Cotton Disease Based On Machine Vision Technology To Identify

In this paper, a cotton disease identification system is established based on machine vision; the significance and the practical application value of this study is also illustrated. The main contents are as follows:(1) Cotton cultivation. First, cultivate seedling both in pit plate and in nutrition plate plug. After some ten days, cotton seedlings are transplanted in soil. In order to get diseased cotton samples, infection testing by artificial inducement has been done.(2) Image acquisition and segmentation. Cotton leaf images are taken both in-doors and out-doors. Median filtering is performed to remove noises in images after acquisition, and then images are binarizied to segment the health part and the diseased part of leaf. Three segmentation methods are compared to get a better result. And the optimal EXCESS-GREEN index with the optimal threshold claim the best segmentation result.(3) Morphological features extraction. Image processing based on mathematical morphology is applied to extract morphological features that can distinguish the diseased leaf images and the healthy ones. The total six features are: the ratio of hole number to leaf area; the ratio of hole area to leaf area; the ratio of erosion times to leaf area; the complexity, i.e. the ratio of the square of boundary circle to leaf area; and the ratio of thinning length to leaf area as well as the ratio of skeleton length to leaf area. Finally, KNN method is used to choose and optimize features.(4) Pattern recognition system establishment. Newly developed pattern recognition method-support vector machine (SVM) is applied here to improve correct recognition rate. Four different common kernel functions are chosen to establish the model, and the grid search method is used to optimize the kernel parameters and the punishment parameter C. It shows that model with radial basis function (rbf) claims the maximum correct recognition rate, which is 98.333%.