Research Of The Identification Arithmetic Of Fabric Defects Based On Computer Vision

In the textile production, the quality control and the examination are extremely important, and the examination of fabric defects is the most important part. The purpose of the fabric defects identification is to discover them in time in the weaving and examining of the cloth process and reduce the fabric quality drop which causes by the defects as far as possible through the repair and the reorganization. With the development of the computer technology and the imagery processing technology, it is possible to identify the fabric defects based on the computer vision. It not only can enhance the production efficiency of textile but also builds the foundation of objective standard to the fabric defects type, the size and the product quality appraisal formulation.Based on the study and research of the development of the automatic detecting technology, we choose the appropriate feature parameters extraction algorithm and the recognition algorithm and further optimize, and proposed the identification algorithm of fabric defects based on computer vision. We decompose and divide the image through applying the wavelet analysis and the image characteristic, and extract the feature parameters in the subimages by using the statistic method of gray, finally identifying the defects by using the BP neural network. The key content of the method includes the feature parameters selection, the feature parameters extraction and the defects identification. This article mainly revolves these three aspects to conduct the fundamental research.First, we introduce different feature parameters and choose the energy, the entropy, the variance and the difference as the feature parameters of this article. After many times of experiments we know these four feature parameters to be possible the good reflection to treat the recognition of the defaults.Second, we introduce the wavelet analysis. We choose the decomposing wavelet and decomposing method according to the fabric texture characteristic and propose the method of texture cycle through the autocorrelation function, then we define the division window size. Through experiments many times we know the decomposing wavelet and decomposing method we choose can reflect texture characteristic and builds the foundation for the feature parameters full extraction. With the feature parameters extracted from the subimages we know the different defects reflected by the different feature parameters.Third, we introduce the theory of neural network and introduce the structure and the work principle of BP neural network. We identify the fabric defects by using 3 layer BP neural network.The identification algorithm we proposed has the better effect in the identification of tabby fabric defects. Through the experiment we detect 5 kinds of most constant defaults on the tabby fabric: end out, double loop stitch, thread out, double loop pick, bore. The correct identification rate reaches to 97%.