基于深度学习的太阳能电池板表面缺陷检测  

Surface Defect Detection on Solar Panels Based on Deep Learning

作　　者：阮龙森 李兴成 RUAN Longsen;LI Xingcheng(School of Mechanical Engineering,Jiangsu University of Technology,Changzhou 213001,China)

机构地区：[1]江苏理工学院机械工程学院,常州213001

出　　处：《组合机床与自动化加工技术》2025年第3期199-203,208,共6页Modular Machine Tool & Automatic Manufacturing Technique

基　　金：国家自然科学基金资助项目(51905235);江苏省自然科学基金资助项目(BK20191037)。

摘　　要：经过对太阳能电池板表面缺陷检测技术的深入研究,发现存在检测效率不高、缺陷规模较小和检测精度不高的问题,提出了一种基于YOLOv8的改进检测方案。设计了EMSC多尺度卷积模块,实现冗余特征信息的剔除,减少冗余计算,并提高模型检测精度;结合C3和ConvNeXt Block设计了CNeB模块改进主干网络,以提取丰富的电池板边缘细粒度信息,促进模型降低误检和漏检的问题;在YOLOv8网络模型的基础上引入BasicRFB模块,以提取丰富的上下文语义信息,促进模型对电池片丰富的表征能力。实验结果表明,改进算法在测试集上,相对于原始网络mAP@50%提升了4.2%,精确率P(Precision)提升了7.5%,参数量相对于YOLOv7-tiny减少了30.46%,能够实现准确的太阳能电池板表面的缺陷检测,基本上满足真实工况下的检测需求。After conducting thorough research on the surface defect detection technology of solar panels,it was identified that issues such as low detection efficiency,small defect sizes,and inadequate detection accuracy exist.Consequently,an improved detection scheme based on YOLOv8 was proposed in this study.The EMSC multiscale convolution module was designed to eliminate redundant feature information,reduce redundant calculations,and enhance the model′s detection accuracy.Furthermore,the CNeB module,incorporating C3 and ConvNeXt Block,was devised to refine the backbone network for extracting rich edge granularity information from the solar panels,thus aiding the model in minimizing false alarms and missed detections.Additionally,the BasicRFB module was introduced to the YOLOv8 network model to extract contextual semantic information,thereby bolstering the model′s ability to comprehensively represent solar cells.Experimental findings indicate that the improved algorithm achieved a 4.2%increase in mAP@50%and a 7.5%boost in Precision compared to the original network on the test set.Moreover,the parameter volume reduced by 30.46%relative to YOLOv7-tiny.This enables accurate detection of surface defects on solar panels,essentially meeting the detection requirements under real-world operating conditions.

关 键 词：YOLOv8 缺陷检测 太阳能电池板 EMSC CNeB BasicRFB 

分 类 号：TH165[机械工程—机械制造及自动化] TG659[金属学及工艺—金属切削加工及机床]