BD RiceSeed: A Multi-Class Image Dataset of Bangladeshi Rice Seed Varieties for Classification Tasks

The dataset was curated from rice seed samples collected across agricultural zones in Bangladesh and features eight distinct rice seed types. The dataset comprises high-resolution seed images categorized into eight varieties: 25, 28, 29, 89, 100, Chinigura, Kata Irri, and Kata Irri Vog. Each category contains 400 original images, each captured at a resolution of 1920 × 1080 pixels.

To enhance the dataset quality and ensure consistency, background removal preprocessing was performed on the original images, resulting in a refined set of seed images optimized for better feature extraction. This preprocessed dataset is stored separately under the "BGRemoved_Image" directory.

In order to mitigate class imbalance and strengthen model performance through diverse visual perspectives, data augmentation techniques were applied. These techniques produced 1,000 synthetic images per class, resulting in a balanced and enriched dataset totaling 8,000 images in the "Augmented_Image" directory.

The dataset is organized into three primary directories:

1) Rice_Image: Contains the original, unaltered images of eight rice seed varieties.
2) BGRemoved_Image: Includes preprocessed images with background removed to improve feature clarity.
3) Augmented_Image: Holds the balanced dataset created via augmentation, used for training and evaluation purposes.

This rice seed classification dataset provides a valuable resource for researchers in the fields of agricultural informatics, machine vision, and deep learning-based crop seed classification. It supports the advancement of AI-driven agricultural technologies for seed identification, classification, and quality assurance.