This plantation group was counting fruit by hand with spot checks, and it had photos and drone imagery it had no way to read.
The company wanted to know how much fruit each harvest produced, but counting was manual. Someone spot-checked a sample by hand. They had a large archive of harvest photos and no way to count from them.
They also had drone and satellite images of their plantations, and no way to pull data out of them, like finding individual trees.
They hadn't run a machine learning project before, and they wanted a partner who would be straight with them about what AI could and couldn't do.
We started with a proof of concept, then built a custom computer vision model. It counted fruit bunches in harvest photos and picked out individual trees and their features from drone footage, with the results available through an API.
Later, we moved the model into the company's own cloud environment so it could run inside their systems.
This was one of our earlier projects, from when a lot of our work was building custom models and training data for machine learning teams. It's a good example of taking a pile of images nobody could use and turning it into data a team can act on.
If you have photos or drone imagery and no good way to get data out of them, this is the kind of work we did here. Talk to us about it.
