基于深度学习的瓷砖在线快速无损检测系统开发

Development of Rapid Online Non-Destructive Inspection System for Ceramic Tile Based on Deep Learning

提出了一种基于双流卷积神经网络的瓷砖缺陷分类和识别系统。设计瓷砖质量检测平台用来采集图像样本,通过图像增强方式扩展样本数量;利用双边滤波器进行图像预处理,采用Sobel算子计算图像的梯度。设计了一种采用最大值融合策略的双流卷积神经网络模型,实现了fc6层的决策级特征融合。该模型以瓷砖的原始图像以及其对应的缺陷区域二值图像作为输入,分别提取特征实现特征融合,然后输入到SVM分类器进行缺陷分类和识别。通过实验分析,证明该模型具有良好的收敛性、准确性和稳定性。

A defect classification and identification system of ceramic tile based on two-stream convolutional neural network is proposed in this paper.The ceramic tile quality detection platform is designed to collect image samples,and the sample number is expanded by image enhancement.The bilateral filter is used to preprocess the image,and Sobel operator is used to calculate the gradient of the image.A two-stream convolution neural network model with maximum fusion strategy is designed to realize feature fusion on the fc6 layer.The original image of ceramic tile and its corresponding binary image of defect area are taken as input to extract features respectively to realize feature fusion which is input them to SVM classifier for defect classification and recognition.Through experimental analysis,it is proved that the model has good convergence,accuracy and stability.