文献类型：期刊文献

中文题名：基于改进Faster R-CNN的工业零件表面缺陷检测

英文题名：Surface Defect Detection of Metallic Parts Based on Improved Faster R-CNN

作者：赵卓然[1];堵威[1];姜庆超[1];曹志兴[1]

机构：[1]华东理工大学信息科学与工程学院,上海200237

年份：2026

卷号：33

期号：5

起止页码：805

中文期刊名：控制工程

外文期刊名：Control Engineering of China

收录：;北大核心:【北大核心2023】；

基金：国家重点研发计划项目(2021YFB3301303)。

语种：中文

中文关键词：瑕疵检测;Faster R-CNN算法;金属零件;小瑕疵和形状不规则瑕疵

外文关键词：Defect detection;Faster R-CNN algorithm;metallic parts;small and irregularly shaped defects

摘要：为提高工业生产线中金属零件表面瑕疵检测精度,提出了一种基于改进Faster RCNN的缺陷检测网络。首先,在特征提取阶段增加多个感受野,以提取到更细致的特征信息;在区域建议阶段使用级联RPN网络和自适应卷积结构,逐步细化建议框的位置;在感兴趣区域特征提取阶段,聚合FPN所有输出层的特征信息,以充分利用获取到的缺陷特征信息,这些改进从网络结构上提高了模型对小瑕疵和形状不规则瑕疵的检测能力。最后,实验结果表明,与当前最优的目标检测网络相比,此方法具有更高的精度,可以广泛应用于检测金属零件表面的腐蚀、划痕、划伤、积屑瘤等缺陷。
In order to improve the accuracy of surface defect detection for metallic parts in Metallic production lines,a defect detection method based on an improved Faster R-CNN is proposed.Firstly,we introduced more receptive fields in the feature extraction stage,thereby allowing finer feature extraction.For the region proposal phase,a cascaded RPN network and adaptive convolution structure were employed to gradually refine the positioning of the proposal box.For the interest region feature extraction,the feature information from all output layers of the FPN was aggregated to fully utilize the defect feature information obtained,enhancing the network's detection capability for small and irregularly shaped defects.Finally,experimental results showed that the detection accuracy of our method outperformed that of the state-of-the-art object detection networks,and our method hence is of high value to be ultilized to detect surface defects such as corrosion,scratches,scuffs,and built-up edge of metallic parts.