Web-Based Application for Medicinal Plant Identification Using Transfer Learning Approach

This research addresses crucial challenges in identifying medicinal plants in India, essential for Ayurvedic Pharmaceutics. It proposes a pioneering approach using image processing, machine learning and a flask-based web solution allowing users to upload and analyze their own photos seamlessly. To conduct this study, a dataset from Kaggle featuring 40 medicinal plant classes is leveraged. Convolutional Neural Networks (CNNs) and transfer learning with DenseNet121 are employed for formulating deep learning model and class prediction. The CNN model achieves an accuracy of 69.58%, and transfer learning with DenseNet121 achieves 77.20%, both demonstrating high precision, recall, and F1-score. While acknowledging limitations in dataset scalability and potential biases, the study recommends future endeavors to focus on dataset expansion, real-time applications, integration of environmental data, continuous model optimization, validation, certification processes, educational initiatives, and global collaboration for conservation efforts. This research represents a significant advancement in ensuring the integrity of raw materials and building trust in traditional medicinal practices within the Ayurvedic Pharmaceutics domain. Such strides contribute to the broader landscape of holistic healthcare by combining traditional wisdom with cutting-edge technologies.