Abstract:

In order to identify cashmere and wool rapedly and accurately, a method based on bag-of-wisual-word was proposed. Optical microscope images of cashmere and wool were taken as experimental specimen in this method. The problem of fiber identification was changed to problem of image classification. Firstly, fiber images were pre-processed to enhance their characteristics. Then, local features were extracted from fiber morphology and these local features were converted to visual words. Fiber images can be classified using visual words mentioned above. The experimental dataset contains 4 400 fiber images. Different mixing ratio of cashmere and wool were selected as train set and test set from the dataset. In this experiment, the highest recognition rate is 86%, and the lowest is 81.5%. The time required to identify 1 000 fibers is shorter than 100 s. The trained classifier can be saved and used for the later detection.

Key words: cashmere, wool, bag-of-vosual-word, image processing, rapid identification