MedPlant-BD: A Visual Dateset of Native Medicinal Plants from Bangladesh for Machine Learning Applications

Our dataset consists of 10 classes representing 10 different species of medicinal plant leaves. It contains a total of 20,000 images, of which 5,000 are original images and 15,000 are augmented images. The dataset is divided into three parts based on percentage distribution: 70% (14,000 images) for training, 20% (4,000 images) for testing, and 10% (2,000 images) for validation. The training directory contains 14,000 images, with 1,400 images per class. The testing directory includes 4,000 images, with 400 images per class. The validation directory has 2,000 images, with 200 images per class. This comprehensive dataset provides sufficient and diverse samples for machine learning and deep learning models to accurately identify medicinal plant species based on leaf images. Each image is of high quality and captured under various conditions to enhance the model’s generalization capability. The structured partitioning of the dataset into training, testing, and validation sets ensures proper model training and reliable performance evaluation.