Abstract:The type and content of the impurities in machineharvested cotton are important parts of the cotton parameters, and they determine the adjustment of the processing technique of cotton. A method based on color and shape features for recognition of machineharvested cotton impurities was presented. Different image processing methods were adopted for large impurities and small impurities, and the detailed algorithm flow chart was formulated. The window filtering, image segmentation, color feature statistics and shape feature extraction were adopted to process image. For the large impurities image, smooth filtering, clustering segmentation, binarization, hole filling were conducted sequentially, and then, shape features such as area, perimeter, eccentricity, rectangle degree and color pixel statistics of the target region were calculated. The impurities which include branches, boll shell, stiff flap and leaf were identified by using combination of color and shape features. For the small impurities image, large and yellow impurities were removed after image sharpening and clustering segmentation, and the area was calculated through color pixel statistics. In order to speed up the calculation and improve the recognition rate, watershed algorithm based on color gradient image and improved fuzzy Cmeans clustering algorithm with specified initial cluster centers were combined to split image. As a result, the recognition and classification of machineharvested cotton impurities can increase the efficiency of cotton processing equipment, reduce damage of cotton fiber, and provide improved guidance for cotton harvest equipment. For the investigated 100 sample images including five types of cotton impurities, a 89% successful recognition rate was achieved.
