A dual oscillatory particle swarm optimization for tomato disease classification

This study presents a novel Dual Oscillatory Particle Swarm Optimization (DOPSO) algorithm for automated tomato disease classification through image analysis, advancing agricultural sensing and computer vision methodologies. DOPSO uniquely introduces anti-phase oscillatory operators that alternate between exploration and exploitation phases, coupled with a dual-subgroup architecture enabling parallel search with information exchange, addressing premature convergence in hyperparameter optimization for image classification tasks. Using a dataset comprising 10,000 images (1,000 per class across 1 healthy state and 9 disease types), we systematically extracted ten discriminative visual features spanning color, texture, and shape characteristics from tomato leaf images. DOPSO achieves superior image classification performance by optimizing light gradient boosting machine (LightGBM) algorithm hyperparameters. Experimental validation demonstrates that DOPSO-LightGBM achieves 82.25% classification accuracy, representing improvements of 6.25 percentage points over support vector machine (SVM) (76.00%), 21.20 percentage points over Logistic Regression (61.05%), 1.90 percentage points over Random Forest (80.35%). Robust performance across precision (82.21%), recall (82.25%), and F1-score (82.16%) metrics establishes reliable tomato disease image classification. This research contributes significantly to automated image analysis for crop disease detection, demonstrating effective integration of optimization algorithms with computer vision techniques for agricultural applications.