Machine Learning Imagery Dataset for Maize Crop: A Case of Tanzania.

Abstract

Maize is one of the most important staple food and cash crops that are largely produced by majority of smallholder farmers throughout the humid and sub-humid tropic of Africa. Despite its significance in the household food security and income, diseases, especially Maize Lethal Necrosis and Maize Streak, have been significantly affecting production of this crop. This paper offers a dataset of well curated images of maize crop for both healthy and diseased leaves captured using smartphone camera in Tanzania. The dataset is the largest publicly accessible dataset for maize leaves with a total of 18,148 images, which can be used to develop machine learning models for the early detection of diseases affecting maize. Moreover, the dataset can be used to support computer vision applications such as image segmentation, object detection and classification. The goal of generating this dataset is to assist the development of comprehensive tools that will help farmers in the diagnosis of diseases and the enhancement of maize yields thus eradicating the problem of fod security in Tanzania and other parts in Africa.

Code and data availability

The paper's core asset is its own maize leaf imagery dataset (18,148 images of healthy/MLN/MSV leaves), publicly deposited by the authors on Harvard Dataverse with an explicit DOI and direct URL. The annotation tools cited (VisiPics, LabelMe, Makerere web annotation tool) are generic third-party tools, not paper assets