Detection, Based On Computer Vision, The Cotton Crop Growth Indicators

Cotton is the most important commercial crop and the second largest agricultural product in China. The output of cotton directly affects the development of agriculture and textile industry, consequently affects the development of national economy. As China has a vast population and a limited farmland, the plant area of cotton is very limited. The primary approach to increase the output of cotton is to enhance the regulation and control in cotton cultivation, improve the quality of cotton population and increase the yield per unit area, which demands on the knowledge of a series of physiological indices of cottons to establish cultivation technique systems with quantified regulations and controls.Indices closely related to cotton's growth conditions, e.g. leaf area index and irrigation index are not only significant to physiological research, but also the basis of taking measures in production. The measurements of these indices are doing manually at present, which are time- and labor-consuming, subjective and inaccurate. Moreover, some methods are destructive to the subjects investigated, which make periodically monitoring impossible. The computer vision technology is able to improve the measuring greatly.The following work is done in the thesis:(1) A semi-supervised image segmenting method, based on color clustering, is used to segment the cotton images. Using colors of a small amount of pixels selected by user as the centers of clustering, cotton leaves can be selected efficiently..(2) On the basis of the results of image segmenting, color features are used to set up BP neural networks to determinate LAI. The results demonstrate that this approach is practical. For the theoretical incompleteness of this method, a thorough analysis on the test result is discussed and explained all the potential sources of the errors.(3) BP neural network models are used to judge the irrigating condition of cotton similarly. Tests on a group of sample images are made. Another set of results are made on the images segmented by the threshold of the histograms of color. The results are analyzed and compared to the result using threshold segmenting with histograms. The result based on the approach mentioned above indicates that the idea is generally practical. More research work should be done on cotton image collecting and appropriate image processing methods for the application in agricultural production.