Saffron-DB: A High-Resolution Image Dataset for Saffron Quality Control, Classification, and Automated Assessment Using Image Analysis and Deep Learning
Description
Saffron-DB is a high-resolution image dataset of Moroccan saffron stigmas (Crocus sativus L.) collected from three Moroccan regions: Meknès, Azilal, and Taliouine. The dataset was prepared for computer vision, image processing, and saffron quality assessment. Key dataset information: • 318 original JPG images corresponding to 318 unique saffron stigma specimens. • Meknès: 107 original JPG images/specimens. • Azilal: 107 original JPG images/specimens. • Taliouine: 104 original JPG images/specimens. • 214 RAW Nikon NEF files are provided for the Nikon-acquired specimens. • 1590 augmented JPG images are provided as derivative processed files. • Total repository content: 2122 image files. The images were acquired indoors under controlled conditions on a uniform white background to ensure consistency and minimize visual noise. A Nikon D3100 camera was used for the Meknès and Azilal samples, while an iPhone 15 Pro Max was used for the Taliouine samples, reflecting differences in acquisition setups. Augmented images, generated through various image processing techniques, are stored separately from the original images to preserve data integrity. Metadata are provided to link each image with its specimen identifier, region of origin, acquisition device, image format, resolution, and processing status, facilitating traceability and reproducibility.
Files
Steps to reproduce
The dataset was prepared through the following workflow: 1. Saffron stigma specimens were collected from three Moroccan regions: Meknès, Azilal, and Taliouine. 2. Images were acquired indoors on a uniform white background. Meknès and Azilal specimens were captured using a Nikon D3100 camera, while Taliouine specimens were captured using an iPhone 15 Pro Max. 3. Images were visually inspected, and blurred, overexposed, poorly framed, or unsuitable images were removed. 4. The retained original JPG images were cropped and standardized to 3024 × 3024 pixels. 5. Augmented JPG images were generated from the original JPG images using Python libraries, including OpenCV and Albumentations. Each original image produced five augmented variants using flipping, rotation, scaling, translation, brightness adjustment, and contrast adjustment. 6. Files were organized under the root folder Saffron-DB by geographical origin and image type: original JPG images, RAW NEF files when available, and augmented JPG images. 7. Metadata were prepared to link each image with its specimen identifier, region, acquisition device, image format, resolution, and processing status.