基于多输入辅助分支训练的太阳能网版表面缺陷检测

doi:10.3778/j.issn.1002-8331.2405-0068

摘要/Abstract

摘要： 针对太阳能网版表面缺陷检测存在目标小、类别数量不均衡和类别目标长宽比失衡问题，提出一种改进的GELAN-C的缺陷检测模型。增加多输入辅助分支和自主学习隐藏相关性融合模块，降低梯度传递深度，改善小目标检测能力。引入DBB重参数化模块构成多分支解耦头，增强末端特征图表达能力，提升类别数量少的样本检测精度。通过ShapeIoU损失函数提高长宽比失衡目标的拟合度。实验结果表明，改进的GELAN-C大小两种模型mAP50和mAP50-95分别为79.0%、50.4%、78.5%、50.2%，最终分别提升4.2、4.5、4.4、6.2个百分点，而参数量和计算量没有增加，实现了太阳能网版表面缺陷的实时高精度检测。

关键词: GELAN-C, 辅助训练, 重参数化, 隐式相关性, ShapeIoU

Abstract: Aiming at the problems of small targets, unbalanced number of categories and imbalance of category target aspect ratio in the detection of defects on the surface of solar stencil, an improved defect detection model of GELAN-C is proposed. The multi-input auxiliary branching and autonomous learning implicit correlation fusion module are added to reduce the gradient transfer depth and improve the small target detection capability. The DBB reparameterization module is introduced to constitute a multi-branch decoupling head, which enhances the end feature map expression ability and improves the detection accuracy of samples with a small number of categories, and also improves the fit of aspect ratio imbalance target by ShapeIoU loss function. The experimental results show that the mAP50 and mAP50-95 of the two improved GELAN-C size models are 79.0%, 50.4%, 78.5%, 50.2%, respectively, and ultimately improved by 4.2, 4.5, 4.4, and 6.2 percentage points, while the number of parameter counts and computation are not increased, realizing the real-time high-precision detection of defects on the surface of solar stencils.

Key words: GELAN-C, auxiliary training, , reparameterization, implicit correlation, ShapeIoU

吉训生, 马佩珏. 基于多输入辅助分支训练的太阳能网版表面缺陷检测[J]. 计算机工程与应用, 2025, 61(15): 258-268.

JI Xunsheng, MA Peijue. Solar Cell Surface Defect Detection Based on Multi-Input Auxiliary Branch Training[J]. Computer Engineering and Applications, 2025, 61(15): 258-268.

参考文献

[1] CAO M Q, MA L, LI R Q. Automated visual inspection of solar cell images using adapted morphological and edge dete-ction algorithms[J]. Journal of Optics, 2024, 53(2): 1293-1301.
[2] 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.
[3] 伊力哈木·亚尔买买提, 邓皓, 谢丽蓉. 基于改进YOLOv4的太阳能电池板缺陷检测[J]. 华南师范大学学报(自然科学版), 2023, 55(5): 21-30.
YILIHAMU Y, DENG H, XIE L R. Solar panel defect detection based on improved YOLOv4[J]. Journal of South China Normal University (Natural Science Edition), 2023, 55(5): 21-30.
[4] 彭雪玲, 林珊玲, 林志贤, 等. 改进的YOLOv5s太阳能电池片缺陷检测算法[J]. 液晶与显示, 2024, 39(2): 237-247.
PENG X L, LIN S L, LIN Z X, et al. Defect detection algorithm of improved YOLOv5s solar cell[J]. Chinese Journal of Liquid Crystals and Displays, 2024, 39(2): 237-247.
[5] 王淑青, 朱文鑫, 张子言, 等. 基于改进YOLOX-S的太阳能电池片表面缺陷检测[J]. 激光杂志, 2024, 45(7): 118-123.
WANG S Q, ZHU W X, ZHANG Z Y, et al. Surface defect detection of solar cells based on improved YOLOX-S[J]. Laser Journal, 2024, 45(7): 118-123.
[6] 周启宸, 王伯超. 基于改进YOLOv7的太阳能电池片表面缺陷检测[J]. 计算机应用, 2023, 43(S2): 223-228.
ZHOU Q C, WANG B C. Surface defect detection of solar cells based on improved YOLOv7[J]. Journal of Computer Applications, 2023, 43(S2): 223-228.
[7] 史册, 南新元. 改进InceptionV3与迁移学习的太阳能电池板缺陷识别[J]. 计算机工程与科学, 2023, 45(4): 646-653.
SHI C, NAN X Y. Improved InceptionV3 and transfer lear-ning for solar panel defect recognition[J]. Computer Engineering & Science, 2023, 45(4): 646-653.
[8] 于舜, 夏炎, 郭朋伟, 等. 一种基于深度卷积神经网络的太阳能电池片缺陷检测方法[J]. 传感技术学报, 2023, 36(7): 1165-1170.
YU S, XIA Y, GUO P W, et al. A defect detection method of solar cell based on deep convolutional neural network[J]. Chinese Journal of Sensors and Actuators, 2023, 36(7): 1165-1170.
[9] 张鹏飞, 王淑青, 黄剑锋, 等. 基于机器视觉的太阳能电池片表面缺陷检测[J]. 组合机床与自动化加工技术, 2022(8): 144-147.
ZHANG P F, WANG S Q, HUANG J F, et al. Surface defect detection method of solar cell based on vision[J]. Modular Machine Tool & Automatic Manufacturing Technique, 2022(8): 144-147.
[10] VLAMINCK M, HEIDBUCHEL R, PHILIPS W, et al. Region-based CNN for anomaly detection in PV power plants using aerial imagery[J]. Sensors, 2022, 22(3): 1244.
[11] RODRIGUEZ-VAZQUEZ J, PRIETO-CENTENO I, FERNANDEZ-CORTIZAS M, et al. Real-time object detection for autonomous solar farm inspection via UAVs[J]. Sensors, 2024, 24(3): 777.
[12] HUANG J, ZENG K Y, ZHANG Z J, et al. Solar panel defect detection design based on YOLOv5 algorithm[J]. Heliyon, 2023, 9(8): e18826.
[13] DWIVEDI D, BABU K V S M, YEMULA P K, et al. Identification of surface defects on solar PV panels and wind turbine blades using attention based deep learning model[J]. Engineering Applications of Artificial Intelligence, 2024, 131: 107836.
[14] YANG X Y, ZHANG Q, WANG S Y, et al. Detection of solar panel defects based on separable convolution and convolutional block attention module[J]. Energy Sources, Part A: Recovery, Utilization, and Environmental Effects, 2023, 45(3): 7136-7149.
[15] WANG C Y, YEH I H, LIAO H Y M. YOLOv9: learning what you want to learn using programmable gradient information[J]. arXiv:2402.13616, 2024.
[16] WANG C Y, BOCHKOVSKIY A, LIAO H M. YOLOv7: trainable bag-of-freebies sets new state-of-the-art for real-time object detectors[C]//Proceedings of the 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2023: 7464-7475.
[17] GE Z, LIU S, WANG F, et al. YOLOX: exceeding YOLO series in 2021[J]. arXiv:2107.08430, 2021.
[18] GUO C X, FAN B, ZHANG Q, et al. AugFPN: improving multi-scale feature learning for object detection[C]//Proceedings of the 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2020: 12592-12601.
[19] HUANG K, WU T, SU H, et al. MonoDTR: monocular 3D object detection with depth-aware transformer[C]//Proceedings of the 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2022: 4002-4011.
[20] LEE C Y, XIE S, GALLAGHER P, et al. Deeply-supervised nets[C]//Proceedings of the Eighteenth International Conference on Artificial Intelligence and Statistics, 2015: 562-570.
[21] WANG C Y, YEH I H, LIAO H Y M. You only learn one representation: unified network for multiple tasks[J]. arXiv: 2105.04206, 2021.
[22] SZEGEDY C, LIU W, JIA Y Q, et al. Going deeper with convolutions[C]//Proceedings of the 2015 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2015: 1-9.
[23] IOFFE S, SZEGEDY C. Batch normalization: accelerating deep network training by reducing internal covariate shift[C]//Proceedings of the 32nd International Conference on Machine Learnin, 2015: 448-456.
[24] DING X H, ZHANG X Y, MA N N, et al. RepVGG: making VGG-style ConvNets great again[C]//Proceedings of the 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2021: 13728-13737.
[25] DING X H, ZHANG X Y, HAN J G, et al. Diverse branch block: building a convolution as an inception-like unit[C]//Proceedings of the 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2021: 10881-10890.
[26] DING X H, GUO Y C, DING G G, et al. ACNet: strengthening the kernel skeletons for powerful CNN via asymmetric convolution blocks[C]//Proceedings of the 2019 IEEE/CVF International Conference on Computer Vision. Piscataway: IEEE, 2019: 1911-1920.
[27] ZHENG Z H, WANG P, REN D W, et al. Enhancing geometric factors in model learning and inference for object detection and instance segmentation[J]. IEEE Transactions on Cybernetics, 2021, 52(8): 8574-8586.
[28] ZHANG H, ZHANG S. Shape-IoU: more accurate metric considering bounding box shape and scale[J]. arXiv:2312. 17663, 2023.
[29] ZHOU B L, KHOSLA A, LAPEDRIZA A, et al. Learning deep features for discriminative localization[C]//Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2016: 2921-2929.
[30] SELVARAJU R R, COGSWELL M, DAS A, et al. Grad-CAM: visual explanations from deep networks via gradient-based localization[C]//Proceedings of the 2017 IEEE International Conference on Computer Vision. Piscataway: IEEE, 2017: 618-626.