多晶硅太阳电池表面缺陷多源异类、小尺度缺陷检测困难。现有基于机器学习图像处理的表面缺陷检测算法在实际使用过程中存在检测精度不高、检测类型少、鲁棒性差等问题。基于深度学习,提出一种可提升精度、便于工业应用的多尺度特征融合的轻量化缺陷检测模型。该模型以YOLOv5目标检测算法为框架,为增强多尺度特征的融合效果,设计浅层通道融合分支和双采样结构,改进了多尺度池化金字塔,提升了模型对小尺度缺陷的检测能力;引入ECA注意力机制,增强颈部网络对有效信息的关注度;最后,设计新的降采样结构和C3（Light-CSP-Darknet53）结构,提升检测精度的同时降低参数量和计算量。实验表明,改进算法的全类平均精度（mAP0.5）比基线模型提高8.5%,参数量减少44.3%,计算复杂度降低33.7%。
Abstract
Detection difficulty caused by multi-source and miscellaneous defects and micro-size defects on the surface of polycrystalline silicon solar cells has bothered photovoltaic industry for a long time. Surface defect detection algorithm driven by machine learning is struggling in the troubles of low-accuracy, less-feature and bad robustness during industrial running. Based on deep learning, a novel lightweight defect detection model with multi-scale feature fusion is put forward to narrow the error gap. Framed by YOLOv5 （the object detection algorithm）, for the goal of enhancing the multi-scale feature fusion effect, shallow channel fusion branch and dual sampling structure are employed, as well the multi-scale pooling pyramid is modified, to improve the detection accuracy of micro-size defects. In this framework, efficient channel attention mechanism is introduced to enhance the neck network’s attention, a new downsampling structure and C3 （Light-CSP-Darknet53） structure are added to increase the accuracy and decrease computation. Experimental results demonstrate that this modified model can increase average accuracy （mAP0.5） by 8.5%, reduce the number of parameters by 44.3% and decrease computation complexity by 33.7%.
关键词
太阳电池 /
表面缺陷 /
目标检测 /
注意力机制 /
多尺度特征融合
Key words
solar cells /
surface defect /
object detection /
attention mechanism /
multi-scale feature fusion
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
[1] HAO D N, QI L F, TAIRAB A M, et al.Solar energy harvesting technologies for PV self-powered applications: a comprehensive review[J]. Renewable energy, 2022, 188: 678-697.
[2] 马新尖, 林涛. 单晶硅太阳电池电致发光缺陷及工艺影响因素分析[J]. 激光与光电子学进展, 2013, 50(3): 031601.
MA X J, LIN T.Analysis of mono-crystalline silicon solar cells electroluminescence defects and process influencing factors[J]. Laser & optoelectronics progress, 2013, 50(3): 031601.
[3] LI Y K, HE C F, LYU Y, et al.Crack detection in monocrystalline silicon solar cells using air-coupled ultrasonic lamb waves[J]. NDT & E international, 2019, 102: 129-136.
[4] KARIMI M, SAMET H, GHANBARI T, et al.A current based approach for hotspot detection in photovoltaic strings[J]. International transactions on electrical energy systems, 2020, 30(9): e12517.
[5] BU C W, LIU T, LI R, et al.Electrical Pulsed Infrared Thermography and supervised learning for PV cells defects detection[J]. Solar energy materials and solar cells, 2022, 237: 111561.
[6] 苏虎, 张家斌, 张博豪, 等. 基于视觉感知的表面缺陷检测综述[J]. 计算机集成制造系统, 2023, 29(1): 169-191.
SU H, ZHANG J B, ZHANG B H, et al.Review of surface defect inspection based on visual perception[J]. Computer integrated manufacturing systems, 2023, 29(1): 169-191.
[7] 任惠, 夏静, 卢锦玲, 等. 基于红外图像和改进MobileNet-V3的光伏组件故障诊断方法[J]. 太阳能学报, 2023, 44(8): 238-245.
REN H, XIA J, LU J L, et al.Research on fault diagnosis of photovoltaic modules based on infrared images and improved MobileNet-V3[J]. Acta energiae solaris sinica, 2023, 44(8): 238-245.
[8] 王道累, 李明山, 姚勇, 等. 改进SSD的光伏组件热斑缺陷检测方法[J]. 太阳能学报, 2023, 44(4): 420-425.
WANG D L, LI M S, YAO Y, et al.Method of hotspot detection of photovoltaic panels modules on improved SSD[J]. Acta energiae solaris sinica, 2023, 44(4): 420-425.
[9] LIU W, ANGUELOV D, ERHAN D, et al.SSD: single shot multibox detector[C]//Computer Vision-ECCV 2016. Cham: Springer International Publishing, 2016: 21-37.
[10] 鲁东林, 王淑青, 鲁濠, 等. 一种改进Faster R-CNN的太阳能电池片缺陷检测方法[J]. 激光杂志, 2022, 43(3): 50-55.
LU D L, WANG S Q, LU H, et al.A defect detection method of solar cell based on improved Faster R-CNN[J]. Laser journal, 2022, 43(3): 50-55.
[11] 周颖, 王如意, 袁梓桐, 等. 一种高效双路径注意力太阳电池缺陷检测网络[J]. 太阳能学报, 2023, 44(4): 407-413.
ZHOU Y, WANG R Y, YUAN Z T, et al.An efficient dual-path attention solar cell defect detection network[J]. Acta energiae solaris sinica, 2023, 44(4): 407-413.
[12] 王淑青, 张鹏飞, 要若天, 等. 基于改进YOLOv5的太阳能电池片表面缺陷检测[J]. 仪表技术与传感器, 2022(5): 111-116.
WANG S Q, ZHANG P F, YAO R T, et al.Surface defect detection of solar cell based on improved YOLOv5[J]. Instrument technique and sensor, 2022(5): 111-116.
[13] ALIMAZEN F M, SEOUD R A A, SHAKER Y O. Deep learning for automatic defect detection in PV modules using electroluminescence images[J]. IEEE access, 2023, 11: 57783-57795.
[14] 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). Seattle, WA, USA, 2020: 11531-11539.
[15] GHOLAMI A, KWON K, WU B C, et al.SqueezeNext: hardware-aware neural network design[C]//2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW). Salt Lake City, UT, USA, 2018: 1719-1728.
[16] HE K M, ZHANG X Y, REN S Q, et al.Deep residual learning for image recognition[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Las Vegas, NV, USA, 2016: 770-778.
[17] 杨佳博. 太阳能电池片复杂表面图像多光谱融合[D]. 天津: 河北工业大学, 2019.
YANG J B.Multispectral fusion of complex surface images of solar cells[D]. Tianjin: Hebei University of Technology, 2019.
[18] REN S Q, HE K M, GIRSHICK R, et al.Faster R-CNN: towards real-time object detection with region proposal networks[J]. IEEE transactions on pattern analysis and machine intelligence, 2017, 39(6): 1137-1149.
[19] BOCHKOVSKIY A, WANG C Y, LIAO H M. YOLOv4: optimal speed and accuracy of object detection[EB/OL].2020: 2004.10934. https://arxiv.org/abs/2004.10934v1.
[20] WANG C Y, BOCHKOVSKIY A, LIAO H Y M. YOLOv7: trainable bag-of-freebies sets new state-of-the-art for real-time object detectors[EB/OL].2022: arXiv: 2207.02696. https://arxiv.org/abs/2207.02696v1.
PDF(2688 KB)
