重庆科技大学学报（自然科学版）2025，Vol.27Issue(6)：59-68,10.DOI:10.19406/j.issn.2097-4531.2025.06.007
基于改进Faster R-CNN的带钢表面缺陷检测方法
A Steel Surface Defect Detection Method Based on an Improved Faster R-CNN
摘要
Abstract
To address the problems of insufficient sample size for steel surface defect detection,poor accuracy in small-target defect detection,and unstable bounding box regression for multi-scale defects,an improved steel sur-face defect detection algorithm is proposed based on Faster R-CNN,which has small-target feature recognition capa-bilities.First,a lightweight FastGAN is employed to synthesize steel surface defect images and augment the data-set,thereby enhancing the model's generalization ability.Second,a ResNet50-DRB feature extraction network in-tegrating structural re-parameterization is constructed to expand the receptive field and improve small-target defect detection accuracy.Finally,a regression loss function combining Smooth L1 and Alpha-IoU is established with the introduction of a power parameter to enhance the model's localization capability for small-target and irregular de-fects.Experimental results demonstrate that the improved algorithm achieves a 2.01 percentage point improvement in PmA@0.5 on the NEU-DET dataset compared to the baseline model,while reducing parameter count and computa-tional complexity by 43.9％and 26.0％,respectively,thus achieving an optimal balance between detection accura-cy and model complexity.关键词
带钢表面缺陷检测/Faster R-CNN/生成对抗网络/结构重参数化/轻量化Key words
steel surface defect detection/Faster R-CNN/generative adversarial network/structural re-parameter-ization/lightweight分类
信息技术与安全科学引用本文复制引用
李文杰,苏盈盈,罗林,张乐,彭杰,杜谦..基于改进Faster R-CNN的带钢表面缺陷检测方法[J].重庆科技大学学报（自然科学版）,2025,27(6):59-68,10.基金项目
重庆市自然科学基金面上项目"热轧钢带智能检测中DeLiGAN和迁移学习的改进及应用"(CSTB2022NSCQ-MSX1425) （CSTB2022NSCQ-MSX1425）
重庆科技大学硕士研究生创新计划项目"PBE-RTDETR改进方法及其在带钢表面小目标缺陷检测中的应用"(YKJCX2421001),"基于Atlas200I DK A2的印刷电路板缺陷智能检测系统设计与实现"(YKJCX2420401) （YKJCX2421001）
