Transparent Machine Vision Techniques For Medicinal Plant Species Identification

Medicinal plants have long been the foundation of the medical system and a source of health and healing, but many people nowadays are unaware of these priceless natural resources or the range of possible applications they provide. Traditional medical professionals known as herbalists mostly depend on their understanding of therapeutic herbs to cure a wide range of illnesses. Unfortunately, identifying medicinal plants by hand from their leaves takes a lot of time, labour, and specialised knowledge. The purpose of this project was to create and construct a web application for the identification of plant species, namely medicinal plants that are present across Uganda. On a publicly available dataset, many machine learning and deep learning models, including Convolutional Neural Networks (CNN), VGG19 and 16, Resnet50, Vision Transformer (ViT), and EfficientNet B0, were trained. Many features in the dataset were extracted from each leaf such as its length, width, texture, area, and color. For the ten classes of medicinal plant species identified in the globe, the model’s accuracy rates were 99%, 95.70%, $\mathbf{9 2. 5 8 \%}, \mathbf{9 7. 9 0 \%}, \mathbf{9 1. 2 0 \%}$, and $\mathbf{3 5. 6 9 \%}$, respectively. In order to improve transparency and visual explanations for its picture classifications, this model makes use of Grad-CAM and SHAP, which are explainable AI (XAI) approaches. It is anticipated that a web-based system for the automatic identification of medicinal plants will contribute significantly to the production of pharmaceutical drugs, help members of the community increase their knowledge of medicinal plants, and assist taxonomists in creating more effective methods for identifying species.