Blue River Technology, a subsidiary of farming equipment maker John Deere, has started putting deep machine learning and to use for farmers to help make their lives easier with smart machines.

Farmers use Blue River’s tools to control weeds and reduce costs with an intelligent sprayer that can “see” weeds and differentiate them from crop plants.

Today, most farming is highly automated and involves a large toolset of machines that furrow and till fields, plant seeds and tend to the crops. As technology has advanced, these machines that tend to crops can also watch them using high-resolution cameras to discern their health, apply pesticides and fungicides to kill harmful insects and fungus, and also spray for weeds.

Unlike the former two pests, weeds can be directly targeted for spraying because they’re plants that pop up amidst crops.

A problem arises, however, because weeds and crops are both plants. As a result, the sprayer needs to be intelligent about how it applies the spray so that it doesn’t hit a crop plant with herbicide and doesn’t leave too many weeds in the field to choke out the crops.

To do that, the machine needs to make real-time decisions on what is a crop and what is a weed. As the machine drives through the field, it uses high-resolution cameras to collect data about the field at a high frame rate.

That data is then ingested by a convolutional neural network using PyTorch, an open-source machine learning library, to identify weed from crop, create a pixel-accurate map of the field and thus guide the robot sprayer to hit only the proper targets.

This entire process takes mere milliseconds of time so that the robot sprayer can traverse the field as quickly as possible and still accurately target weeds while leaving crops untouched.

Because the machines only spray exactly where the weeds are found, this approach reduces the amount of herbicide used to control weeds by 90%. Thus, farmers save a lot of money on expensive chemicals and at the same time can promote more sustainable agricultural practices that reduce the number of toxins sprayed onto the soil.

To make this technology possible, Blue Technologies makes use of two industry tools for machine learning: PyTorch and Weights & Biases.

PyTorch is an open-source library for machine learning primarily developed by Facebook Inc.’s AI Research lab. It is the library behind numerous computer vision and adaptive learning technologies such as Tesla Autopilot, a self-driving car technology, and Uber’s Pyro, an AI smart agent that matches drivers and riders.

Weights & Biases is a machine learning modeling platform used to refine data models. In the case of computer vision, it is used to assist in the training of machine learning algorithms to recognize the difference between weeds and crops.

Just like a human, a machine learning program must be taught to understand what it’s expected to do. As part of Blue River’s process, it’s often necessary to teach the algorithm to recognize what a cotton plant or another crop looks like as well as train it to know what an invasive weed such as pigweed looks like.

To train it, the machine learning system receives a giant variety of images featuring cotton plants, to identify “good” plants, and then a huge variety of pictures of pigweed, to identify “bad” plants. After training, the computer vision system can then identify any plant it sees and rate its confidence that it’s a crop or a weed.

The result is a computer model that can quickly digest a rapidly moving image from a camera pointed at the field and model which plants are weeds and which are crops. That not only allows the sprayer to hit the weed directly but also keep the spray a safe distance away from crops.

In the United States alone there are 915 million acres of farmland, and food-related industries contributed more than $1.1 trillion to the U.S. gross domestic product in 2017. The world’s population is expected to grow to nearly 10 billion by 2050, which will increase the global demand for food by 50%. There isn’t much land left to farm, so making more efficient use of that land will become even more necessary.

With these advances in ML technology and the ability to reduce costs while increasing efficacy for farmers, Blue River hopes to continue using AI to pave the way to that future.

Photo: Blue River Technology