针对多晶硅太阳电池片缺陷尺度不一、跨度大、表面纹理背景复杂、缺陷微小微弱的特性,基于Yolov4提出一种新型检测框架EDANet,构建两个新型的模块：跨尺度空间增强模块（CGSE）和自校准压缩-激励模块（SCSE）。CGSE模块作为空间注意力以多尺度的形式将特征进行融合、抑制背景、突出前景、重加权特征图,引导网络学习正确的前景背景特征分布;SCSE模块以通道注意力的形式对高层特征的每个空间位置建立长距离依赖关系,通过显式合并信息帮助网络生成更多的辨识性表示,以区分微小微弱缺陷。实验结果表明：该网络的均值平均精度（mAP）值达到92.07%,在多晶硅太阳电池缺陷检测精度上有明显提升。
Abstract
Aiming at the characteristics of different defect scales, large span, complex surface texture background and small defects of polycrystalline silicon solar cells, a new detection framework EDANet is proposed based on Yolov4, and two new modules are constructed： cross-scale group space enhancement （ CGSE ）module and self-calibrated squeeze and excitation（SCSE） module. CGSE module integrates features, as spatial attention, in the form of multi-scale, suppresses background, highlights foreground, and reweights feature maps and guides the network to learn the correct foreground and background feature distribution. The SCSE module establishes a long-distance dependence on each spatial location of the high-level feature in the form of channel attention, and helps the network to generate more identification representations by explicitly merging information to distinguish small and weak defects. The experimental results show that the mAP value of the network reaches 92.07%, and the accuracy of defect detection of polycrystalline silicon solar cells is significantly improved.
关键词
目标检测 /
太阳电池 /
卷积神经网络 /
多尺度融合 /
注意力机制
Key words
object detection /
solar cells /
convolutional neural network /
multi-scale fusion /
attention mechanism
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
[1] JEN C Y, RICHTER C.Doping profile recognition applied to silicon photovoltaic cells using terahertz time-domain spectroscopy[J]. IEEE transactions on terahertz science and technology, 2014, 4(5): 560-567.
[2] 李洁, 袁知博, 秦嘉悦. 基于Sobel算子边缘检测的太阳电池缺陷特征提取方法[J]. 太阳能学报, 2021, 42(1): 63-68.
LI J, YUAN Z B, QIN J Y.Research on solar cells defects feature extraction based on Sobel operator edge detection[J]. Acta energiae solaris sinica, 2021, 42(1): 63-68.
[3] 王理顺, 钟勇, 李振东, 等. 基于深度学习的织物缺陷在线检测算法[J]. 计算机应用, 2019, 39(7): 2125-2128.
WANG L S, ZHONG Y, LI Z D, et al.On-line fabric defect recognition algorithm based on deep learning[J]. Journal of computer applications, 2019, 39(7): 2125-2128.
[4] 李林升, 曾平平. 改进深度学习框架Faster-RCNN的苹果目标检测[J]. 机械设计与研究, 2019, 35(5): 24-27.
LI L S, ZENG P P.Apple target detection based on improved Faster-RCNN framework of deep learning[J]. Machine design & research, 2019, 35(5): 24-27.
[5] 孙彦, 丁学文, 雷雨婷. 基于目标检测模型的人脸识别技术[J]. 计算机与网络, 2019, 45(22): 68-71.
SUN Y, DING X W, LEI Y T.Face recognition technology based on target detection model[J]. Computers & network, 2019, 45(22): 68-71.
[6] 顾佳晨, 高雷, 刘路璐. 基于深度学习的目标检测算法在冷轧表面缺陷检测中的应用[J]. 冶金自动化, 2019, 43(6): 19-22, 27.
GU J C, GAO L, LIU L L.Application of object detection algorithm based on deep learning for inspection of surface defect of cold rolled strips[J]. Metallurgical industry automation, 2019, 43(6): 19-22, 27.
[7] HOU W, TAO X, XU D.Combining prior knowledge with CNN for weak scratch inspection of optical components[J]. IEEE transactions on instrumentation and measurement, 2021, 70: 1-11.
[8] ZHANG X, HAO Y W, SHANGGUAN H, et al.Detection of surface defects on solar cells by fusing multi-channel convolution neural networks[J]. Infrared physics & technology, 2020, 108: 103334.
[9] CHEN H Y, PANG Y, HU Q D, et al.Solar cell surface defect inspection based on multispectral convolutional neural network[J]. Journal of intelligent manufacturing, 2020, 31(2): 453-468.
[10] SU B Y, CHEN H Y, CHEN P, et al.Deep learning-based solar-cell manufacturing defect detection with complementary attention network[J]. IEEE transactions on industrial informatics, 2021, 17(6): 4084-4095.
[11] TAO X, XU D, ZHANG Z T, et al.Weak scratch detection and defect classification methods for a large-aperture optical element[J]. Optics communications, 2017, 387: 390-400.
[12] YI X, SONG Y H, TANG X.Weak supervised surface defect detection method based on selective search and CAM[C]// 2019 Chinese Automation Congress(CAC), IEEE Hangzhou, China, 2019: 4386-4391.
[13] LI W, LIU K, ZHANG L Z, et al.Object detection based on an adaptive attention mechanism[J]. Scientific reports, 2020, 10(1): 11307.
基金
国家自然科学基金（62173124）; 河北省自然科学基金（F2019202305）
PDF(2357 KB)
