FloraMediX: AI-enabled recognition and therapeutic profiling analysis of indian medicinal flora.

Introduction: Accurate identification and profiling of medicinal plants are essential for integrating traditional herbal medicine into mainstream healthcare. However, plant identification remains challenging due to morphological similarities among species and the declining availability of taxonomic expertise. Methods: We developed FloraMediX, an AI-enabled platform that combines deep learning-based visual recognition with structured pharmacological knowledge retrieval. The system employs a Swin Transformer Base (Swin-B) architecture for medicinal plant classification, a human-in-the-loop curation pipeline for iterative model refinement, and integration with phytochemical, therapeutic, and geographical databases including IMPPAT and OSADHI. The model was trained on 166,684 images representing 75 Indian medicinal plant species from the PlantNet-300K dataset and evaluated on an independent test set of 20,880 images. Results: The Swin-B model achieved a Top-1 accuracy of 93.70%, Top-5 accuracy of 99.68%, Macro F1-score of 91.38%, and an AUC of 0.9987. Performance exceeded baseline architectures, including ResNet-CBAM (90.57%), ConvNeXt (92.22%), and CoAtNet (92.22%) on the same evaluation dataset. Discussion: FloraMediX provides a reliable and accurate framework for medicinal plant recognition while simultaneously delivering curated phytochemical, therapeutic, and geographical information. The integration of advanced deep learning, expert-guided curation, and knowledge databases supports applications in research, education, biodiversity conservation, and the preservation of India's traditional medicinal plant knowledge.