应用深度卷积神经网络的色织物缺陷检测

针对织物缺陷检测时传统人工的误检率、漏检率较高问题，提出一种应用深度卷积神经网络的色织物缺陷检测算法。因织物图像采集过程中含有较多噪声且信噪比较低，先对缺陷织物进行最优尺寸高斯滤波，有效滤除细节噪声；再根据织物图像特征建立深度卷积神经网络，利用径向基神经网络的非线性映射能力作用于卷积神经网络，并通过反向传播算法调整权值参数，获取无缺陷样本与训练样本之间的映射函数；最后，利用映射函数及特征字典重构图像并提取特征，根据Meanshift算法分割缺陷，确定缺陷位置。结果表明：应用深度卷积神经网络的缺陷检测算法对色织物图像库中的缺陷图像可实现提高检测效率、缩短检测时间，获取准确缺陷位置的目的。

Yarn-dyed fabric defect detection based on deep-convolutional neural network

Focusing on the problems of high error detection and omission rate of traditional artificial fabric defect detection，this paper presents a yarn-dyed fabric defect detection method, which based on thedeep-convolutional neural network. The fabric image contains much noise and has low signal noise ratio (SNR), and optimal dimension Gauss filter as preprocessing is conducted firstly for the sampled images to remove the detailed noise. Secondly, the deep-convolutional neural network is established based on the features of fabric samples, nonlinear mapping ability of radial basis function neural network acts upon convolutional neural network, weight parameters are adjusted via back propagation algorithm, and a mapping function between defect free samples and training samples can be obtained. Finally, the mapping function and features dictionary are used to reconstruct image and extract features, according to the Meanshift algorithm to segment the defects and determine the fabric defect position by two value. The experimental results demonstrate that the method based on the deep-convolutional neural network can achieve the purpose ofimproving efficiency, shortening the time of measurement, and obtaining an accurate defect image.