基于EBS-YOLOv5的钢铁表面缺陷检测
Steel Surface Defect Detection Based on EBS-YOLOv5
耿冰 1廖宇 1冯旭 1崔琨 1胡钰航 1崔益博1
作者信息
- 1. 湖北民族大学智能科学与工程学院,湖北恩施 445000
- 折叠
摘要
随着人工智能在工业检测领域的应用日益增加,如何快速准确地检测缺陷成为了一个热点问题.钢铁是一种重要的工业原材料,针对钢铁表面缺陷的检测问题,提出了一种基于YOLOv5的改进模型.该模型主要有2点改进:一是在头部加入SENet(压缩和激励网络)注意力机制,增强全局检测能力,使模型能够学习不同通道的权重信息;二是加入EVC(显性视觉中心)模块,使得模型在关注全局信息的同时,也不忽略层内局部特征和角部信息.使用了公开的数据集NEU-DET(东北大学缺陷检测任务)验证了其可行性,试验结果表明,EBS-YOLOv5模型比起YOLOv5s的准确率有了较明显的提升,mAP50(平均精度指标)提高了6.2％,仍然保持了154的检测帧率,满足工业场景下对钢铁表面的检测需求.
Abstract
With the increasing application of artificial intelligence in the field of industrial inspection,the rapid and accurate detection of defects has become a focal issue.Steel is a crucial industrial raw material,and addressing the challenge of surface defect detection in steel has led to the proposal of an improved model based on YOLOv5.The model incorporates two key enhancements:Firstly,it integrates the SENet attention mechanism into the head to augment global detection capabilities,allowing the model to learn the weight information from different channels;Secondly,it introduces the EVC module,enabling the model to focus on global information without neglecting local features and corner details within each layer.Feasibility testing was conducted using the publicly available NEU-DET dataset.Experimental results indicate that the EBS-YOLOv5 model demonstrates a significant improvement in accuracy compared to YOLOV5s,with a 6.2％increase in Map50,while maintaining a detection frame rate of 154 frames per second,thus satisfying the requirements of steel surface defect detection in industrial scenarios.
关键词
钢铁表面/缺陷检测/EBS-YOLO/SENetKey words
steel surface/defect detection/EBS-YOLO/SENet引用本文复制引用
出版年
2024
NETL
NSTL
万方数据