Saffron-DB: An image dataset for saffron quality control, classification, and automated assessment based on image analysis and deep learning.

Saffron-DB is an image dataset of Moroccan saffron stigmas (Crocus sativus) prepared for computer vision, image processing, and automated saffron quality assessment. The dataset contains high-resolution images collected from three Moroccan geographical origins: Meknès–Sidi Slimane, Azilal, and Taliouine. The images were acquired under indoor conditions using a uniform white background to support visual consistency and reproducibility. The dataset includes different image formats depending on the acquisition device. For the Meknès–Sidi Slimane and Azilal samples, images were captured using a Nikon D3100 digital camera and are provided in both RAW Nikon format (.NEF) and corresponding JPG format. The RAW files preserve the original acquisition information, while the JPG files provide ready-to-use images for visualization, preprocessing, and computational analysis. For the Taliouine sample, images were acquired using an iPhone 15 Pro Max and are provided in JPG format. The dataset content is summarized as follows: • Original JPG images: 318 images in total. • RAW NEF files: 214 files in total, available for the Nikon-acquired samples. • Augmented JPG images: 1590 images in total. • Total image files: 2122 files when the RAW NEF files are included. • Meknès–Sidi Slimane folder: 107 original JPG images, 107 RAW NEF files, and 535 augmented JPG images. • Azilal folder: 107 original JPG images, 107 RAW NEF files, and 535 augmented JPG images. • Taliouine folder: 104 original JPG images and 520 augmented JPG images. The dataset also includes augmented JPG images generated from the original JPG images using Python-based image processing tools, including OpenCV and Albumentations. The augmentation process applied controlled geometric and photometric transformations, including flipping, rotation, scaling, translation, brightness and contrast adjustment, and Gaussian blur. Each original image produced five augmented variants, which are stored separately from the original images to preserve traceability. Saffron-DB is organized in a hierarchical folder structure. Each sample folder contains subfolders for the available image formats, including NEF images when available, original JPG images, and augmented JPG images. File names include sample identifiers, regional information, and image indices to support reproducible use of the dataset. This dataset can be reused for saffron image classification, visual quality assessment, color and morphology analysis, feature extraction, preprocessing evaluation, and the development of machine learning or deep learning models for agricultural product inspection. It may also support studies related to food quality control, digital agriculture, post-harvest evaluation, and traceable saffron classification.