基于卷积神经网络的钣金件表面缺陷分类识别方法

Classification and Recognition Method of Sheet Metal Parts Surface Defects Based on Convolution Neural Network

摘要:

针对国防军工、电子信息等领域对多批次、小批量钣金零件快速、智能制造的需求,提出了一种基于卷积神经网络的少样本钣金件表面缺陷分类识别方法。首先基于卷积神经网络的网络架构,搭建出了经典的分类模型,并在实验中进行了参数修改,以达到实际生产中的表面缺陷检测要求；其次利用缺陷分割提取的方法获得卷积网络训练模型的样本集,并进行数据增强。实验结果表明,该模型的准确度可达97.02%；最后利用窗口滑移检测方法使待检测零件与模型进行对比,实现了对缺陷的分类和缺陷位置的标记。经实验验证,该方法的准确性和实时性均可满足实际工业生产要求。

Abstract:

In order to meet the requirement of rapid and intelligent manufacturing of multi-batch and small-batch sheet metal parts in the fields of national defense, military industry and electronic information, a method of surface defect detection for sheet metal parts with few samples based on convolution neural network is proposed. Firstly, based on the network model of convolution neural network, the classical classification model is built, and the parameters are modified in the experiment to meet the needs of surface defect detection in actual production. Secondly, the sample set of convolution neural network training model is obtained by defect segmentation and extraction method, and the data are enhanced. The experimental results show that the accuracy of the model can reach 96.88%. Finally, the window sliding detection method is used to compare the part to be tested with the model to realize the classification of defects and the marking of defect location. Experiments show that the accuracy and real-time performance of the method meet the requirements of actual industrial production.