基于Faster R-CNN的刨花板表面缺陷检测研究

为了克服刨花板表面缺陷人工目视检测的局限性,实现对多种缺陷准确、实时检测,提出一种基于Faster R-CNN的检测方法。运用从工厂生产现场获取的各种表面缺陷图,制作成一个包含3566张刨花板表面缺陷图像数据集,其中主要包括胶块、水印、砂痕、杂物、粗刨花5种缺陷类型。通过用该数据集对Faster R-CNN在ZF、VGG16和ResNet101不同特征提取网络下的不同锚点(Anchor)设置模型分别进行训练、验证和测试,并对比了不同参数对检测精度的影响。结果显示,该方法能有效检测刨花板表面缺陷,且模型在ResNet101作为特征提取网络时准确率最高。在对训练好的Faster R-CNN模型的鲁棒性进行评估和验证中,模型对122张新图像的5种缺陷类型进行检测,测试的5种缺陷类型识别率分别为92.31%、91.84%、90.57%、96.88%和95.24%,平均检测率为93.37%,测试结果表明该方法能为基于机器视觉刨花板表面缺陷检测系统提供良好支撑。

Research on Surface Defect Detection of Particleboard Based on Faster R-CNN

In order to overcome the limitation of artificial visual inspection of surface defects of particleboard and realize accurate and real-time detection of various defects,a detection method based on Faster R-CNN is proposed.Using various surface defect maps acquired from the factory production site,a set of surface defect image data of 3566 particleboard was produced,which mainly includes five types of defects:rubber block,watermarking,sand mark,debris and rough particleboard.The data set is used to train,validate and test different anchor setting models of Faster R-CNN under different feature extraction networks of ZF,VGG16 and ResNet101,and the effects of different parameters on detection accuracy are compared.The results show that this method can effectively detect the surface defects of particleboard,and the model has the highest accuracy when ResNet101 is used as feature extraction network.In the evaluation and validation of the robustness of the trained Faster R-CNN model,the model detects five types of defects in 122 new images.The recognition rates of the five types of defects tested are 92.31%,91.84%,90.57%,96.88%and 95.24%,respectively.The average rate is 93.37%.The test results show that the method can provide a good performance for the surface defect detection system of particleboard based on machine vision.