Enhancing Pennisetumglaucum Disease Classification Through Hybrid Optimization Strategies

Abstract

This research endeavours to elevate the efficacy of Pennisetum glaucum Leaf Disease classification within the agricultural domain by integrating Particle Swarm Optimization (PSO) and Ensemble Bayesian Optimization (BO) techniques into the realm of Convolutional Neural Networks (CNNs).

The foundation of our investigation rests upon a fundamental CNN model architecture, serving as the baseline for subsequent enhancements. This swarm intelligence-based technique explores the hyperparameter space, seeking optimal configurations that significantly augment the CNN's predictive performance. Here Bayesian Optimization is incorporated to refine hyperparameter tuning, specifically targeting the number of filters and learning rate. Intriguingly, the five most optimal configurations identified by Bayesian Optimization are amalgamated to form an Ensemble model, harnessing the collective intelligence of diverse CNN architectures. This ensemble model adopts a voting approach to consolidate predictions, aiming for improved robustness and generalization. This paper results not only contribute valuable insights into optimizing CNNs for plant disease classification but also underscore the significance of tailored optimization techniques in enhancing model performance within specific domains.