| 摘要: |
| 焊接技术在多个领域广泛应用，近年来焊缝缺陷的自动检测已成为研究的热点。本文针对铝合金熔焊焊缝的X射线图像，采用VGG-16卷积神经网络作为基础网络，提出了一种SC-VGG的新型网络结构。该结构通过引入多尺度合成卷积层来替代传统的单一卷积层，优化了训练过程中的损失函数，使网络更加聚集于焊缝缺陷类型的精确预测。实验结果表明，SC-VGG网络结构在训练过程中展现出了良好的收敛性。与其他网络相比，SC-VGG网络在提取焊缝缺陷特征方面表现优异，其平均准确率和召回率分别达到了95.86%和98.33%，为焊缝缺陷的自动化分类提供了算法支撑。 |
| 关键词: 焊缝检测 缺陷识别 VGG-16模型 合成卷积 |
| 基金项目: |
|
| Research on Weld Defect Recognition Technology Based on Improved VGG-16 Network Structure |
|
Liu Xiaojia Cao Lijun Liu Huan Wang Fei Wei Quan
|
|
Shanghai Aerospace Precision Machinery Research Institute, Shanghai 201600
|
| Abstract: |
| Welding technology is widely used in multiple fields, and the automatic detection of weld defects has become a research hotspot in recent years. In this paper, aiming at the X-ray images of aluminum alloy fusion welding seams, a new network structure called SC-VGG is proposed, using the VGG-16 convolutional neural network as the basic network. This structure replaces the traditional single convolutional layer with a multi-scale synthetic convolutional layer and optimizes the loss function in the training process, making the network more focused on accurate prediction of weld defect types. Experimental results show that the SC-VGG network structure exhibits good convergence during the training process. Compared with other networks, the SC-VGG network performs excellently in extracting weld defect features, with an average accuracy and recall rate reaching 95.86% and 98.33% respectively, providing algorithm support for the automatic classification of weld defects. |
| Key words: weld inspection defect identification VGG-16 model synthetic convolution |
