Helmets Labeling Crops: Kenya Crop Type Dataset Created via Helmet-Mounted Cameras and Deep Learning

Catherine Nakalembe , Ivan Zvonkov, Hannah Rae Kerner, Diana Botchway Frimpong, Kenneth Mwangi, Jane Kioko, Bhanu Tokas, Kartik Jawanjal, Iman Smith, Anjali Paliyam, Christopher Wakhanala, Ana M Tárano, Shreya Jha, Priscilla Mawuena Loh

Accurate, up-to-date agricultural monitoring is essential for assessing food production, particularly in countries like Kenya, where recurring climate extremes, including floods and droughts, exacerbate food insecurity challenges. In regions dominated by smallholder farmers, a significant obstacle to effective agricultural monitoring is the limited availability of current, detailed crop-type maps. Creating crop-type maps requires extensive field data. However, the high costs associated with field data collection campaigns often make them impractical, resulting in significant data gaps in regions where crop production information is most needed. This paper presents our inaugural dataset comprising 4,925 validated crop-type data points from Kenya’s 2021 and 2022 long-rain seasons. Collaborating with institutional partners and an extensive citizen science network, we collected georeferenced images across Kenya using GoPro cameras. We developed and implemented a deep learning pipeline to process images into crop-type datasets. Our methodology incorporates rigorous quality control measures to ensure the integrity and reliability of the data. The resulting dataset represents a significant contribution to open science and a valuable resource for evidence-based agricultural decision-making.