What Is Precision Spraying?

Precision spraying applies herbicide, fungicide or pesticide only to the plants that need it, rather than spraying the whole field at a set rate. It is also called spot spraying or targeted spraying, and it is one of the most practical forms of precision agriculture.

A precision sprayer carries cameras along its boom, looking at the ground just ahead of the nozzles. As the machine moves, a computer on board analyses each image, decides whether it contains a weed, works out where that weed will be when it reaches the nozzle line, and opens the right nozzle for a fraction of a second. The rest of the boom stays shut. On a field with scattered weeds, most of the ground is never sprayed at all.

There are two kinds of detection, and they need very different computing.

Green-on-brown can rely on color alone, so it has been available for decades. Green-on-green is the harder problem: a weed seedling can look very like the crop, so the system needs a deep learning model trained on thousands of labeled images, and an AI computer fast enough to run it on every frame from every camera.

Every part of the vision chain has to keep up with the machine and survive life on it.

A sprayer at 15 mph covers almost 7 meters every second. If the system takes 50 milliseconds from capturing an image to opening a nozzle, the boom has moved around a third of a meter in that time. The cameras therefore sit ahead of the nozzles, and the computer has to finish every frame within a fixed time budget, every time, or the spray lands behind the weed.

That is why precision spraying runs on the machine rather than in the cloud. Sending images over a mobile link would add delay and depend on coverage that many fields do not have. The spot spraying case study describes an autonomous platform that does this at 15 mph, with GigE cameras, a sealed M12 switch and a rugged Jetson AGX Orin computer.

A green-on-green model is only as good as the images it learned from. Developers collect pictures from the sprayer’s own cameras across many fields, growth stages, soil types and lighting conditions, then label each weed and crop plant. The model is trained on GPU servers and then compressed, typically to INT8, so it runs fast enough on the computer on the machine.

Each season adds new images, especially of the cases the model got wrong, and the improved model is pushed back to the fleet. That loop of capture, label, retrain and redeploy is why systems keep getting better, and why the edge computer needs enough headroom to run next year’s larger model as well as this year’s.

The main benefit is using far less chemical. Savings depend on weed pressure: where weeds are scattered, most of the field is never sprayed. Less herbicide cuts input cost, reduces residue and drift, protects the crop from unnecessary contact and helps growers meet stricter rules on pesticide use. Some systems also map weed patches as they go, which helps plan future treatments.

The limits are real. Dense weed infestations leave little to save, models trained in one region and crop may need retraining elsewhere, and dust, low sun and wet leaves all challenge the cameras. Systems are improving each season as training data grows. Our rugged Jetson computers are built for the sealed, vibration-heavy conditions these machines work in.

We supply rugged edge AI computers and sealed networking for precision sprayers, with cameras, switches and the AI software stack integrated and tested as standard.