Where this conversation describes AI being used to build or deliver a product. Part of the AI in practice tracker.
Engineering & simulationBlue River See & Sprayroughly 70% savings in herbicides
The system uses computer vision and deep learning to detect weeds versus crops in real-time, enabling the sprayer to target only weeds and reduce herbicide usage by roughly 70%.
“by putting cameras across the spray boom, um, we can take pictures of the ground, detect the crop, detect the weeds and spray herbicides and only the weeds. The result is like roughly 70% savings in herbicides for farmers…”
Engineering & simulationJohn Deere Autonomous Tractor (AR)
The team used stereo cameras and deep learning image segmentation to identify objects like rocks or cows, designing the network to heavily penalize missing unexpected objects to ensure safety in dusty, low-speed farm environments.
“we ended up with stereo cameras. Okay. And the other thing, you know, because we were such a deep learning, heavy company, you know, image segmentation just really isn't a huge problem anymore. With enough well labeled data. You can kind of figure out what you're looking at in any given scene.”
Engineering & simulationBlue River Machine Learning Models
The company developed a system where models identify their own weaknesses, automatically capturing and labeling that specific data to retrain the system, allowing the technology to self-heal and improve performance over time without manual intervention.
“what we're starting to do now is actually make models to discover where the model's weak. And capture that data. So pick it off, you know, while it's running on the system and and say, hey, this is this is getting cute for upload. That data gets labeled, that data ends up in a train set and the performance improves.”