Blue River's See & Spray uses computer vision to spray weeds, not crops. Labelbox's platform produced the image-segmentation signal that trains those models, cutting signal-production time by over 50%.

The challenge

Blue River Technology builds smart agriculture equipment for sustainable farming through robotics and computer vision. Weeds steal nutrients and stunt crops, and farmers face a bad tradeoff: broad automated spraying that can damage crops, or costly manual spraying. Its See & Spray technology uses computer vision to spray only the weeds and preserve the crops. Behind it are machine learning models that identify weed versus crop, trained on hundreds of thousands of images. The hard part is the signal: image segmentation, where every plant needs a specific class — the most intricate labeling task there is — which made producing it expensive.

The approach

Labelbox produced the signal with model-assisted labeling. Instead of labeling each image from scratch, Blue River's model generated masks and contributors corrected the errors — faster, and focused on the model's real problem areas: the edges and the boundary between weed and crop within contiguous segments of organic material. The work centered on correcting classes and labels rather than hand-labeling, standardizing how the team creates and manages signal in one place.

The outcome

Blue River cut signal-production time by over 50% — worth millions of dollars in savings per year — and freed its team to focus on correction instead of labeling from scratch. The technology's goal is to reduce herbicide use by 70-80%, one of the costliest line items in a modern farmer's budget. Further gains come from active learning and additional automation in Labelbox.

Where this goes

See & Spray is embodied intelligence in the field. Better segmentation signal, produced faster, is what lets a machine make the right call on every plant.