Benefits of a Multi-Species Medicinal Leaf Dataset: Azadirachta indica, Centella asiatica, Kalanchoe for Deep Learning Applications

The study utilizes a primary image dataset of medicinal plants, specifically curated to facilitate automated disease detection and health monitoring. The data acquisition campaign was conducted over a period of 45 days, from September 8, 2025, to October 22, 2025. To ensure environmental diversity, samples were gathered from three distinct geographical locations in Bangladesh: Hajigonj in Chandpur (for Kalancho, Centella, Mikania), Daffodil Smart City in Dhaka (for Azadirachta Indica), and Shakipur in Tangail (for Moringa). The images were captured using smartphone cameras—specifically the OnePlus 7T, OnePlus Nord CE2, and Realme 7i—under natural lighting conditions. The original images were captured in high resolution, with pixel dimensions ranging from 3000 to 4000 pixels. To ensure uniformity and computational efficiency for the deep learning model, all images were preprocessed and resized to a standard dimension of 512×512 pixels.The dataset comprises a total of 1,578 original images categorized into 12 distinct classes across five crop species. The distribution includes Azadirachta Indica (Neem) separated into Disease and Healthy categories, and Centella Asiatica with similar Disease and Healthy labels. Kalancho images are classified into Asymptomatic, Chewing Insect Damage, and Complex Foliar Damage. The Mikania Micrantha species includes Disease Affected, Distorted, and Healthy classes, while Moringa samples are labeled as either Healthy or Yellow. For the development and rigorous evaluation of the proposed model, the dataset is partitioned using a stratified sampling approach into an 80:10:10 ratio, allocating 80% for training, 10% for validation, and 10% for testing.