IDENTIFICATION FOR ANIMAL FIBERS WITH ARTIFICIAL NEURAL NETWORK

Scale pattern of animal fibers is different and that is a major reference distinguishing them from each other. Usually, there are four basic shape parameters, including fiber diameters, scale interval, scale perimeter and scale area, to be used for describing the cuticle scale pattern of animal fiber. In present paper, two kinds of animal fiber are checked up under light microscope with a magnification of 40xfor objective and their images are captured by a CCD camera fixed on the microscope. After using a series of image operators on them, the skeletonized binary images only having one pixel wide can be obtained. Then, these basic shape parameters of scale are measured and the database composed of numerical data of four comparable indexes, including fiber diameter, scale interval, normalized scale perimeter and normalized scale area, are established. Finally, a multi-parameter neural network classifier, including four input nodes, five hidden nodes and two output nodes, are developed to classify the two kinds of animal fibers. Two sets of classification rules are applied to the classifier respectively and the simulation results show that whether rule 1 or 2, the neural network classifier can always distinguish cashmere from fine wool (70s) effectively and the average classification performance is higher than 90 percent.