Optimizing Rice Plant Diseases Recognition in Image Processing and Decision Tree Based Model

Abstract

The objective of this paper is to design rice plant diseases recognition system and optimize the recognition efficiency of the system for new test datasets. In this research, the images from rice plant field had been captured by Charged Couple Device digital camera in Joint Photographic Experts Group format in day lighting. The total 6 categories of images with 5 categories of disease infected and one category of non-infected images had been captured. These acquired images had been pre-processed and segmented using three-level of the threshold to extract hybrid features which are a combination of color, texture and discrete cosine coefficient. The hybrid features of each image represent unique feature pattern of individual categories. The inverse multi-quadrics radial basis function had been applied on extracted hybrid features to make features localized and non-singular to enhance the uniqueness of the feature patterns. These transformed features had been used to design rice plant diseases recognition system using a decision tree. The uses of radial function drastically optimize the average recognition efficiency of diseased and non-diseased rice plant from 16.67% to 83.34%. This method can be generalized to design a monitoring system for plant diseases to help farmers and government agencies for on-location inspection and assessment of severity of diseases and take precautionary measure to control the spread of diseases.