基于卷积神经网络的织物表面缺陷分类方法

针对织物缺陷检测时疵点种类繁多且传统人工检测方法漏检率高的问题，提出了一种基于卷积神经网络的织物表面缺陷分类方法。因卷积神经网络（CNN）训练时参数多、样本量大，且极易陷入过拟合，利用微调卷积神经网络模型Alexnet对织物疵点图像进行特征提取，初始化采用原网络的参数而非随机初始化参数；再针对特定目标下的训练样本对网络参数进行微调；最后利用softmax回归算法进行预测分类。分别用三种方法和两种织物进行测试，结果表明：针对特定目标微调后的Alexnet网络，在两类织物测试中均能达到95%以上的分类准确率。

Classification Method of Fabric Surface Defects Based on Convolution Neural Network

A new method for classification of fabric surface defects based on convolution neural networks is proposed,which is mainly used to solve a wide range of defects and the high error rate of manual detection method.Convolution neural networks need to train a large number of parameters,the training process requires a large number of samples,and it is easy to fall into over-fitting,so the fine-tuning Alexnet network model is adopted for feature extraction.The original network parameters were used to initialize the network,and then fine-tune the parameters for a specific sample.Finally,the softmax algorithm was used for prediction and classification.The experiment was carried out using three methods and tested with two types of fabrics.The results show that the fine-tuned Alexnet can achieve more than 95% classification accuracy in both types of fabric tests.