Maize Variety Dataset for Classification
Description
The collection consists of images of three varieties of maize seeds: Wang Dataa, Sanzal Sima, and Bihilifa, obtained from Heritage Seeds Ghana. These maize cultivars are frequently grown in the northern area of Ghana. Heritage Seeds Ghana categorized and labeled the seeds for each of the three kinds manually at the collection station. The images were taken with a 12-megapixel camera on a phone, resulting in original JPEG images of different sizes. The photos were resized to dimensions of 224 by 224, ensuring uniformity. A blue background was employed during the capture process to maintain uniformity and enhance visibility in daylight situations. The photos were subsequently categorized into their corresponding classes: Wang Dataa, Sanzal Sima, and Bihilifa. In total, the dataset used for this study consists of 17,724 color photos.
Files
Steps to reproduce
Maize classification has been achieved through the utilization of deep learning techniques with hyperspectral imaging. However, the high power consumption of GPUs frequently prevents the deployment of these automated methods on embedded devices, necessitating the allocation of significant processing and computing resources. Access to local maize data in Ghana to perform classification tasks is also exceedingly challenging. To overcome these obstacles, the objective of this research is to generate a straightforward dataset that includes three distinct varieties of local maize seeds in Ghana: Wang Dataa, Sanzal Sima, and Bihilifa. The objective is to expedite the creation of a maize classification tool that is both efficient and cost-effective, while simultaneously minimizing human involvement in the seed grading process for marketing and production. The dataset is organized into a single folder named MaizeData, which contains a total of 17,724 entries. Three subfolders from each of the three varieties (classes) are included in the MaizeData folder. In particular, the Bhihilifa class contains 6,480 files, the Sanzal Sima class contains 5,100 files, and the Wang Dataa class contains 6,144 images. All of the images have undergone validation by experts from Heritage Seeds Ghana and are open for use by researchers.