RGB Maize Kernels Image Dataset for grain quality screening: A use case- Detection of Aflatoxins and Fumonisins in Maize Using Computer Vision

This dataset comprises 5,143 consistently captured RGB images of maize kernels, designed to advance computer vision for food safety. Acquired under controlled lighting and background conditions, it minimizes extraneous variability to reliably train models for detecting aflatoxin and fumonisin contamination. Each image is annotated in YOLO format, with labels validated against laboratory measurements to ensure accuracy. Critically, images contain multiple kernels, mirroring real-world batch inspection scenarios and making the dataset particularly suitable for developing practical, edge-AI applications for use in resource-constrained environments like smallholder farms. To facilitate immediate use, the data is organized for compatibility with standard YOLO pipelines; the ‘images’ folder is split into sequential 7-Zip archives (begin extraction from images.7z.001), while labels are provided in a single labels.zip file. Users are advised to verify annotation alignment after extraction. Curated with reproducibility in mind, this resource adheres to FAIR data principles, supporting scalable, non-destructive screening research and fostering the development of low-cost, mobile tools for maize quality assessment.