Segmentation of field cotton image based on competitive learning network

In order to distinguish field cotton image exactly, cotton and its nature background, including bracteole, leaf, rhizome and land, were classified into two categories and segmented based on competitive learning network. 10000 pixels with two categories extracted from some typical cotton images were regarded as training data, their color components and mixing of components were classified into two categories based on K-means clustering in HSI, Lab, Ohta, RGB color space, and error rates of color components and their mixtures are lower in RGB color space, particularly blue component. Competitive learning networks were trained only onec with one input value of blue component or three inputs of red, green and blue components of training data in RGB color space and all pixels of an image as test data were input into them, comparing the results of competitive learning networks with K-means clustering after morphological filtering show that competitive learning network with one input value of blue component was optimized, 907 cotton images were segmented with an accuracy of 92.94%. Combining supervisory learning arithmetic, avoiding iterative and over-fitting of K-means clustering, competitive learning network has good performance and high efficiency in image segmentation.