Multi-variety Green Coffee Beans and Adulterated Samples

This dataset contains coffee bean image data in two independent parts: a multiclass coffee bean variety dataset and an adulteration-level image dataset. It includes image data and accompanying metadata only; no training code, model weights, or prediction outputs are provided. The multiclass subset contains 17,958 PNG images of individual coffee beans across 14 classes. The original data split is retained, with 14,367 images in the training set and 3,591 images in the validation set. Each image is associated with a unique identifier, class label, split assignment, and relative file path. This subset is intended for research on coffee bean variety image classification. The adulteration subset contains 30 BMP images organized into six reported adulteration levels: 0%, 1%, 5%, 10%, 20%, and 30%, with five images at each level. Its metadata provide the adulteration level, file path, and replicate identifier. This subset may support adulteration-level recognition, image feature exploration, and preliminary evaluation of related methods. The dataset includes sample manifests, class and adulteration-level definitions, data dictionaries, version history, and SHA-256 checksums to support discovery, integrity verification, and reproducible reuse. The Zenodo record is the canonical archival and citation source; the Kaggle dataset is a corresponding mirror.