Medicinal Plant Recognition Using Particle Swarm Optimized Neural Network Cascade

Abstract

The accurate classification of medicinal plants is vital for ensuring safe and effective use in traditional and modern healthcare systems. However, manual identification by experts often suffers from inconsistencies and errors, potentially endangering human lives. This study proposes a robust and intelligent classification framework that integrates deep learning and optimization techniques to enhance identification accuracy. A novel cascaded network architecture is developed, leveraging the powerful ResNet50 model for deep feature extraction. These features are then refined using Particle Swarm Optimization (PSO), a metaheuristic technique that effectively selects the most relevant attributes, reducing computational complexity and enhancing classification performance. The optimized feature set is subsequently evaluated using seven machine learning classifiers: Support Vector Machine (SVM), Random Forest, Decision Tree, K-Nearest Neighbors (KNN), XGBoost, Naïve Bayes, and Logistic Regression. Experiments are conducted on a publicly available Kaggle dataset containing images of seven distinct medicinal plant species. The results highlight that Logistic Regression achieves a high classification accuracy of 99.18%, while fine-tuned SVM outperforms all others with a peak accuracy of 99.75%. This demonstrates the efficacy of the proposed cascaded network in accurately distinguishing plant species. The synergy of ResNet50 for robust feature extraction, PSO for intelligent feature selection, and diverse classifiers provides a comprehensive and scalable solution for medicinal plant classification. This research contributes significantly to the automation of botanical identification, offering practical applications in ethnobotany, pharmacology, and agriculture, thereby reducing human error, and enhancing medicinal plant research.