Coconut capture app
Field data that existed only at farm gates, captured tree by tree — the dataset a yield model gets trained on.
- 500+ farms
- 8 guided angles per tree
- geolocation-tagged dataset
The challenge
Coconut yield prediction needed consistent, standardised field imagery — but there was no reliable way for agents to capture prescribed photos of each tree and tie them to the right tree and farm.
What we built
An Android app that guides a field agent step by step to photograph each tree from 8 prescribed angles, tags every image set to the specific tree and farm, captures geolocation, and maintains the database feeding the ML pipeline.
The outcome
500+ farms targeted across South India (20–50 trees each) captured into the yield-prediction training dataset. Annotation happens in CocoDag (CS-15).
Bring the problem. Leave with a scope, a timeline, and the number it should move.