PaddyVision: A Structured Image Classification Dataset for Bangladeshi Paddy Varieties using Machine Learning

Md Mijanur Rahman, Pallabi Karmaker, Abdullah, Tanjim Tabassum Urmi, Akhir Ahmed Akash

This dataset includes an exploratory collection of Bangladeshi paddy variety images withvariety-based labels. The dataset was developed for research and experimentation purposesin the fields of computer vision, machine learning, and agricultural artificial intelligence. Thedataset includes 6,000 images of 60 Bangladeshi paddy varieties, with 100 images pervariety, collected between January 2025 and April 2026. Images were captured usingsmartphone cameras in real-world agricultural environments including paddy fields, cultivationlands, and nearby farming regions. Images were collected under different lighting conditions,weather environments, and camera perspectives to improve variability and robustness. Allimages were manually inspected, resized to a uniform resolution of 224 × 224 pixels, andorganized into class-specific directories. The dataset contains balanced image distributionsacross 60 paddy variety classes, each containing 100 images, ensuring uniformrepresentation for machine learning and deep learning applications. Images follow astandardized naming convention, and a metadata CSV file is provided containing imageidentifiers and corresponding paddy labels. This dataset can be used for exploratory imageclassification, transfer learning, crop recognition, smart farming research, and agricultural AIsystem development. The dataset also supports educational purposes related to responsiblemachine learning and agricultural computer vision applications.