Ensemble CNN Model for Detecting Rice Leaf Diseases by Integrating MobileNetV2 and EfficientNet-B1 Architectures

Abstract

Rice diseases significantly impact agricultural productivity, requiring effective methods for early detection to prevent substantial damage. This paper introduces an ensemble Convolutional Neural Network (CNN) model that integrates MobileNetV2 and EfficientNet-B1 architectures for the identification of rice leaf diseases. The algorithm employs a dataset comprising 5,936 rice leaf photos from Mendeley Data to classify five disease categories: Bacterial blight, Blast, Brownspot, Tungro, and Healthy. The ensemble method employs hyperparameter optimization by Bayesian and Random techniques, integrating dropout, learning rate adjustment, and kernel regularization to achieve a compromise between model accuracy and generalization. The model attained a training accuracy of 99.99% and a validation accuracy of 99.90%, showcasing robustness and accuracy. The TensorFlow Lite format facilitates integration into mobile applications, providing farmers with a portable and effective solution. This system enhances agriculture by equipping stakeholders with advanced disease management tools, increasing yields, and promoting sustainable practices.