缺陷汽车玻璃检测方法

Defect car glass detection method

汽车玻璃生产过程中会造成断裂、划痕、漏点等表面缺陷,本文结合机器视觉与深度学习提出了自动识别缺陷玻璃的方法。首先,利用玻璃前景、背景人工合成缺陷样本解决负样本不足的问题;将玻璃缺陷细分为多个类别,同时对样本进行分类;将玻璃图片进行频域处理过滤背景噪音,再将其与玻璃灰度化后图片进行合成作为分类网络的输入;构造以Alexnet网络为模板的多分类网络进行训练和预测。经过实验验证,该方法准确有效,为玻璃缺陷检测提供了一种可靠的检测方法。

In the production process of automobile glass,surface defects such as breaks,scratches,leaks,and fractures will be caused. This paper proposes a method to automatically identify defective glass by combining machine vision and deep learning. First,use the glass foreground and background to artificially synthesize defect samples to solve the problem of insufficient negative samples;subdivide the glass defects into multiple categories and classify the samples at the same time;process the glass image in the frequency domain to filter the background noise,and then combine it with the glass After graying,the picture is synthesized as the input of the classification network;a multi-classification network with the Alexnet network as the template is constructed for training and prediction. After experimental verification,this method is accurate and effective,we provides a reliable detection method for glass defect detection.