| The economic value of the cashmere products is very high. It is very necessary that the identification of cashmere and wool fiber for standardizing the market and protecting the legal rights and interests of consumers. Fiber detection technology, which is based on digital image processing and image recognition technology, as it has a higher recognition rate and the detection result is relatively objectively, is widely used in wool cashmere fiber inspection field, and it has already became a popular research direction in this field.In this paper, we choose cashmere 100 fiber images and 100 wool fiber images which are extracted by differential interference contrast microscope, we preprocess every image and extract the features form it. At last, we use the characteristics to classify two kinds of fiber images by classifier.In the stage of image preprocessing, we mainly use of digital image processing techniques to process the fiber images to simplify the process and improve the efficiency of feature extraction.In the stage of feature extraction, we extract 16 fiber characteristics including 9 fiber structure characteristics and 4 fiber statistical characteristics, take the fiber speckle parameters as one of the fiber characteristics, and discuss the affine invariance of fiber characteristics. At last, we model the rake angle and thickness of fiber scale and present a new corner detection algorithm, based on the new algorithm, we realize the extraction of the two kinds of characteristics above.In the stage of image recognition, through the comparison of three kinds of classifiers, we choose the support vector machine(SVM) as the final classifier to recognize and classify the feature we extract from the fiber images. We get the recognition rate of the single fiber characteristic from the 16 fiber features by calculating the average of the recognition rate through the cross validation method. Then we take the characteristics as an eigenvector and realize the feature dimension reduction by using principal component analysis(PCA). Through the cross validation, the comprehensive recognition rate is 90.25%. |
