A Novel Hybrid Severity Prediction Model for Blast Paddy Disease Using Machine Learning

Abstract

Hypothesis: Due to the increase in the losses in paddy yield as a result of various paddy diseases, researchers are working tirelessly for a technological solution to assist farmers in making decisions about disease severity and potential danger to the crop. Early prediction of infection severity would facilitate resources for the treatment of the infection and prevent contamination to the whole field. Methodology: In this study, a hybrid prediction model was developed to predict various levels of severity of blast disease based on diseased plant images. The proposed model is a four-fold severity prediction model. The level of severity is defined based on the percentage of leaf area affected by the disease. The image dataset is derived from both primary and secondary resources. Tools: The features are first extracted with the help of the Convolutional Neural Network (CNN) approach. Then the identification and classification of the severity level of blast disease are conducted using a Support Vector Machine (SVM). Conclusion: Mendeley, Kaggle, GitHub, and UCI are the secondary resources used for dataset generation. The number of images in the dataset is 1908. The proposed hybrid model achieves 97% accuracy.