91.9	94.2	60.2	96.6	97.5
Ours	88.9	97.1	64.9	92.6	89.5	98.5	73.5	97.1	98.2
Note: The bold fonts represent the best in each category.

下载: 导出CSV

参考文献(26)

[1]	刘喜梅, 马俊杰. 泛在电力物联网在电力设备状态监测中的应用[J]. 电力系统保护与控制, 2020, 48(14): 69-75. LIU X M, MA J J. Application of the ubiquitous power Internet of Things in state monitoring of power equipment[J]. Power System Protection and Control, 2020, 48(14): 69-75.
[2]	赵洪山, 张则言, 孟航, 等. 基于高压绝缘套管纹理特征的红外目标检测[J]. 红外技术, 2021, 43(3): 258-265. http://hwjs.nvir.cn/article/id/6faf1628-98fa-4a3a-ade2-7c8d0abf7dae ZHAO H S, ZHANG Z Y, MENG H, et al. Infrared target detection of high voltage insulation bushing based on textural features[J]. Infrared Technology, 2021, 43(3): 258-265. http://hwjs.nvir.cn/article/id/6faf1628-98fa-4a3a-ade2-7c8d0abf7dae
[3]	郑婷婷, 周浩, 王秋忆. 基于改进YOLOv5的电力设备检测算法[J]. 电子测量技术, 2023, 46(4): 155-160. ZHENG T T, ZHOU H, WANG Q Y. Power equipment detection algorithm based on improved YOLOv5[J]. Electronic Measurement Technology, 2023, 46(4): 155-160.
[4]	任新辉. 基于红外技术的变电站设备识别与热故障诊断[D]. 成都: 西南交通大学, 2016. REN X H. Substation Equipment Recognition and Thermal Fault Diagnosis Based on Infrared Technology[D]. Chengdu: Southwest Jiaotong University, 2016.
[5]	尹阳. 基于红外图像的变电站设备识别与热状态监测系统研究[D]. 西安: 西安科技大学, 2018. YIN Y. Research on Substation Equipment Recognition and Thermal Status Monitoring System Based on Infrared Images[D]. Xi'an: Xi'an University of Science and Technology, 2018.
[6]	Girshick R. Fast R-CNN[C]//2015 IEEE International Conference on Computer Vision(ICCV), IEEE, 2015: 1440-1448.
[7]	REN S, HE K, Girshick R, et al. Faster R-CNN: towards real-time object detection with region proposal networks[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2017, 39(6): 1137-1149.
[8]	Redmon J, Divvala S, Girshick R, et al. You Only Look Once: Unified Real-Time Object Detection [C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2016: 779-788.
[9]	Redmon J, Farhadi A. YOLO9000: Better, faster, stronger[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2017: 7263-7271.
[10]	Redmon J, Farhadi A. YOLOv3: An incremental improvement[J/OL]. arXiv: 1804.02767, 2018, https://arxiv.org/abs/1804.02767.
[11]	Bochkovskiy A, WANG C Y, LIAO H Y M. YOLOv4: optimal speed and accuracy of object detection[J/OL]. arXiv: 2004.10934, 2020, https://arxiv.org/abs/2004.10934.
[12]	LIU W, Anguelov D, Erhan D, et al. SSD: single shot multibox detector[C]//European Conference on Computer Vision, 2016: 21-37.
[13]	LIN T Y, Goyal P, Girshick R, et al. Focal loss for dense object detection[C]//Proceedings of the IEEE International Conference on Computer Vision, 2017: 2980-2988.
[14]	汤踊, 韩军, 魏文力, 等. 深度学习在输电线路中部件识别与缺陷检测的研究[J]. 电子测量技术, 2018, 41(6): 60-65. TANG Y, HAN J WEI W L, et al. Research on deep learning for component recognition and defect detection in transmission lines[J]. Electronic Measurement Technology, 2018, 41(6): 60-65.
[15]	LIU Z Q, WANG H F. Automatic detection of transformer components in inspection images based on improved Faster R-CNN[J]. Energies, 2018, 11(12): 3496. DOI: 10.3390/en11123496
[16]	徐凯, 梁志坚, 张镱议, 等. 基于GoogLeNet InceptionV3模型的电力设备图像识别[J]. 高压电器, 2020, 56(9): 129-135, 143. XU K, LIANG Z J, ZHANG Y Y, et al. Power equipment image recognition based on GoogLeNet InceptionV3 model[J]. High Voltage Apparatus, 2020, 56(9): 129-135, 143.
[17]	王旭红, 李浩, 樊绍胜, 等. 基于改进SSD的电力设备红外图像异常自动检测方法[J]. 电工技术学报, 2020, 35(S1): 302-310. WANG X H, LI H, FAN S S et al. Automatic detection method for abnormal infrared images of power equipment based on improved SSD[J]. Transactions of the Chinese Electrotechnical Society, 2020, 35(S1): 302-310.
[18]	Howard A G, ZHU M, CHEN B, et al. Mobilenets: Efficient convolutional neural networks for mobile vision applications[J/OL]. arXiv: 1704.04861, 2017, https://arxiv.org/abs/1704.04861.
[19]	ZHENG Hanbo, LI Jinheng, LIU Yang, et al. Infrared target detection model for power equipment based on improved YOLOv3[J]. Transactions of the Chinese Electrotechnical Society, 2021, 36(7): 1389-1398.
[20]	苏海锋, 赵岩, 武泽君, 等. 基于改进RetinaNet的电力设备红外目标精细化检测模型[J]. 红外技术, 2021, 43(11): 1104-1111. http://hwjs.nvir.cn/article/id/3233a6a1-cbf0-4110-baa5-2a56e551f092 SU H F, ZHAO Y, WU Z J, et al. Refined infrared target detection model for power equipment based on improved RetinaNet[J]. Infrared Technology, 2021, 43(11): 1104-1111. http://hwjs.nvir.cn/article/id/3233a6a1-cbf0-4110-baa5-2a56e551f092
[21]	段中兴, 张雨明, 马佳豪. 基于改进YOLOv4的电力设备红外图像识别[J]. 激光与光电子学进展, 2022, 59(24): 29-36. DUAN Z X, ZHANG Y M, MA J H. Infrared image recognition of power equipment based on improved YOLOv4 [J]. Advances in Laser and Photonics, 2022, 59(24): 29-36.
[22]	YANG L, ZHANG R Y, LI L, et al. Simam: A simple, parameter-free attention module for convolutional neural networks[C]//International Conference on Machine Learning, PMLR, 2021: 11863-11874.
[23]	WANG J, XU C, YANG W, et al. A normalized Gaussian Wasserstein distance for tiny object detection[J]. arXiv preprint arXiv: 2110.13389, 2021.
[24]	HU J, SHEN L, SUN G. Squeeze-and-excitation networks[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2018: 7132-7141.
[25]	WANG Q L, WU B G, ZHU P F, et al, ECA-Net: efficient channel attention for deep convolutional neural networks[C]//2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2020: 11531-11539.
[26]	WOO S, PARK J, LEE J Y, et al. CBAM: Convolutional Block Attention Module[C]//Proceedings of the European Conference on Computer Vision (ECCV), 2018: 3-19.