摘要
传统的图像处理方法对生产过程中各种金属板材表面缺陷检测效率低,难以满足工业生产的需求。为了提高金属板材表面缺陷检测的精度,文章提出了一种基于优化Faster R-CNN算法的金属板材表面缺陷检测方法,以残差网络ResNet50作为主干特征提取网络。首先,融合特征金字塔网络和可变形卷积网络以提高对小目标和不规则性缺陷的检测能力。然后,采用RoI Align和K-means++聚类算法对候选框进行优化,实现缺陷的精准定位。最后,将提出的模型运用在NEU-DET数据集中进行多次实验。实验结果表明,优化后的Faster R-CNN算法在此数据集上的mAP为78.7%,与原始网络相比提高了7.7%,并且其检测性能优于SSD、YOLOv5s和YOLOv7三类目标检测算法。
Traditional image processing methods have low efficiency in detecting surface defects of various metal sheets during the production process,making it difficult to meet the needs of industrial production.In order to improve the accuracy of metal sheet surface defect detection,a metal sheet surface defect detection method based on optimized Faster R-CNN algorithm is proposed.Using the residual network ResNet50 as the backbone feature extraction network.Firstly,the Feature Pyramid Network and Deformable ConvNets v2 are fused to improve the detection ability for small objects and irregular defects.Then,RoI Align and K-means++clustering algorithms are used to optimize the candidate boxes and achieve precise defect localization.Finally,the proposed model is applied to multiple experiments in the NEU-DET dataset.The experimental results shows that the mAP of the optimized Faster R-CNN algorithm on this dataset is 78.7%,which is 7.7%higher than the original network,and its detection performance is better than SSD,YOLOv5s,and YOLOv7 object detection algorithms.
作者
孔思曼
周晨阳
王家华
李林
孙践知
KONG Siman;ZHOU Chenyang;WANG Jiahua;LI Lin;SUN Jianzhi(School of Computer Science and Engineering,Beijing Technology and Business University,Beijing 100048,CHN)
出处
《制造技术与机床》
北大核心
2024年第1期171-178,共8页
Manufacturing Technology & Machine Tool
