Identification of Dry Ayurvedic Herbs (Fruits and Seeds) through Computer Vision Technology

Abstract

The study investigates the use of Computer Vision Technology (CVT) combined with Convolutional Neural Networks (CNNs) to address challenges in the identification of dry Ayurvedic herbs (fruits and seeds). A dataset of 50,000 high-resolution images, encompassing 50 herb species, was utilized to train a CNN model. The architecture comprised convolutional layers with filters and Dropout layers, ensuring efficient feature extraction and overfitting mitigation. The model achieved a peak training accuracy of 91.86% and a validation accuracy ranging from 81% to 83%, with an inference time of 36 milliseconds per step, indicating its potential. Performance evaluations, including accuracy metrics and confusion matrices, highlighted high prediction rates for distinct species. However, misclassifications among visually similar herbs underscored the need for dataset expansion and further optimization. Recommendations include incorporating robust database, additional species, diverse angles, and lighting conditions, as well as addressing class imbalances through data augmentation or resampling. Advanced regularization techniques, are proposed to enhance generalization. This research work Bridges the Traditional identification methods and Modern methods with Technology and establishes a robust framework for leveraging AI and computer vision in Ayurvedic herb identification, contributing significantly to the modernization and quality assurance of traditional herbal medicine. The findings emphasize scalability and future integration of cloud-based systems for large-scale applications.