Research Of Cashmere Identification Mechanism Based On Image Analysis And Statistical Analysis

Cashmere is a rare and special animal fibers, it is an extremely valuable textile raw materials, the fiber elongated and felt gentle, it known as the "King of Fibers." Because the cashmere have very high quality and price, the market appear some man-made adulterated products, mixed fibers situation is complicated. Therefore, how accurate, effective, rapid and economical identification of cashmere fiber is very important. This paper is an exploration that it is identified the fibers by using a computer. The purpose is to improve the rate of automatic recognition of the fiber image acquired under the optical microscope; first, process the image by using the way of image processing technology, and then statistical analysis by the data, finally calculate the recognition rate.In this paper, at first, the cashmere and wool fiber's structure is compared and analyzed, then, processing the fiber image which obtained by Germany Zeiss (Zeiss) Axioskop2 mot plus differential interference contrast (DIC) microscope image acquisition system, by using image processing techniques. Firstly, sharpening the image by using Laplacian, and then turn the image into a gray image; the gray image that it just obtained is used the median filter to remove noise, and then it does edge detection by using the canny, the final adoption of the expansion and corrosion in morphology modified the fiber image, refining the edge of the fiber and link it, you can get a more complete image of the edge of the fiber.In this paper, it is using two kinds of methods to identify the six characteristics data that it collected to the 10 kinds of cashmere and wool fiber diameter, the diameter of high ratio, scale height, scale projection width, scale thickness at right angles and scales diameter difference, One is the Bayesian identification: First, it carried out the distribution fitting test, and then uses Bayesian identification method to identify, the results show that, cashmere No. 1, wool No. 1, sheep cashmere No. 1, and stretch wool obey the normal distribution in the characteristics of these six parameters. Re-use of Bayesian model identification, to be identify the cashmere No. 1 and wool No. 1 is the best, recognition rate is as high as a percentage of 99, identify the wool No. 1 and stretch wool achieve a higher level, the recognition rate achieved a percentage of 96.315; another is the support vector machine method: using support vector machine training function (svmtrain) for training, the kernel function select the radial basis function, sigma values were obtained 1 to 9, calculation cashmere, wool fiber's overall recognition rate. The results show that, when the training samples is the total number of 480 and sigma values is 9, it gets the lowest classification error rate a percentage of 4.167.