Computer Image Recognition Method Of Cashmere And Wool

As the excellent properties of cashmere fiber, as well as many similarities of the structure and quality indicators with wool, so the falsified phenomenon that wool are mixed in cashmere are becoming increasingly serious recently, which makes a fast, accurate and effective method to identify cashmere and wool become a top priority, meanwhile,detection of cashmere and wool have been recognized as the most difficult issues in the field of fiber testing.In this paper,the computer image processing method to identify cashmere and wool was studied, which includes: research on the image processing program of cashmere and wool , methods of feature extraction and research on the identification model of BP neural network . In the part of the image processing program of cashmere and wool,this paper initially developed the overall processing program and course by analyzing two kinds of SEM photographs: image de-noising, image enhancement, image segmentation and modification treatment. In order to obtain continuous fiber edge and scale edge information, multiple methods of each processing stage are required to study, so as to find the most suitable approach to the image. Therefore, this paper adapts Matlab7.0 to do experiment, a large number of image processing methods are tested and compared to choose the best method, and finally obtain the image processing programs: transferring to the gray image, using the Sobel operator to the edge reinforcement, using median filter to de-noising, using the watershed to detect the edge, and using of morphology with the expansion and refinement algorithm modification treatment. Eventually, binary images only with edge information are obtained by a series of image processing.In the part of the extraction of the characteristic parameters of cashmere and wool fibers, relative indicators of visual indicators which are used to characterize the cashmere and wool fibers are compared in this article, and the effectively distinguish indicators are chosen, which are: the fiber diameter, scale height and diameter/height. In the measurement, this paper analyze the Hough transform, the least squares methods and their shortcomings. Then, an improved method of fitting a straight line to measure fiber diameter and height scales is put forward and measurement data obtained are analyzed.In the part of the identification phase of cashmere and wool with BP neural network model, which adapts the structure: three neurons in input layer, seven hidden layer neurons, two output layer neurons, and the middle layer neurons and output layer neurons activation functions are S-transfer function. In our experimental conditions, the average recognition rate of cashmere and wool is 92.5% by using this algorithm, which illustrate that the proposed method of cashmere and wool identification is effective.