基于改进YOLO v5s算法的光伏组件故障检测

孙建波; 王丽杰; 麻吉辉; 高玮

基于改进YOLO v5s算法的光伏组件故障检测

孙建波^{1, 2,},
王丽杰^{1, 2, ,},
麻吉辉^{1, 2},
高玮^{1, 2}

1.
哈尔滨理工大学测控技术与通信工程学院, 黑龙江哈尔滨 150080
2.
哈尔滨理工大学黑龙江省激光光谱技术及应用重点实验室, 黑龙江哈尔滨 150080

基金项目:

国家自然科学基金项目 61975047

详细信息

作者简介:
孙建波（1995-），男，硕士研究生，主要研究方向为计算机视觉、目标检测。E-mail: 987277295@qq.com

通讯作者:
王丽杰（1971-），女，教授，硕士生导师。主要研究方向为图像分析与识别等。E-mail: wlj@hrbust.edu.cn

中图分类号: TP181
计量
- 文章访问数: 255
- HTML全文浏览量: 55
- PDF下载量: 49
出版历程
- 收稿日期: 2022-05-22
- 修回日期: 2022-06-23
- 刊出日期: 2023-02-19

Photovoltaic Module Fault Detection Based on Improved YOLOv5s Algorithm

SUN Jianbo^{1, 2,},
WANG Lijie^{1, 2, ,},
MA Jihui^{1, 2},
GAO Wei^{1, 2}

1.
School of Measurement and Control Technology and Communication Engineering, Harbin University of Science and Technology, Harbin 150080, China
2.
Heilongjiang Province Key Laboratory of Laser Spectroscopy Technology and Application, Harbin Univesity of Science and Technology, Harbin 150080, China

摘要

摘要: 针对无人机在光伏组件巡检任务中红外故障图像识别准确率低、检测速度慢的问题，提出一种特征增强的YOLO v5s故障检测算法。首先对损失函数进行优化，将原有的回归损失计算方法由GIOU（generalized intersection over union）改为功能更加强大的EIOU（efficient intersection over union）损失函数，并自适应调节置信度损失平衡系数，提升模型训练效果；随后，在每个检测层前分别添加InRe特征增强模块，通过丰富特征表达增强目标特征提取能力。最后，用创建的红外光伏数据集进行对比验证。实验结果表明：本文方法均值平均精度（mean average precision, mAP）为92.76%，检测速度（frame per second，FPS）达到42.37 FPS，其中热斑、组件脱落两种故障类型平均精度分别为94.85%、90.67%，完全能够满足无人机自动巡检的需求。
- 故障检测 /
- YOLO v5s /
- EIOU /
- 特征增强模块
Abstract: Infrared fault images have the limitations of a low recognition accuracy and low detection rate in a PV module inspection task using a UAV. To address these issues, a feature enhanced YOLO v5s fault detection algorithm is proposed. First, the loss function is optimized, the original regression loss calculation method is changed from GIOU to EIOU, and the confidence loss balance coefficient is adjusted adaptively to improve the model training. The InRe feature enhancement module is then added before each detection layer to enhance the ability of the target feature extraction by enriching the feature expression. Finally, comparative experiments are conducted using the infrared photovoltaic dataset created in this study. The experimental results show that the detection mAP of our method is 92.76%, whereas the detection speed is 42.37 FPS. The mean average precisions of the hot spot and component falling off were 94.85% and 90.67%, respectively, which can fully meet the requirements of the automatic inspection of the UAV.
- fault detection /
- YOLOv5s /
- EIOU /
- feature enhancement module

HTML全文

图 1 两种常见光伏组件故障类型样例

Figure 1. Examples of two common PV module failure types

下载: 全尺寸图片幻灯片

图 2 YOLO v5s结构示意图

Figure 2. Structural diagram of YOLO v5s

下载: 全尺寸图片幻灯片

图 3 GIOU示意图

Figure 3. Schematic diagram of GIOU

下载: 全尺寸图片幻灯片

图 4 InRe特征增强模块

Figure 4. InRe feature enhancement module

下载: 全尺寸图片幻灯片

图 5 改进后的YOLO v5s模型

Figure 5. Improved YOLO v5s model

下载: 全尺寸图片幻灯片

图 6 损失变化

Figure 6. Loss diagram

下载: 全尺寸图片幻灯片

图 7 实验六损失函数变化

Figure 7. Loss diagram of experiment 6

下载: 全尺寸图片幻灯片

图 8 检测结果对比

Figure 8. Comparison of test results

下载: 全尺寸图片幻灯片

表 1 消融实验

Table 1 Ablation experiment

Experiment	EIOU	Balance parameter	InRe	mAP%	FPS
Experiment 1	-	-	-	85.57	44.62
Experiment 2	√	-	-	88.67	46.91
Experiment 3	-	√	-	85.97	46.93
Experiment 4	√	√	-	88.54	47.50
Experiment 5	-	-	√	89.81	40.82
Experiment 6	√	√	√	92.76	42.37

下载: 导出CSV

表 2 对比实验

Table 2 Comparative experiment

Algorithm	AP/%			Parameter quantity (M)	mAP/%	FPS
Algorithm	Damage	Hot spot	Model size/MB	Parameter quantity (M)	mAP/%	FPS
Faster-RCNN	92.45	87.95	522.91	137.08	90.20	16.33
SSD	90.95	78.37	90.58	23.75	84.66	36.50
YOLOv3	92.31	88.05	236.32	61.95	90.18	41.31
YOLOv4	91.08	85.76	243.92	63.94	88.42	25.87
YOLOv5s	90.36	78.06	26.96	7.07	84.21	44.56
YOLOv4-mobileNetv2	92.24	87.62	46.11	12.15	89.93	38.26
YOLOv4-tiny	76.19	54.89	22.42	5.88	65.54	56.18
YOLOv5-mobileNet	77.44	60.21	7.42	2.82	68.83	43.58
Proposed	94.85	90.67	30.52	7.41	92.76	42.37

下载: 导出CSV

参考文献(16)

[1]	杨店飞, 郭宇杰, 沈桂鹏. 基于BP神经网络的光伏组件故障类型诊断[J]. 智慧电力, 2016, 44(2): 15-21. https://www.cnki.com.cn/Article/CJFDTOTAL-XBDJ201602006.htm YANG Dianfei, GUO Yujie, SHEN Guipeng. Fault type diagnosis of photovoltaic module based on BP neural network[J]. Smart Power, 2016, 44(2): 15-21. https://www.cnki.com.cn/Article/CJFDTOTAL-XBDJ201602006.htm
[2]	李大勇, 陈如亮, 崔岩, 等. 基于Pspice的光伏组件热斑现象仿真[J]. 哈尔滨工业大学学报, 2006, 38(11): 1888-1892, 1897. https://www.cnki.com.cn/Article/CJFDTOTAL-HEBX200611018.htm LI Dayong, CHEN Ruliang, CUI Yan, et al. A research of hot spot on PV module with Pspice[J]. Journal of Harbin Institute of Technology, 2006, 38(11): 1888-1892, 1897. https://www.cnki.com.cn/Article/CJFDTOTAL-HEBX200611018.htm
[3]	YU Ruei Tzeng, Chia che Ho, Ming Hwa Sheu, et al. Application of thermal camera in PV plants shelter detection[C]//2017 IEEE International Conference on Consumer Electronics, 2017: 411-412.
[4]	时亚涛, 戴芳, 杨畅民. 太阳能光伏电池缺陷检测[J]. 电子测量与仪器学报, 2020, 34(4): 157-164. https://cdmd.cnki.com.cn/Article/CDMD-10561-1019621693.htm SHI Yatao, DAI Fang, YANG Changming. Defect detection of solar photovoltaic cell[J]. Journal of Electronic Measurement and Instrument, 2020, 34(4): 157-164. https://cdmd.cnki.com.cn/Article/CDMD-10561-1019621693.htm
[5]	毛峡, 石天朋. 光伏热斑图像有效区域分割算法研究[J]. 太阳能学报, 2018, 39(5): 1270-1276. https://www.cnki.com.cn/Article/CJFDTOTAL-TYLX201805014.htm MAO Xia, SHI Tianming. Research on effective region segmentation algorithm of photovoltaic hot spot image[J]. Acta Energiae Solaris Sinica, 2018, 39(5): 1270-1276. https://www.cnki.com.cn/Article/CJFDTOTAL-TYLX201805014.htm
[6]	LIU W, Anguelov D, Erhan D, et al. SSD: single shot multibox detector[C]//European Conference on Computer Vision, 2016: 21-37.
[7]	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.
[8]	Girshick R, Donahue J, Darrell T, et al. Rich feature hierarchies for accurate object detection and semantic segmentation[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2014: 580-587.
[9]	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.
[10]	CAI Z, Vasconcelos N. Cascade R-CNN: delving into high quality object detection[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2018: 6154-6162.
[11]	WEI S, LI X, DING S, et al. Hotspots infrared detection of photovoltaic modules based on Hough line transformation and faster-RCNN approach[C]//2019 6th International Conference on Control, Decision and Information Technologies, 2019: 1266-1271.
[12]	Bartler A, Mauch L, Yang B, et al. Automated detection of solar cell defects with deep learning[C]//2018 26th European Signal Processing Conference, 2018: 2035-2039.
[13]	程起泽, 陈泽华, 张雲钦, 等. 基于CNN-LSTM的太阳能光伏组件故障诊断研究[J]. 电子技术应用, 2020, 46(4): 66-70. CHENG Qize, CHEN Zehua, ZHANG Yunqin, et al. Research on fault diagnosis of solar photovoltaic module based on CNN-LSTM[J]. Application of Electronic Technology, 2020, 46(4): 66-70.
[14]	ZHANG Y F, REN W, ZHANG Z, et al. Focal and efficient IOU loss for accurate bounding box regression[EB/OL]. [2021-01-20]. https://doi.org/10.48550/arXiv.2101.08158.
[15]	Cipolla R, Gal Y, Kendall A. Multi-task learning using uncertainty to weigh losses for scene geometry and semantics[C]//IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2018: 7482-7491.
[16]	Christian Szegedy, Sergey Ioffe, Vincent Vanhoucke, et al. Inception-v4, Inception-ResNet and the impact of residual connections on learning[EB/OL]. [2016-08-23]. https://doi.org/10.48550/arXiv.1602.07261.

施引文献(4)

期刊类型引用(2)

1.	肖桐，田昌会，王军，孟真，范琦，高志强，谢晓伟，田晓霞. 一种实现中红外非对称传输的手征超表面设计. 红外技术. 2021(03): 272-278 . 本站查看
2.	孟真，田昌会，黄思宁，范琦，杨百愚，田晓霞. 一种基于六边形环状结构的双阻带红外频率选择表面. 红外技术. 2020(06): 528-533 . 本站查看