Leveraging Deep Learning for Medicinal Plant Identification

Medicinal Plants are the backbone of Indian Ayurveda and herbal treatments. However, the correct recognition of medicinal plants is essential for their safe and effective use. Traditional methods rely heavily on the expertise of Vaidya’s (Ayurveda practitioners) and herbalists, whose extensive knowledge base can be a tedious task to acquire and error-prone. Additionally, with the growing popularity of herbal remedies, the demand for proper identification methods is escalating. The proposed model builds a robust model based on deep learning for identification of medicinal plants through image processing. It outlines methods involved in data collection, preprocessing, feature extraction, selection, training, and evaluation. The model targets to classify 30 species of medicinal plants with almost 500 images of each species based on plant leaves. The model enables users to easily and accurately identify plants and access information about their medicinal properties and uses. The model has received accuracy of 94 % for testing and 98 % for training data, with improvements observed at different epochs. Model selection focuses on optimizing architectures and hyper parameters through cross-validation and grid search. Training of models is done on the annotated datasets. This work explores the potential of capturing plant images and leveraging algorithms that perform processing of images to draw out key features such as edges, textures, and colors from the image. They might detect parts of the leaf, like veins, shapes, or patterns specific to certain plant types. These features will then be harnessed to train an ML model, empowering it to recognize unknown medicinal plants from new image inputs. By automating the identification process, this research aims to bridge the gap in accessibility and contribute to a more efficient and reliable way of harnessing the power of nature’s medicinal bounty.