Based On Machine Vision Automatic Test Machine System Research

China’s textile products in the international textile fabric market occupy very highshare, but most of them are some middle and low grade products, so fabric surface defectinspection in recent years to become a very important link. And the traditional test is doneby the human eye, this method reduces automation degree in the textile production process,make inspection efficiency reduced greatly, it is easy to appear a phenomenon of residualand error detection, and by the individual factors, subjective experience influence is larger,so in order to speed up the development of China’s textile industry, improve the efficiencyof the textile enterprises, it is more important that carry out high efficiency and highquality automatic test machine system research. This paper mainly from two aspects ofresearch that the defect feature extraction and recognition classification, and the specificwork is as follows:This paper adopt a kind of improved Grey-Level Co-occurence Matrix to fabric defecton the extraction of characteristic value, first find out symbiotic matrix of four directions,and then find respectively five characteristic parameter values that each co-occurrencematrix, according to the comprehensive gray level co-occurrence matrix formula, first findout the weighting coefficient of each direction, and then find out five comprehensive graylevel co-occurrence matrix, and finally find out five characteristic value, this kind ofmethod overcomes the problem that the single direction symbiosis matrix can’t completedescribe the texture characteristics, makes the final error rate is reduced, shortens theclassification time for later defect classification work, reduces the calculated amount, butalso improves the classification speed and classification accuracy in the meantime.In the defect classification, this paper studies the radial basis function (RBF) neuralnetwork and learning vector quantization (LVQ) network, and compares their classificationeffect, analyzes the different characteristics parameters of the RBF neural network,different training samples groups, different training times for classification effect, the RBFneural network has faster classification speed and higher classification accuracy, so moresuitable to be applied to fabric defect classification.In addition, it also studies the influence of the fast principal component analysis method to classification effect, contrast the characteristic value that adopt the fast PCAmethod and not, characteristic value dimension and amount are reduced significantly afterimprovement, but also eliminate the redundancy and some correlation between the features,accelerate the classification speed, make error rate reduced greatly.