Abstract: Scale and pattern of cashmere and fine wool are different, which is used as a major reference to distinguish them. A commonly used technique is to analyze cuticle scale edge height (CSH) of fiber from SEM images. However,
it is expensive and has an average error of 8%. A new method is presented in this paper. After the fiber images are captured with a CCD camera, they are transformed into skeletonzied binary images which are only one pixel
wide and can show fiber and scale edge details. Four relative shape parameters of the fiber scale are extracted. A multi-parameter Bayes classification model is then developed. Numerical experiment results show that, by using an ordinary microscopy, the proposed Bayes model has the performance similar to that based on a scanning electronic microscopy in differentiating cashmere and fine wool (70 S), with accuracy rate approaching 90%.

Key words: cashmere, relative shape parameter, scale pattern, Bayes classification model