Recognition And Classification Of Fabric Defects Based On Wavelet Analysis And Neural Network

Recognition and detection of fabric defects is an important step in the productionprocess of textiles. According to the defects and shortcomings of artificial works, thispaper based on the existing fabric defect detection theory and method, deeply studiedfabric image preprocessing and automatic recognition method based on wavelet analysisof flaws. And the feasibility and validity of this method is verified by experiment.Firstly, in image processing commonly used pre processing method, analyses therespective characteristics, and through the experiment, compared to a variety of preprocessing results, choose suitable pretreatment methods, improve the visual effect,highlights the flaw of images and the edge contour feature.Secondly, the automatic defects recognition method is proposed based on waveletanalysis. In order to reduce the environment impact on defect identification and at thesame time to improve the speed of defect detection, wavelet transform is used in thesegmentation of images. After wavelet decomposition, comparing through gray mean ofimages, abandon over the flawless windows of that threshold is no more than set. Thosewhose threshold over set are as a further inspection area, extracted the characteristicvalues in wavelet analysis method, effectively reduce the image characteristic values ofthe extraction and calculation of the number of windows. After wavelet decomposition,five feature values energy, entropy, variance, range and deficit moment wereextracted from the latitude and longitude sub-images. After normalization, theresponsiveness of the different feature values is checked in a unified metric examined,to confirm whether there is a defect and the defect judgment accurate position.Finally, analysis of the structure characteristics, design method and parameters ofthe method to determine of a neural network, a3layer BP neural network is designedaccording to the characteristic of the fabric flaws. Inputting the feature values to the BPneural network for automatic detection, through the experimental comparison, verify the feasibility and effectiveness of fabric defect recognition and classification method inthis paper.