RATINGS OF RICE LEAF BLAST DISEASE BASED ON IMAGE PROCESSING AND STEPWISE REGRESSION

In this research, an evaluation method involving digital image processing and stepwise regression was studied to establish an efficient and accurate rating system for studying rice blast disease. For this purpose, the R-G image was segmented by using maximum interclass variance method in which the lesion and naturally withered region was extracted from the leaves. Then, 240 lesion areas and 240 natural yellow areas were selected as samples. During the experiment, ten morphological features and five texture features were extracted. Subsequently, for lesion identification, stepwise regression analysis, SVM and BP neural network were used. In the results, regression analysis of naturally yellow areas showed the highest accuracy in lesion identification, reaching 93.33% for disaster-level assessment of identified lesion areas. On the basis of the results, it is evident that 153 samples were correctly classified into divisions of 160 tested different rice blast leaves, with 95.63% classification accuracy. This study has introduced a new method for objective assessment of leaf blast disease.