Deep learning based approaches for automatic rice leaf disease classification

Abstract

Rice is a fundamental food source for more than half of the global population. Rice newlineis considered the third most widely cultivated crop, making its health and productivity newlinecritical to global food security. However, rice crops are vulnerable to various diseases newlinethat can significantly impact both yield and quality. Conventional methods of disease newlineclassification, which are highly dependent on visual inspections, are often time consuming, newlinelabor intensive, subjective, and error-prone. This underscores the need for newlineautomated leaf disease classification methods. Recent advances in image processing, newlinemachine learning, and deep learning have been explored to address this need. newlineAmong these methods, deep learning has proven effective by automatically learning newlinehierarchical and discriminative features from raw leaf images, eliminating the need for newlinemanual feature extraction and reducing subjectivity in disease identification. However, newlineeffective training of deep learning models requires substantial amounts of data. More newlineextensive datasets enable better generalization and performance on unseen samples. newlineHowever, current research in automatic rice leaf disease classification using deep learning newlinefaces several limitations, including the limited number of disease classes, a small newlinenumber of images per class, and the use of limited pre-trained models. These limitations newlineresult in biased results, poor model generalization, overfitting, and high error rates, rendering newlinethese models unreliable in real-world applications. To address these limitations, newlinethis research investigates the automatic classification of rice leaf diseases using deep newlinelearning, particularly through transfer learning and an ensembling approach, to enhance newlinelearning capabilities, computational efficiency, and model performance. newlineThe first objective of this research is to improve the classification of rice leaf diseases newlineby training and fine-tuning pre-trained deep learning models using data augmentation newlineand transfer learning techniques. We achieved this by collecting and integ