Abstract: In the current riveting production process of Passat roof, manual visual inspection is used to check whether there is rivet omission, which is inefficient, and there are problems such as wrong inspection and missing inspection. In this paper, the deep learning convolution neural network is introduced into the roof riveting detection, and a convolution neural network algorithm based on ReLU activation function is proposed. Experiments show that the algorithm has high accuracy and strong robustness, which provides a new idea for the application of deep learning to vehicle body visual detection, The success rate of experimental detection reaches 99.87%, meet the quality assurance requirements in the production process, which has guiding significance for the engineering application and promotion.

Key words: Automobile manufacturing; Visual recognition; Deep learning; Neural network

摘要： 在 Passat 车顶拉铆生产过程中，目前采用人工目视的方法检查是否存在铆钉遗漏问题， 该方法检查效率较低且存在错检漏检等问题。文章将深度学习卷积神经网络引入到车顶拉铆 检测中，提出了一种基于 ReLU 激活函数的卷积神经网络算法，并通过实验表明该算法具有 精度高、鲁棒性强等特点，为深度学习应用于车身视觉检测提供了一种新的思路。实验检测 成功率可以达到 99.87%，满足生产过程中的质量认定要求，对工程应用推广具有指导意义。

关键词: 汽车制造；视觉识别；深度学习；神经网络