Brief

We wanted to deliver a technically practical system that was fast enough to distinguish and identify plants needing treatment. To be commercially viable, the system had to be rugged and affordable. This demanded a move away from powerful specialist platforms with delayed data processing towards ‘AI at the edge’ – where data is processed immediately and acted on locally.

Approach

We developed our Fafaza solution to run on off-the-shelf components and with an AI compute platform costing less than $100. We added a low-cost camera to identify the target by texture, and machine vision probabilistic programming to track the target and determine where to aim and when to fire. Despite it being low-cost, we enabled Fafaza to identify, classify and apply treatment in real time, while moving at the speed of a tractor.

By targeting specific undesirable weed leaves in a crop, Fafaza dramatically reduces the amount of chemical used and virtually eliminates drift and run-off. As well as helping to save farmers’ money, it cuts ground water pollution, prevents wind-borne contamination and eases regulatory compliance.

Benefit

As connectivity is often poor in rural environments, the ability to run complex machine learning-based applications on a tractor implement creates countless opportunities to collect, process and act on data there and then. Fafaza demonstrates how ‘AI at the edge’ smart farming approaches have become practical, cost-effective and deployable today.