Identification of Toxic and Medicinal Plants using Deep Learning

Abstract

This research proposes the development of a deep learning system utilizing YOLOv8-seg technology that can classify plant leaves as medicinal, toxic, and non-toxic. 4,000 labeled images were gathered from Kaggle, botanical databases, and actual plant images taken from the field. The network was trained on the Tesla T4 GPU for 100 epochs, resulting in an 88.8% Box mAP50 score, 79.0% Mask mAP50 score, 89.2% precision score, and a macro F1-score of 85.0%. The per class accuracy obtained 99.8% for the medicinal plant category, 81.8% for toxic plants, and 73.5% for non-toxic plants. A robustness test conducted under various situations (complicated backgrounds, poor lighting, overlapping plants) achieved an average mAP50 score of 75.7%. This study designed an interface using Gradio that supports both Arabic and English