Ayurvedic Herb Classification using Transfer Learning based CNNs

For approximately over 4000 years, medicinal herbs are being utilized in traditional medicine. Ayurvedic herbs have gained popularity in medical industry due to their fewer side effects and lower costs compared to modern medicine. However, it's difficult for agronomists and pharmaceutical companies in the biomedical domain to identify the suitable herb leading to the loss of age's old Ayurvedic knowledge. This paper aims at implementing a system that examines an Ayurvedic leaf using Transfer Learning based Convolution Neural Networks (CNN). Ayurvedic Herb recognition is based on their images. The experiment considers 1835 images of Herbs across thirty species of healthy medicinal plants. The success rates for Herb identification were found to be as high as 98.97% in testing accuracy for DenseNet121 Convolution Neural Network. VGG16, MobileNetV2 and InceptionV3 models reported a test accuracy of 93.79%, 96.21% and 97.59% respectively. Additionally, the future research will be further focused on dataset expansion to benefit agronomists and stakeholders thus enriching about knowledge of Ayurvedic Herbs.