基于轻量化YOLO网络的热轧带钢表面缺陷检测
Detection of Surface Defects in Hot-rolled Strip Steel Based on Lightweight YOLO Network
夏旭 1阮佩1
作者信息
- 1. 西安工程大学电子信息学院,陕西 西安 710048
- 折叠
摘要
针对热轧带钢表面缺陷检测中检测精度不高、卷积特征对尺度敏感的问题,设计了高效的特征提取模块(FEM)和增强的多尺度特征模块(MFM),并提出了一种基于深度学习的轻量化的热轧带钢表面缺陷检测方法,即Better Lightweight YOLO(BL-YOLO).实验结果表明,该缺陷检测网络在性能和消耗之间达到了很好的平衡,以 61.9 fps达到了 80.1的mAP.
Abstract
Aiming at the problems of low detection accuracy and scale sensitivity of convolutional features in hot-rolled strip surface defect detection,this paper designs an efficient feature extraction module(FEM)and an enhanced multi-scale feature module(MFM),and proposes a deep learning-based lightweight hot-rolled strip surface defect detection method,Better Lightweight YOLO(BL-YOLO).Experimental results show that this defect detection network achieves a good balance between performance and consumption,achieving a mAP of 80.1 with 61.9 fps.
关键词
深度学习/缺陷检测/NEU-DET/轻量化技术/多尺度策略Key words
deep learning/defect detection/NEU-DET/lightweighting technique/multi-scale strategy引用本文复制引用
出版年
2024
NETL
NSTL
万方数据