基于改进Faster RCNN的钢板表面缺陷检测研究  

Research on Surface Defect Detection of Steel Plate Based on Improved Faster RCNN

作　　者：卢勇拾 张滢雪 司占军[1] 于彦辉 王庆 LU Yong-shi;ZHANG Ying-xue;SI Zhan-jun;YU Yan-hui;WANG Qing(College of Artificial Intelligence,Tianjin University of Science and Technology,Tianjin 300457,China)

机构地区：[1]天津科技大学人工智能学院,天津300457

出　　处：《印刷与数字媒体技术研究》2024年第3期244-251,共8页Printing and Digital Media Technology Study

摘　　要：钢铁是我国工业生产的重要原材料之一,其表面质量问题会直接影响产品的使用,从而带来无法预知的风险,故对钢铁表面进行缺陷检测具有重要意义。而在缺陷检测过程中,存在因裂痕缺陷特征不明显,导致缺陷定位不准确以及检测难度高等问题。针对以上问题,本研究提出一种改进的Faster RCNN算法,在主干特征提取网络上引入自适应模块,增强网络提取有效特征的能力,同时使用DBSCAN聚类算法取得合适的先验框,大大提高了算法的检测效率。实验结果表明,改进的Faster RCNN算法模型对不明显的缺陷特征检测能力大幅度的提升,相比其他检测算法,在钢板表面缺陷检测中能达到高质量、缺陷定位准确、分类成功率高的效果。Steel is one kind of the important raw materials in China’s industrial production.The surface quality problems of steel will directly affect the use of the products,which will bring unpredictable risks.Therefore,it is significance to carry out defect detection on the surface of steel.In the defect detection process,some defect features may not be obvious,resulting in inaccurate defect localization and high detection difficulty.An improved Faster RCNN algorithm was proposed in this study,which would introduce an adaptive module on the backbone feature extraction network,enhance the ability of the network to extract more effective features.DBSCAN clustering algorithm was used to obtain a suitable anchor frame,which greatly improved the dete ction efficiency of the algorithm.The experimental results showed that the proposed Faster RCNN model can achieve a substantial improvement in the detection of obscure defect features.Compared with other detection algorithms,the proposed algorithm can aobtain accurate defect localization and a high classification success rate in the detection of defects on the surface of steel plate.

关 键 词：Faster RCNN DBSACN聚类 目标检测 锚框 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]