Boosting crop yield predictions for 10,000+ smallholder farmers with 30,000+ annotated satellite & drone images, achieving 85% accuracy across East Africa.

Client Overview

An agri-tech startup focused on precision farming in East Africa aimed to build a predictive analytics platform for smallholder farmers. They required annotated geospatial data to train models on crop health and yield forecasting, but lacked access to region-specific datasets.

Project Scope

DataLens annotated 30,000+ satellite and drone images covering maize, cassava, and coffee crops across varying terrains. Tasks encompassed:

• •Bounding box labeling for pest infestations
• •Pixel-level segmentation for soil erosion
• •Temporal tracking for growth stages

Data was sourced ethically from public and client-provided imagery.

Challenges

• •Seasonal data gaps due to cloud cover in equatorial regions.
• •Linguistic barriers in labeling indigenous crop varieties.
• •Scalability for real-time farmer advisories.

DataLens Solution

Leveraging supervised annotation workflows, DataLens deployed field verifiers for ground-truth collection in Kenya and Tanzania. The platform's geospatial tools facilitated vector-based labeling, with iterative feedback loops refining models for climate-adaptive predictions.

Results

• •Delivered datasets with 97% inter-annotator agreement.
• •Boosted the client's yield prediction model accuracy to 85%, increasing farmer incomes by an estimated 15-20%.
• •6-week project timeline, enabling seasonal rollout to 10,000+ farmers.