Ayurvedic Plant Identification Through Deep Learning Approaches A Case Study in Sri Lanka

Medicinal plants provide a cost-effective alternative to modern medicine with fewer side effects, yet their benefits remain underutilized due to challenges in accurate identification. Manual identification demands expert knowledge, making it inaccessible to non-specialists. This study addresses these challenges by leveraging deep learning to identify 32 medicinal plant species accurately. A dataset comprising 1,500 images per species was collected from notable Sri Lankan locations, encompassing various plant parts such as leaves, roots, tubers, flowers, and fruits. Data preprocessing techniques enhanced model performance, including resizing, noise removal, and augmentation. The dataset was split into training (70%), testing (20%), and validation (10%) sets. After evaluating MobileNetV3, ResNet50, and EfficientNetB0. The fine-tuned EfficientNetB0 emerged as the optimal model with a 98.03% accuracy. Additionally, YOLO achieved an F1 score of 0.8 for object detection. The proposed hybrid approach combining DCNN and EfficientNetB0 offer a reliable solution for medicinal plant identification.