基于改进Faster R-CNN的热轧带钢表面缺陷检测  

Surface Defect Detection of Hot-rolled Strip Steel Based on Improved Faster R-CNN

作　　者：邓慧[1] 曾磊 DENG Hui;ZENG Lei(School of Intelligent Manufacturing and Construction Engineering,YongZhou Vocational Technical College,Yongzhou 425000,China;School of Automotive and Traffic Engineering,Jiangsu University,Zhenjiang 212000,China)

机构地区：[1]永州职业技术学院智能制造与建筑工程学院,湖南永州425000 [2]江苏大学汽车与交通工程学院,江苏镇江212000

出　　处：《控制工程》2024年第4期752-759,共8页Control Engineering of China

摘　　要：热轧带钢是钢铁行业的重要产品,其表面缺陷是影响产品质量的重要因素。针对传统缺陷检测算法存在的过程繁琐、精度不足和效率低下等问题,提出一种基于改进更快速区域卷积神经网络(faster region-based convolutional neural network,Faster R-CNN)的检测算法,实现对热轧带钢表面缺陷的高效、高精度检测。首先,采用特征相加的方法对底层细节特征和高层语义特征进行融合;然后,采用精准的感兴趣区域池化(precise region of interest pooling,Precise ROI Pooling)获取固定大小的特征向量,避免特征出现位置偏差;最后,利用均值偏移聚类算法对带钢数据集进行聚类,获得适用于热轧带钢表面缺陷检测的先验框尺寸。实验结果表明,所提算法在热轧带钢表面缺陷检测数据集上的平均精度均值达到了85.34%,检测速度为23.5帧/s,且鲁棒性良好,满足实际的工业检测需求。Hot-rolled strip steel is an important product in the steel industry,and its surface defects are crucial factors affecting product quality.In view of the cumbersome process,insufficient accuracy,and low efficiency of conventional defect detection algorithms,a detection algorithm based on improved faster region-convolutional neural network(Faster R-CNN)is proposed to achieve efficient and accurate defect detection for surface defects of hot-rolled strip steel.Firstly,the method of feature summation is adopted to fuse the underlying detailed features and the high-level semantic features.Then,precise region of interest pooling(Precise ROI Pooling)is used to obtain a fixed-size feature vector to avoid the positional deviation of the features.Finally,the mean-shift algorithm is used to obtain the priori frame size applicable to the surface defect detection of hot-rolled strip steel by clustering on the strip dataset.The experimental results show that the proposed algorithm achieves a mean average precision of 85.34%on the surface defect detection dataset of hot-rolled strip steel with a detection speed of 23.5 frames/s,and good robustness,which meets the needs of practical industrial detection.

关 键 词：表面缺陷检测 Faster R-CNN 特征融合 Precise ROI Pooling 均值偏移 

分 类 号：TP278[自动化与计算机技术—检测技术与自动化装置]