Animal Fiber Classification Based on Fuzzy Neural Network

Scale pattern of cashmere and fine wool is different and that is a major reference distinguishing them from each other. A technique commonly used consists of analyzing their SEM image for detecting cuticle scale edge height (CSH) of fiber. However, the method is expensive and only a small quantity of fiber samples is detected, which make the average error approach 8 percent. In this paper, a new method is presented. The color light microscope images of fiber captured by CCD camera are transformed into skelrtonzied binary images only having one pixel wide and showing only fiber and scale edge details. Then four basic shape parameters of fiber scale are measured and a database composed of numerical data of four comparable indexes, which are fiber diameter, scale interval, normalized scale perimeter and normalized scale area, is established. A FNN classifier is developed based on them. The simulation results show that whether on training set or testing set, the model can always distinguish cashmere from fine wool (70s) effectively and the average classification accuracy are higher than 90 percent.