Ayurvedic Plant Identification using Inception V3 and VGG16 Model

Across India, numerous pastoral communities hold a solid conviction in floral diversity. They gather beneficial stores and condiments and employ them alongside traditional wisdom and rituals. Another comparable ancient approach that leads to the application of therapeutic condiments is Ayurveda. Although almost 10,000 plants are used medicinally in India, not all of them are included in the official Ayurvedic Pharmacopoeia. Every plant must be carefully inspected before being formally used in Ayurvedic medicine. For this purpose, identifying condiments remains the crucial initial phase. Many of these classifications are wholly reliant on human senses, allowing space for mistakes and misinterpretations. Therefore, it becomes essential to create an efficient system utilizing computer vision, pattern recognition, and image processing methods together with the availability of various feature detection techniques alongside diverse classifiers commonly applied in building an automated recognition framework for herbal leaves by employing leaf images and conveying its relevant details to acquire insights and wisdom. Two deep neural network models, namely VGG-16 and InceptionV3 are applied to study the performance of the algorithms for classification of Indian plant leaf data into Two Hundred different classes. To handle the limited dataset, we used the transfer learning technique with a fine-tuning method, optimizing the model output. Our study showed a classification accuracy of 85.00% with VGG-16 and 87.50% with InceptionV3. Performance evaluation has been carried out using confusion matrix, recall, precision, F1 score. Our analysis showed that fine-tuned InceptionV3 architecture outperforms the other model for image classification task.