深度神经网络模型用于图像斑点微瑕疵检测

Speckled Micro-defects Detection Based on Deep Neural Network Learning

在印染行业中,产品外观经常会出现很多斑点状缺陷,此类缺陷属于微瑕疵,其人工检测成本高且检测难度大。针对产品外观斑点状缺陷,本文提出了一种基于机器视觉的图像微瑕疵检测方法,其主要思想是基于Faster-RCNN框架构建轻量化网络模型,并利用样本梯度特征信息进行非端对端网络训练。所设计的非端对端训练模式不仅能有效缩短模型训练时间,还可以提升模型推理能力。实验结果表明,针对不同类型班点微瑕疵,本文提供的检测算法具有高效的局部检测精度,且可应用于其他领域的类似斑点任务检测。

In the printing and dyeing industry,a lot of speckled defects will often occur on product appearances.The speckled defect commonly belongs to the micro-defect,and manual detection is costly and difficult.In this paper,a micro-defect detection method based on machine learning is proposed for the speckled defect of product appearance.The main idea is to build a lightweight network based on the Faster-RCNN framework,and use the characteristic information of sample gradient graph to train the network.Our designed non-end-to-end training mode can effectively simplify the training time and improve the reasoning ability of the model.The experimental results show that the proposed detection scheme not only has high local detection accuracy for different types of micro blemish spots,but also can be transferred and applied to similar detection tasks in other fields.