Research Of Classification System Of Cashmere And Wool Based On Digital Image Processing

The research of this paper is based on the 《Research of Parameter Detection of Cashmere Fiber in China-Arabia Cashmere Cooperation Industry)) of China-Arabia university cooperation research program. The cashmere and wool fiber of Lingwu market were selected randomly as the research object. 1500 sample images of the two kinds of fiber were collected respectively with the digital image collecting system. Process the fiber image with the technology of digital image processing.8 characteristic parameters and 4 combined parameters who can indicate the characteristics of the fiber were extracted. Then did statistic analysis to each parameters emphatically. Finally we constructed an identification model of the two different fibers based on Bayesian theory. The constructed model can realize auto identification of cashmere and wool with high accuracy. The main research contents and achievements are listed as follows:(1) Research the best collecting condition for cashmere and wool images. The optical microscope needed to be reflex lighting way. The objective lens should be 20 times with big depth of field and long working distance. The sample extraction should be in accordance with ISO 17751:2007 strictly.(2) Research and confirm the best preprocessing method for fiber image. The processing order was graying, edge enhancement, denoising, edge extraction and finally the morphology processing. The best processing method in each stage was also researched. Select weighted average way as the grayscale method. Use Laplace filter to enhance the edges. Use median filtering to denoise. Set 128 as the binaryzation threshold. Choose dilation and erosion to do the morphology processing.(3) Select 8 key characteristic parameters (the fiber diameter, the scale height, the scale perimeter, the scale area, the relative scale perimeter, the relative scale area, the diameter to height ratio, the squareness) and 4 combined parameters ((dh)2,h2-Ad, (dh)2 / Ad, (dh)2/(Ad·Pd))of the two kinds of fibers. In the process of parameter measurement, we designed an areal survey method of diameter measurement. The designed method can function well on curve fibers. Its measurement accuracy is much higher than the traditional method of triangle and axle wire.(4) Do an intensively statistic analysis to distribution characteristics, two kinds of error probability and correlation of the 8 characteristic parameters and 4 combined parameters. The results show that every parameter approximatively follows logarithmic normal distribution. According to the parameters’ critical value, the two error probability value and the overlapping area, we can test and verify the 8 characteristic parameters’accuracy. The sequence decreasingly was the relative scale area, the diameter to height ratio, the relative scale perimeter, the fiber diameter, the scale height, the squareness, the scale area, the scale perimeter. When it came to the 4 combined parameters, the sequence decreasingly was h2 · Ad,(dh)2,(dh)2/ Ad,(dh)2/(Ad-Pd). The relationship among all the parameters can be linear correlation, linear independence or nonlinear correlation.(5) According to the parameters’correlation and accuracy, Bayesian classification model based on 4 combined parameters was constructed. The analysis to 4 identification model results shows that all the parameters’ number, the parameter’s accuracy and the correlation among parameters can affect the accuracy of the identification model. The situation is the identification accuracy can be much higher as long as the parameters’ number is larger, the parameters’ accuracy is higher and the coefficient of association among different parameters is smaller. On the basis of the above rule, the best Bayesian identification model with 9 combined parameters was finally constructed. The experiment results show that the identification accuracy can reach 98.9%. The computer auto identification of cashmere and wool with high accuracy is realized successfully.