基于AFSD-YOLOv7的钢材表面缺陷检测
投稿时间:2023-04-07  修订日期:2023-05-12  点此下载全文
引用本文:
摘要点击次数: 151
全文下载次数: 0
作者单位邮编
付帅 上海工程技术大学电子电气工程学院 201620
凌铭* 上海工程技术大学电子电气工程学院 201620
楚东港 上海工程技术大学电子电气工程学院 
基金项目:上海市技术标准项目
中文摘要:钢材表面缺陷对于钢材行业来说是一个巨大的挑战。针对传统的钢材缺陷检测方法存在着效率低,检测精度不高等问题,基于YOLOv7设计了一种AFSD-YOLOv7模型进行实时地钢材表面缺陷检测。首先,在YOLOv7模型中使用一种轻量化卷积结构替换标准卷积结构以加速模型的推理过程;然后采用快速空间金字塔池化结构替换原始空间金字塔池化结构以加速网络的特征提取过程;最后添加改进的ECA-Net注意力机制以提升模型检测精度。实验结果表明,AFSD-YOLOv7能够对钢材缺陷进行有效识别,相比于YOLOv7模型,计算量减少了54.8%,mAP提高了3.2%,对于钢材表面缺陷检测具有实际应用价值。
中文关键词:钢材  缺陷检测  YOLOv7  神经网络  深度学习  注意力机制  标准卷积
 
Steel Surface Defect Detection Based on AFSD-YOLOv7
Abstract:Surface defects on steel are a significant challenge for the steel industry. Traditional steel defect detection methods suffer from low efficiency and accuracy. To address these issues, an AFSD-YOLOv7 model has been designed for real-time steel surface defect detection. First, a lightweight convolutional structure was used to replace the standard convolutional structure in the YOLOv7 model, speeding up the inference process. Next, a fast spatial pyramid pooling structure was used to replace the original spatial pyramid pooling structure to accelerate the network's feature extraction process. Finally, an improved ECA-Net attention mechanism was added to enhance the model's detection accuracy. Experimental results show that AFSD-YOLOv7 can effectively identify steel defects. Compared to the YOLOv7 model, AFSD-YOLOv7 reduces computation by 54.8% and improves mAP by 3.2%, indicating significant practical value for steel surface defect detection.
keywords:Steel  defect detection  YOLOv7  neural networks  deep learning  attention mechanism  standard convolution
查看全文   查看/发表评论   下载pdf阅读器