Image Recognition Of Various Fibers And Quantitative Analysis For Blended Fabric

Digital image processing technology has been widely applied in industry measurement and other fields. To replace time-consuming manual detections in textile industry, applying digital image processing techniques and developing automated testing equipment can obviously enhance the detection accuracy and efficiency. The image processing methods to measure warp/weft densities and blending ratio in fabrics are studied in the thesis. The main contributions are as follows:1. To measure the warp and weft densities in fabrics, a rapid and reliable method based on image recognition is proposed. First of all, several warp and weft RGB signals are extracted from the color image of fabric, and the RGB signals are converted into gray signals whose backgrounds are removed. Secondly, the gray signal curves are divided into several groups, and the curves in the same group are linked to the end of the previous curve one by one. Finally, the linked signal curves are converted into frequency signals by Fourier transforms. The amplitude-frequency curves with the highest SNR (signal to noise ratio) are used to calculate the curve cycle. Warp/weft density is then computed with the curve cycle and the organization structure of fabric. The experiment result shows that the method is rapid, accurate and reliable.2. For cotton-flax blended fabric, a new automatic fiber identification method based on the fiber microscopic images in the longitudinal view is proposed. This method includes the following four parts:image preprocessing, boundary contour extraction of fiber, feature extraction and selection, recognition of fibers. The image preprocessing is composed of median filtering denoising, image background correction and image binarization. Fiber contours are obtained by combining morphological close computing and background regional growth algorithm. Then the regional image and binary image of single fiber are obtained as well. 3. The regional image, binary image and its thin image in the vertical direction of fiber skeleton are obtained. Their vertical integral projections are computed, and then the coefficient variation (CV) and mean for each projection are calculated. Such six statistical parameters are analyzed and five of them are selected to be as characteristic variables to distinguish cotton and flax fibers. The above five characteristic variables are used to construct the BP neural network classifier and the least square support vector machine classifier. The experiment results show that the least square support vector machine classifier performs better in the recognition of cotton/flax fiber, and its mean identification accuracy is 93.6%.4. Based on the proposed fiber recognition algorithm, the initial detection software for blending ratio with good interface is developed by using MATLAB and VC++ mixing programming technique.