Lightweight Dataset for Maize Classification on Resource-Constrained Devices

The dataset comprises images of three types of maize seeds: Wang Dataa, Sanzal Sima, and Bihilifa, sourced from Heritage Seeds Ghana. These maize varieties are commonly cultivated in the northern region of Ghana. At the collection point, Heritage Seeds Ghana manually sorted and labeled the images as either 'good' or 'bad' for each of the three varieties. The 'good' category represented high-quality maize seeds suitable for productive yields, while the 'bad' category included damaged, infected, or low-quality seeds not suitable for production. The images were captured using a 12-megapixel phone camera, resulting in original JPEG images of varying dimensions. Additionally, the augmented images were standardized to a size of 128 by 128. During capture, a blue background was used to ensure consistency and clarity during daylight, with no specific attention to lighting conditions. The images were then organized into their respective classes of 'good' and 'bad.' Overall, the dataset for this study comprises both raw (4,846 images) and augmented (28,910 images) color images.

本数据集包含三类玉米种子的图像：旺达塔（Wang Dataa）、桑扎尔西马（Sanzal Sima）以及比希利法（Bihilifa），数据来源为加纳遗产种子公司（Heritage Seeds Ghana）。上述玉米品种均为加纳北部地区常见栽培品种。在数据采集环节，该公司针对三个品种的图像分别进行了人工分拣与标注，将其划分为‘优质’与‘劣质’两类。其中‘优质’类别代表适用于高产种植的高品质玉米种子，‘劣质’类别则涵盖受损、染病或质量低下、不适宜用于生产的种子。图像由1200万像素的手机相机拍摄，原始图像格式为JPEG，尺寸各异。此外，数据增强后的图像已被统一调整为128×128的尺寸。拍摄过程中采用蓝色背景以确保日间拍摄时的一致性与清晰度，未对光照条件进行特殊控制。随后，所有图像被分别归入‘优质’与‘劣质’两个类别中。综上，本研究使用的数据集共包含原始彩色图像4846张与数据增强后的彩色图像28910张。