摘要
针对硅钢片表面缺陷检测容易出现漏检、检测区域不准确、多重检测等问题,提出一种改进YOLOv7-tiny的硅钢片表面缺陷检测算法:SMCS-YOLOv7 tiny算法。首先,基于SimAM注意力机制构建ELAN-SIM模块,增强模型对目标特征信息的提取能力;其次,使用Mish激活函数代替原网络中的Leaky ReLU激活函数,提高模型的非线性特征提取能力;再次,在Neck层添加CoordConv模块,增强模型的空间感知能力;最后,采用SIoU损失函数加快模型收敛速度。实验结果表明,SMCS-YOLOv7 tiny算法在硅钢片缺陷数据集上的准确度P、召回率R、mAP@0.5分别达到88%、78.1%和85.7%,较原YOLOv7-tiny算法分别提高了2.2%、3%和2.5%。相比传统的硅钢片表面缺陷检测方法,提出的算法实现了更精准检测效果。
Aiming to address the issues of missed detections,inaccurate detection areas,and multiple detections in the surface defect detection of silicon steel sheets,we propose an improved algorithm called SMCS-YOLOv7 tiny.Firstly,we construct the ELAN-SIM module based on the SimAM attention mechanism to enhance the model′s capability of extracting target features.Secondly,we replace the Leaky ReLU activation function with the Mish activation function to improve the model′s non-linear feature extraction ability.Thirdly,we add the CoordConv module to the Neck layer to enhance the model′s spatial awareness.Finally,we adopt the SIoU loss function to accelerate model convergence.Experimental results demonstrate that the SMCS-YOLOv7 tiny algorithm achieves accuracy(P),recall(R),and mAP@0.5 of 88%,78.1%,and 85.7% respectively on the silicon steel sheet defect dataset,resulting in improvements of 2.2%,3%,and 2.5% compared to the original YOLOv7-tiny algorithm.Compared to traditional methods for surface defect detection on silicon steel sheets,our proposed algorithm achieves more accurate detection results.
作者
李克讷
陈福丁
李永革
樊香所
陈健民
LI Kene;CHEN Fuding;LI Yongge;FAN Xiangsuo;CHEN Jianmin(College of Automation,Guangxi University of Science and Technology,Liuzhou 545000,China;Guangxi Liuzhou Special Transformer Co.,Ltd.,Liuzhou 545000,China)
出处
《组合机床与自动化加工技术》
北大核心
2025年第2期171-176,共6页
Modular Machine Tool & Automatic Manufacturing Technique
基金
广西科技重大专项基金项目(AA23062091)
国家自然科学基金资助项目(61663003)。
