A visual dataset for recognition of rice varieties

Area of Science:

• Agricultural Science
• Computer Vision
• Data Science

Background:

• Rice is a staple food in Bangladesh, with numerous local varieties.
• Accurate identification of rice varieties is crucial for agriculture and food security.
• Existing datasets may not capture the diversity of local Bangladeshi rice strains.

Purpose of the Study:

• To create and present a comprehensive, publicly available dataset of Bangladeshi rice varieties.
• To facilitate research in agricultural image analysis and machine learning applications.
• To document the visual characteristics of 20 distinct local rice types.

Main Methods:

• Collected 4,730 low-resolution images of 20 rice varieties using smartphone cameras across Bangladesh.
• Included augmented data, resulting in a total of 23,650 images.
• Ensured diverse representation of rice colors, sizes, and market origins.

Main Results:

• A dataset featuring 20 distinct rice varieties, including Subol Lota, Bashmoti (Deshi), BR-28, and Shorna-5.
• Images capture variations in color, size, and texture for each variety.
• The dataset exhibits non-uniform distribution across varieties.

Conclusions:

• The curated dataset provides a valuable resource for agricultural research and development in Bangladesh.
• It supports the development of automated systems for rice variety identification.
• Potential applications include crop monitoring, quality control, and breeding programs.