Image recognition of green weeds in cotton fields based on color feature

In order to realize automation of herbicide spraying precisely in cotton fields, the research on automatic recognition green weeds from cotton fields was developed based on color feature. The cotton seedling stem’s dark red feature was considered mainly. Firstly, the gray images of excess red feature and excess green feature were processed by Otsu’s threshold method, and cotton seedling stem’s and green plants’ binary images were gained respectively. Secondly, cotton seedling stem’s coordinates were extracted from its binary images, and the location information fusion was done between cotton seedling stem images and green plants’ binary images, then cotton seedling were obtained from green plants’ binary images. Finally, all weeds’ regions might be recognized and its image features, regions’ centroids and regions’ areas, were also calculated. The tests of 15 mixed images between cotton seedlings and green weeds showed that the green weeds could be recognized completely, and the recognition rate of the cotton seedling was 74% in the case that the cotton seedling stems were not blocked by their leaves and there were no overlap between cotton seedlings and green weeds.