A DEEP LEARNING MODEL FOR IMPROVING THE RICE PLANT DISEASE DETECTION PERFORMANCE

Abstract

Rice is one of the most utilized grains in India. It is a seasonal crop which mostly grows between June to October. This crop mostly grows in natural conditions and its production has a significant influence on different diseases in the plant. Early stage detection of diseases can help in improving the production. In this paper, an analysis and study on deep learning models for getting accurate rice plant disease detection is presented. In this context, first the recent contributions on detecting the diseases by analysing the plant leaf images are reviewed. Then, a comparison among sequential model and 2D-CNN model has been performed. The experimental analysis demonstrates that 2D-CNN outperforms as compared to the simple sequential model. The experiments are extended by including the different image feature selection models. In order to extract features, sobel based edge detection, Local Binary Pattern (LBP) based texture analysis and their combinations i.e. sobel and LBP, Sobel, LBP and color, and a combination of color and sobel are used. The experiments are performed on Kaggle based rice plant disease detection dataset and the performance in terms of precision, recall, f1-score and accuracy has been measured. The experimental evaluation highlights two major points (1) the CNN does not require additional features for better classification consequences (2) the highly trained models are able to respond faster as compared to less trained models. Based on the obtained performance, a more accurate model for plant disease detection is designed.

Keywords

• plant disease detection
• machine learning
• image processing
• food security