Published Paper Details:

This paper presents a comprehensive deep learning framework for automated identification and classification of South Indian medicinal plants, with particular emphasis on species exhibiting closely similar leaf morphologies. The increasing loss of traditional botanical expertise combined with growing demands for reliable, scalable identification technologies in ethnomedicine, biodiversity conservation, and phytopharmaceuticals motivates this research. We developed a multi-source image dataset encompassing six vital medicinal plant species--Alfalfa, Aloevera, Fenugreek, Kadamba, Neem, and Papaya--systematically organized by organ type (leaves, flowers, fruits). The proposed framework implements both single-part and multi-part classification workflows using ResNet-34 architecture, optimized through transfer learning and validated using contemporary best practices. Results demonstrate marked improvement in classification performance when organ-level images are combined using ensemble voting techniques, achieving weighted average accuracy of 84% and F1-score of 0.88. The inclusion of reproductive organs (flowers, fruits) alongside vegetative parts effectively resolves ambiguities in morphologically similar species, establishing a robust, scalable, and interpretable workflow suitable for real-world ethnobotanical and healthcare applications.