Hybrid metaheuristic optimization of convolutional neural networks for tomato leaf disease classification

Convolutional Neural Networks (CNNs) are widely used for plant disease detection, yet their performance is strongly influenced by hyperparameter selection. Traditional manual tuning or random search approaches are inefficient and may lead to suboptimal solutions. In this study, we propose and evaluate four hybrid metaheuristic strategies, Ant Lion Optimizer combined with Whale Optimization Algorithm (ALO-WOA), Ant Lion Optimizer with Dragonfly Algorithm (ALO-DA), Ant Lion Optimizer with Particle Swarm Optimization (ALO-PSO), and Particle Swarm Optimization with Whale Optimization Algorithm (PSO-WOA), for automatic hyperparameter tuning of CNNs. The methods were applied to a tomato leaf disease dataset comprising 21,421 training, 4,586 validation, and 4,602 test images across 10 classes. The CNN architecture was fixed with three convolutional blocks and a tunable dropout and learning rate. The experimental results show that ALO-DA achieved the highest test accuracy of 97.83%, closely followed by ALO-WOA (97.67%) and PSO-WOA (97.52%), while ALO-PSO achieved 95.26%. These findings demonstrate that hybrid metaheuristics can effectively improve CNN hyperparameter search compared to single optimizers, balancing exploration and exploitation more efficiently. Limitations and future research directions are discussed.