What We Do
This project will create unprecedented ML-ready labeled datasets for crop type, crop pest and disease, and market prices in the main food production regions in five African countries. The team will use novel and innovative approaches that include rapid point data collection with cameras mounted on the hoods of vehicles—“helmets”—combined with crowdsourcing to create point and polygon labels. By partnering with local universities, this project will create opportunities for training future African researchers to use remote sensing and machine learning.
How Satellites Make This Work
Satellite data availability, rapid developments in cloud and computational capabilities, and big data analytics are revolutionizing the remote sensing field and open new opportunities to realize the long-held promise of remotely sensed data of providing accurate, reliable and timely crop-specific information at parcel to national levels across cropping systems. Earth Observation (EO) data can hence play an increasingly central role in agriculture, from informing government policies and humanitarian aid, stimulating resilience, to monitoring progress towards the intensification needed to meet global food needs sustainably. Access to EO, data processing infrastructure as well capacity to develop methods, have improved substantially in large producer countries. However, these are still lacking in smallholder systems that form a large percentage of agriculture in sub-saharan Africa, where these data are even more critical to farmers making decisions that impact their only source of livelihood. It is therefore critical to increase the number of training datasets to derive better crop type maps, and capacity development in critical institutions blended with Citizen science provides an opportunity to improve this situation.
Kenya, Mali, Rwanda, Tanzania, and Uganda are considered to be among Africa’s top food-producing regions and will be the countries of focus for this new and innovative approach to quickly collecting large amounts of labeled data through cooperative ground surveys. This project is a collaborative initiative between the NASA Harvest Africa Program based at The University of Maryland, The Regional Universities Forum for Capacity Building in Agriculture (RUFORUM), Center for Earth Observation and Citizen Science at the International Institute for Applied Systems Analysis (EOCS-IIASA), The Regional Center for Mapping of Resource for Development (RCMRD), The Eastern Africa Grain Council (EAGC), Lutheran World Relief Mali, International Maize and Wheat Improvement Center (CIMMYT), Radiant Earth Foundation, and Mapillary. Together, the team will approach data collection by implementing camera-mounted equipment to the hoods of vehicles (ie. the “helmets”) and combining this information with additional crowdsourced data in order to maximize the data points available for labeling. Achieving a quality dataset that can be usefully applied to machine learning algorithm development requires a massive amount of data, which is notably challenging to obtain on a widespread scale especially during a pandemic and in high-conflict regions. It is for this reason that “Helmets Labeling Crops” places a strong focus on rapid and easy data collection supported by citizen scientists and academic researchers alike. The models, systems and information products are user-driven, science-based, actionable, and the methods build capacity at the local level to effectively utilize the data and tools and ensure their uptake and long term sustainability by national and regional entities.
Not only will this project support the development of new African agricultural datasets, but a cross-benefit of this citizen science approach is that students, farmers, and others in the agricultural community will have an opportunity to explore image segmentation and receive training on how to collect and label the data. The hope is that individuals who are not usually involved in ‘typical’ academic research processes will feel empowered to use the tools already available to them (i.e. their smartphones) to take part in a community-based effort to bolster food security in their home countries.
The Lacuna Fund
Guided by machine learning professionals worldwide, Lacuna Fund provides data scientists, researchers, and social entrepreneurs with the resources they need to either produce new labeled datasets to address an underserved population or problem, augment existing datasets to be more representative, or update old datasets to be more sustainable.