RiceLeafBD: A Field-Collected Rice Leaf Disease Dataset from Bangladesh

RiceLeafBD is a field-collected rice leaf disease image dataset developed to support research in plant disease detection, deep learning, computer vision, explainable artificial intelligence, and precision agriculture. The dataset was collected from paddy fields located in Savar Upazila, Akran, Birulia, and Savar Municipal Area of Dhaka District, Bangladesh, during October–November 2024.

The dataset contains 2,545 annotated images distributed across five classes: Blast (339 images), Narrow Brown Spot (329 images), Sheath Blight (454 images), Tungro (500 images), and Normal Leaf (923 images). Images were captured under real agricultural field conditions using consumer-grade smartphone cameras with varying lighting conditions, backgrounds, and disease severity levels.

All images were manually verified by agricultural experts and extension officers. Ambiguous samples, poor-quality images, and leaves showing multiple concurrent infections were excluded to ensure annotation reliability.

The dataset is intended for research in image classification, transfer learning, convolutional neural networks (CNNs), vision transformers (ViTs), explainable AI, and smartphone-based disease diagnosis systems. It was used in the study titled “A Comparative Study of CNN and Vision Transformer Models for Rice Leaf Disease Detection Using a Field-Collected Bangladeshi Dataset.”

This dataset is publicly released to support reproducibility, benchmarking, and future research in agricultural artificial intelligence.