基于双通道多模态卷积网络的电子元器件缺陷分类检测

Defect Classification in Electronic Components Using Dual-Channel Multi-Modal Convolutional Network

摘要: 目前在电子元器件的缺陷检测领域，主要是基于机器学习的传统图像处理算法和基于深度学习的智能图像算法。深度卷积网络可以自动提取图像更深层的特征，避免了传统图像处理算法特征提取的复杂性和盲目性。因此，本文提出一种基于双通道多模态卷积网络的电子元器件缺陷分类方法，将电子元器件影像分为Pass和Fail两个类别。首先，为提高模型的泛化能力，本文使用了Top光源和Side光源两个模态的数据。其次，为解决训练样本不足和类间样本不平衡的问题，使用PCA Jittering对数据集进行扩增。最后，为了实现模型对不同模态数据的有效覆盖，本文设计了一种基于特征融合的双通道卷积神经网络。实验表明，本文的电子元器件缺陷分类方法能够更有效地处理训练样本不足的问题，并在训练过程中通过特征融合提高了模型的性能。这将为电子信息产业的发展提供重要的技术支持和应用前景。

Abstract: In the field of defect detection for electronic components, the main methods are traditional image processing algorithms based on machine learning and intelligent image algorithms based on deep learning. Deep convolutional networks can automatically extract deeper features of images, avoiding the complexity and blindness of feature extraction in traditional image processing algorithms. Therefore, this paper proposes a method for classifying electronic component defects based on a dual-channel multimodal convolutional network, dividing electronic component images into two categories: Pass and Fail. Firstly, to improve the generalization ability of the model, this paper uses data from two modalities: Top light and Side light. Secondly, to solve the problem of insufficient training samples and class imbalance between samples, PCA Jittering is used to augment the dataset. Finally, in order to achieve effective coverage of the model for different modal data, this paper designs a dual-channel convolutional neural network based on feature fusion. Experiments show that the electronic component defect classification method proposed in this paper can more effectively deal with the problem of insufficient training samples and improve the performance of the model through feature fusion during training. This will provide important technical support and application prospects for the development of the electronic information industry.