RiceLeafBD: A Real-Field Image Dataset for Rice Leaf Disease Detection and Classification in Bangladesh

The RiceLeafBD dataset is a newly developed, high-quality image collection designed to support research on rice leaf disease detection and classification using computer vision, machine learning, and deep learning techniques. The dataset was created to address the lack of authentic, real-field agricultural datasets for rice crops in Bangladesh. It includes a total of 1,555 RGB images of rice leaves captured under natural field conditions in Sylhet and Dhaka, Bangladesh, covering four major categories: Healthy, Bacterial Leaf Blight, Brown Spot, and Tungro Virus.

Images were taken during the Amon rice season (August–December) using smartphone cameras (Samsung Galaxy S21 Ultra and Redmi Note 9), ensuring diversity in resolution, lighting, and background conditions. Each image was captured from the adaxial (upper) surface of the leaf under varying weather and environmental situations such as cloudy, sunny, and windy days. This diversity enhances the dataset’s robustness for developing and testing deep learning models that generalize well to real-world agricultural environments.

All images were validated with the assistance of agricultural experts from Jamalpur Krishi Gobeshona Institution, ensuring biological accuracy and disease classification reliability. The dataset also includes an accompanying CSV annotation file specifying disease labels and metadata.

RiceLeafBD is suitable for applications in:

- Image-based disease detection and classification

- Transfer learning and model benchmarking

- Precision agriculture and smart farming solutions

- Computer vision research in plant pathology

Key Features:

- Total images: 1,555

- Image type: RGB

- Categories: Healthy (252), Bacterial Leaf Blight (417), Brown Spot (356), Tungro Virus (530)

- Data format: Images (.jpg/.png)

- Source locations: Sylhet & Dhaka, Bangladesh

- Collection period: August–December (Amon season)

Potential Uses:
This dataset can be used to train and evaluate deep learning models for rice leaf disease identification, perform transfer learning experiments, and support comparative analyses in agricultural image processing research. It serves as a reliable open-access resource for researchers, practitioners, and educators working toward sustainable agriculture and food security.

Citation:
If you use this dataset, please also cite:
Rimi, S. A., Chowdhury, M. J. U., Abdullah, R., Ahmed, I., Mim, M. A., & Rahman, M. S. (2025). Empowering Agricultural Insights: RiceLeafBD – A Novel Dataset and Optimal Model Selection for Rice Leaf Disease Diagnosis through Transfer Learning Technique. arXiv preprint arXiv:2501.08912.