An extensive image dataset for deep learning-based classification of rice kernel varieties in Bangladesh

Area of Science:

• Agricultural Science
• Genetics
• Computer Vision

Background:

• Understanding rice genetic diversity is crucial for crop improvement.
• Existing datasets may lack the resolution or diversity needed for advanced analysis.
• Collaborations between agricultural institutes enhance research capabilities.

Purpose of the Study:

• To introduce a comprehensive, high-resolution image dataset of local rice varieties.
• To support research in rice genetics, breeding, and agricultural practices.
• To leverage advanced imaging and data augmentation for agricultural data enrichment.

Main Methods:

• Development of a dataset featuring 38 local rice varieties.
• Acquisition of 19,000 original high-resolution microscopic images.
• Application of data augmentation techniques (scaling, rotation, lighting) to generate 57,000 additional images, totaling 76,000 images.

Main Results:

• A rich, augmented dataset of 76,000 images detailing 38 rice varieties.
• Images capture unique characteristics like color, size, and agricultural utility.
• The dataset simulates diverse environmental conditions through augmentation.

Conclusions:

• The dataset provides a valuable resource for analyzing rice genetic diversity.
• It facilitates the development of innovative agricultural practices and breeding programs.
• This resource empowers researchers to effectively study and utilize rice genetic resources.