基于改进YOLOv5的变电站典型缺陷自动检测算法

doi:10.16182/j.issn1004731x.joss.23-0914

摘要/Abstract

摘要：

针对变电站缺陷识别场景下存在的复杂变电缺陷和样本不均衡等问题，提出一种改进的YOLOv5算法。向YOLOv5网络结构中引入Transformer模型，使用自注意力机制捕捉特征之间的长距离依赖关系。引入基于焦点损失的对损失函数进行优化，提高对小样本变电站缺陷的检测准确性和鲁棒性。为适应变电站缺陷识别任务，构建专门的数据集，采用聚类算法对真实标注框进行聚类，生成更准确的先验框。通过遗传算法选择适应该数据集的超参数，提升算法性能。实验结果表明，本文所提算法在变电站缺陷识别任务中取得了较好的性能表现。相较于传统的YOLOv5算法，改进算法在复杂变电缺陷和小样本目标的识别上具有更强的能力。

关键词: 目标检测, YOLOv5, Transformer, 自注意力机制, 变电站缺陷识别

Abstract:

In response to the challenges present in the context of defect recognition in substations, such as complex substation defects and sample imbalance, an improved YOLOv5 algorithm was proposed. The Transformer model was introduced into the YOLOv5 network structure, leveraging the self-attention mechanism to capture long-range dependencies among features. A focal loss-based optimization was employed to improve the loss function, as well as the detection accuracy and robustness of defects of small sample substations. To meet the requirements of substation defect recognition, a dedicated dataset was constructed. A clustering algorithm was applied to the real annotation boxes to generate more accurate prior boxes. The genetic algorithm was utilized to select hyperparameters that are specifically tailored to the dataset, further enhancing the algorithm's performance. Experimental results demonstrate that the proposed algorithm achieves favorable performance in substation defect recognition tasks. In comparison to the traditional YOLOv5 algorithm, the proposed algorithm exhibits superior capabilities in recognizing complex substation defects and small sample targets.

Key words: target detection, YOLOv5, Transformer, self-attention mechanism, substation defect recognition

中图分类号:

TP391

徐忠锴,刘艳玲,盛晓娟等 . 基于改进YOLOv5的变电站典型缺陷自动检测算法[J]. 系统仿真学报, 2024, 36(11): 2604-2615.

Xu Zhongkai,Liu Yanling,Sheng Xiaojuan,et al . Automatic Detection Algorithm for Typical Defects of Substation Based on Improved YOLOv5[J]. Journal of System Simulation, 2024, 36(11): 2604-2615.

图/表 8

图1

图2

表1

表2

表3

表4

表5

图3

参考文献 33

1	李锋, 谢俊, 兰金波, 等. 智能变电站继电保护配置的展望和探讨[J]. 电力自动化设备, 2012, 32(2): 122-126.
	Li Feng, Xie Jun, Lan Jinbo, et al. Prospect and Discussion of Relay System Configuration for Intelligent Substation[J]. Electric Power Automation Equipment, 2012, 32(2): 122-126.
2	司文荣, 傅晨钊, 卞晓亮, 等. 变电站接地网缺陷诊断无损评价方法综述[J]. 电气工程学报, 2015, 10(8): 14-21.
	Si Wenrong, Fu Chenzhao, Bian Xiaoliang, et al. Overview of the Nondestructive Evaluation Methods for Substation Grounding Grids Fault Diagnosis[J]. Journal of Electrical Engineering, 2015, 10(8): 14-21.
3	杨凯, 李锐, 罗林, 等. 基于深度学习的车轮踏面表面缺陷检测研究[J]. 信息技术, 2021, 45(7): 93-97.
	Yang Kai, Li Rui, Luo Lin, et al. Research on Wheel Tread Surface Defect Detection Based on Deep Learning[J]. Information Technology, 2021, 45(7): 93-97.
4	王慧玲, 綦小龙, 武港山. 基于深度卷积神经网络的目标检测技术的研究进展[J]. 计算机科学, 2018, 45(9): 11-19.
	Wang Huiling, Qi Xiaolong, Wu Gangshan. Research Progress of Object Detection Technology Based on Convolutional Neural Network in Deep Learning[J]. Computer Science, 2018, 45(9): 11-19.
5	李琳. 低光照图像目标检测关键技术研究[D]. 合肥: 合肥工业大学, 2021.
	Li Lin. Research on Key Technology of Low Illumination Image Object Detection[D]. Hefei: Hefei University of Technology, 2021.
6	Syed Sahil Abbas Zaidi, Ansari M S, Aslam Asra, et al. A Survey of Modern Deep Learning Based Object Detection Models[J]. Digital Signal Processing, 2022, 126: 103514.
7	Redmon J, Divvala S, Girshick R, et al. You Only Look Once: Unified, Real-time Object Detection[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2016: 779-788.
8	Redmon J, Farhadi A. YOLO9000: Better, Faster, Stronger[C]//2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2017: 6517-6525.
9	Redmon J, Farhadi A. YOLOv3: An Incremental Improvement[EB/OL]. (2018-04-08) [2023-01-08]. .
10	Bochkovskiy A, Wang C Y, Liao Hongyuan. YOLOv4: Optimal Speed and Accuracy of Object Detection[EB/OL]. (2020-04-23) [2023-01-15]. .
11	褚文杰. 基于YOLOv5的坦克装甲车辆目标检测关键技术的研究[D]. 北京: 北京交通大学, 2021.
	Chu Wenjie. Research on the Key Technology of Tank and Armored Vehicle Target Detection Based on YOLOv5[D]. Beijing: Beijing Jiaotong University, 2021.
12	Ultralytics. Ultralytics Yolov5[EB/OL]. [2022-06-01]. .
13	赵文清, 康怿瑾, 赵振兵, 等. 改进YOLOv5s的遥感图像目标检测[J]. 智能系统学报, 2023, 18(1): 86-95.
	Zhao Wenqing, Kang Yijin, Zhao Zhenbing, et al. A Remote Sensing Image Object Detection Algorithm with Improved YOLOv5s[J]. CAAI Transactions on Intelligent Systems, 2023, 18(1): 86-95.
14	张云佐, 郭威, 蔡昭权, 等. 联合多尺度与注意力机制的遥感图像目标检测[J]. 浙江大学学报(工学版), 2022, 56(11): 2215-2223.
	Zhang Yunzuo, Guo Wei, Cai Zhaoquan, et al. Remote Sensing Image Target Detection Combining Multi-scale and Attention Mechanism[J]. Journal of Zhejiang University(Engineering Science), 2022, 56(11): 2215-2223.
15	张锦, 屈佩琪, 孙程, 等. 基于改进YOLOv5的安全帽佩戴检测算法[J]. 计算机应用, 2022, 42(4): 1292-1300.
	Zhang Jin, Qu Peiqi, Sun Cheng, et al. Safety Helmet Wearing Detection Algorithm Based on Improved YOLOv5[J]. Journal of Computer Applications, 2022, 42(4): 1292-1300.
16	李文举, 张干, 崔柳, 等. 基于坐标注意力的轻量级交通标志识别模型[J]. 计算机应用, 2023, 43(2): 608-614.
	Li Wenju, Zhang Gan, Cui Liu, et al. Lightweight Traffic Sign Recognition Model Based on Coordinate Attention[J]. Journal of Computer Applications, 2023, 43(2): 608-614.
17	刘晓倩, 李士心, 李保胜, 等. 基于改进YOLOv5的交通标志检测算法研究[J]. 计算机科学与应用, 2022, 12(9): 2161-2168.
	Liu Xiaoqian, Li Shixin, Li Baosheng, et al. Research on Traffic Sign Detection Algorithm Based on Improved YOLOv5[J]. Computer Science and Application, 2022, 12(9): 2161-2168.
18	程松, 杨洪刚, 徐学谦, 等. 基于YOLOv5的改进轻量型X射线铝合金焊缝缺陷检测算法[J]. 中国激光, 2022, 49(21): 130-138.
	Cheng Song, Yang Honggang, Xu Xueqian, et al. Improved Lightweight X-Ray Aluminum Alloy Weld Defects Detection Algorithm Based on YOLOv5[J]. Chinese Journal of Lasers, 2022, 49(21): 130-138.
19	张宇杰, 蔡乐才, 成奎, 等. 改进YOLOv5金属表面缺陷检测方法[J]. 四川轻化工大学学报(自然科学版), 2022, 35(4): 32-41.
	Zhang Yujie, Cai Lecai, Cheng Kui, et al. An lmproved YOLOv5 Method for the Detection of Metal Surface Defects[J]. Journal of Sichuan University of Science & Engineering(Natural Science Edition), 2022, 35(4): 32-41.
20	陈婷, 周旻, 韩勤, 等. 基于改进YOLOv4的变电站缺陷检测[J]. 计算机系统应用, 2022, 31(6): 245-251.
	Chen Ting, Zhou Min, Han Qin, et al. Defect Detection for Substation Based on Improved YOLOv4[J]. Computer Systems & Applications, 2022, 31(6): 245-251.
21	罗箫瑜, 张志. 基于改进YOLOX的变电站设备缺陷检测方法[J]. 吉林大学学报(信息科学版), 2023, 41(5): 848-857.
	Luo Xiaoyu, Zhang Zhi. Defect Detection for Substation Based on Improved YOLOX[J]. Journal of Jilin University(Information Science Edition), 2023, 41(5): 848-857.
22	伍艺佳, 华雄, 王丽蓉, 等. 基于注意力机制学习的变电设备缺陷检测方法[J]. 计算机与现代化, 2021(2): 7-12, 17.
	Wu Yijia, Hua Xiong, Wang Lirong, et al. Method of Substation Equipment Defect Detection Based on Attention Mechanism Learning[J]. Computer and Modernization, 2021(2): 7-12, 17.
23	谢政峰. 基于深度学习的新材料地板缺陷检测技术研究[D]. 成都: 四川大学, 2021.
	Xie Zhengfeng. Research on Defect Detection Technology of New Material Flooring Based on Deep Leaning[D]. Chengdu: Sichuan University, 2021.
24	谷玉海, 曹梦婷, 修嘉芸, 等. 基于YOLOv4网络的违章行为检测算法[J]. 重庆理工大学学报(自然科学), 2021, 35(8): 114-121.
	Gu Yuhai, Cao Mengting, Xiu Jiayun, et al. Algorithm for Detecting Violations Based on YOLOv4 Network[J]. Journal of Chongqing University of Technology(Natural Science), 2021, 35(8): 114-121.
25	Vaswani A, Shazeer N, Parmar N, et al. Attention is All You Need[C]//Proceedings of the 31st International Conference on Neural Information Processing Systems. Red Hook: Curran Associates Inc., 2017: 6000-6010.
26	Devlin J, Chang Mingwei, Lee K, et al. BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding[C]//Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. Stroudsburg: ACL, 2019: 4171-4186.
27	Radford A, Narasimhan K, Salimans T, et al. Improving Language Understanding by Generative Pre-training[EB/OL]. [2023-02-01]. .
28	Peters M E, Neumann M, Iyyer M, et al. Deep Contextualized Word Representations[C]//Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. Stroudsburg: ACL, 2018: 2227-2237.
29	侯丽微, 胡珀, 曹雯琳. 主题关键词信息融合的中文生成式自动摘要研究[J]. 自动化学报, 2019, 45(3): 530-539.
	Hou Liwei, Hu Po, Cao Wenlin. Automatic Chinese Abstractive Summarization with Topical Keywords Fusion[J]. Acta Automatica Sinica, 2019, 45(3): 530-539.
30	刘文婷, 卢新明. 基于计算机视觉的Transformer研究进展[J]. 计算机工程与应用, 2022, 58(6): 1-16.
	Liu Wenting, Lu Xinming. Research Progress of Transformer Based on Computer Vision[J]. Computer Engineering and Applications, 2022, 58(6): 1-16.
31	Dosovitskiy A, Beyer L, Kolesnikov A, et al. An Image is Worth 16×16 Words:Transformers for Image Recognition at Scale[C]//ICLR 2021. New York: ICIR, 2021: 1-22.
32	Lin T Y, Goyal P, Girshick R, et al. Focal Loss for Dense Object Detection[C]//2017 IEEE International Conference on Computer Vision (ICCV). Piscataway: IEEE, 2017: 2999-3007.
33	Lin T Y, Maire M, Belongie S, et al. Microsoft COCO: Common Objects in Context[C]//Computer Vision-ECCV 2014. Cham: Springer International Publishing, 2014: 740-755.

类别	标签	样本数量	占比%
表盘模糊	bj_bpmh	2 442	11.76
表盘破损	bj_bpps	953	4.59
地面油污	sly_dmyw	2 558	12.32
硅胶变色	hxq_gjbs	4 637	22.33
硅胶筒破损	hxq_gjtps	266	1.28
绝缘子破裂	jyz_pl	1 460	7.03
箱门闭合异常	xmbhyc	1 975	9.51
挂空悬浮物	yw_gkxfw	2 967	14.29
鸟巢	yw_nc	2 985	14.37
未穿安全帽	wcaqm	524	2.52

模型	参数规模/M	计算量/B	推理时间/ms
本文方法	7.0	15.8	1.5
YOLOv5s	7.2	16.5	1.7
Faster R-CNN	273.0	280.6	46
SSD	78.4	111.7	37

模型	精确率	召回率	mAP@0.5
本文方法	0.782	0.781	0.715
YOLOv5s	0.761	0.718	0.704
Faster R-CNN	0.682	0.734	0.698
SSD	0.744	0.636	0.686

类别	基线模型		本文方法
类别	精确率	召回率	精确率	召回率
整体效果	0.761	0.718	0.782	0.781
表盘模糊	0.843	0.825	0.839	0.828
表盘破损	0.790	0.712	0.864	0.709
地面油污	0.547	0.406	0.645	0.648
硅胶变色	0.852	0.879	0.813	0.894
硅胶筒破损	0.794	0.481	0.907	0.863
绝缘子破裂	0.716	0.520	0.749	0.752
箱门闭合异常	0.841	0.699	0.746	0.734
挂空悬浮物	0.846	0.736	0.806	0.779
鸟巢	0.813	0.742	0.761	0.734
未穿安全帽	0.901	0.861	0.913	0.826

方法	Transformer	Focal Loss	超参数进化	精确率	召回率	mAP@0.5	mAP@0.5:0.95
基线模型				0.761	0.718	0.704	0.420
改进1	√			0.813	0.692	0.711	0.428
改进2	√		√	0.769	0.729	0.707	0.418
本文	√	√	√	0.782	0.781	0.715	0.442