Abstract
The overall objective of this research is to develop a supporting system for web based diagnosis of rice diseases using the automatic analysis of image features. In order to establish an online discrimination method, the relationship between the discrimination conditions and the accuracy of 6 types of pattern discrimination analysis methods (Support Vector Machine (SVM), Ensemble Learning, Tree-Based Model, Neural Network, Linear Discriminant Analysis, and Quadratic Discriminant Analysis) was examined with 4 classes of 3 rice diseases (Leaf blast, Sheath blight, and Brown spot) based on the shape and color features of the disease lesions. In the case when 7 variables were used, the accuracy results of each analysis method in the discrimination of 4 classes at once were 86%, 81%, 76%, 80%, 78% and 81%, respectively. Furthermore, the accuracy of SVM was 94% in pair-wise discrimination. Hence, SVM was considered to have good potential for discriminating rice diseases.
