Robust GOA-Driven CNN Parameter Optimization for Effective Rice Leaf Disease Classification

Abstract

Precision agriculture has emerged as a solution to global challenges, including population growth, declining fertile land, and climate change. As a result, accurate nutrient management in crops is becoming increasingly critical. However, this task is challenging due to genetic diversity among plant populations and fluctuating environmental conditions. Because plant health directly affects nutrient requirements, accurate classification of leaf diseases is essential. This study aims to improve the classification performance of rice leaf diseases by optimizing the parameters of a Convolutional Neural Network (CNN) using various optimization algorithms. The dataset comprises publicly available images of rice leaf diseases, and a CNN is employed for its strength in visual pattern recognition. To enhance CNN performance, several metaheuristic optimization algorithms, including the Grasshopper Optimization Algorithm (GOA) and the Salp Swarm Algorithm (SSA), are applied, along with other comparative methods. These algorithms are used to fine-tune parameters such as the number of epochs, dropout rate, number of hidden neurons, learning rate, activation function, and optimizer. The results show that GOA achieved the highest accuracy of 72.88 percent, due to its optimal configuration of 8 epochs, a 0.1 dropout rate, 512 neurons, a learning rate of 0.001, the ReLU activation function, and the Adam optimizer. SSA demonstrated strong generalization capability with faster training, while GTO offered a balanced trade-off between accuracy and computational cost. These findings suggest that optimization algorithms have considerable potential to enhance plant disease classification systems. Future research should consider hybrid optimization strategies and larger datasets to improve robustness and practical applicability in field settings.