Published on August 7th, 2020 📆 | 6972 Views ⚑0
Blue River Technology Uses Facebook AI For Weed Control
With crop prices in the dumpster and the world’s population growing among a changing climate, artificial intelligence is becoming a life-saving measure for many farmers. From automated planting and harvesting to unmanned vehicles for cultivation and soil sampling, AI has begun to make it more cost efficient for producers to do their job.
One of the largest roadblocks is herbicides. According to a 2016 University of Illinois study, the chemical prices are on the rise and pose a big threat to a farmer’s bottom line.
“Growth rates of fertilizer, pesticide, and seed costs have been higher in years following 2006 than they were between 1990 and 2006. In 2015, the sum of fertilizer, pesticide, and seed costs were 48% of crop revenue, much higher than the 36% average from 1990 to 2006. These costs need to decrease, particularly if corn prices remain below $4.00. Otherwise, it will be difficult for revenues to be less than total costs.”
Now, Facebook is getting into the farming game. Artificial intelligence co-maintained by the company, Pytorch, is being put to use by Blue River Technology — a subsidiary of John Deere — to improve weed control and bring down the costs of herbicide use.
“Farmers today face a huge challenge — feeding a growing global population with less available land,” said Alexandru Voica, Technology Communications Manager for Facebook in an email. “This, paired with variability inherent in farming, like changing weather conditions, and threats like weeds and pests, impact a farmer’s ability to produce food. (Blue River Technology shows how) PyTorch-enabled deep learning technology powers agricultural robots that make real-time decisions in the field; analyze images of crops in a matter of milliseconds; and support the building of ‘weed detection models’ for more efficient farming.”
The weed-controlled device works by integrating cameras, computer vision, machine learning and robotics to make an intelligent sprayer which drives through fields to target and spray weeds, leaving the crops intact, working essentially the same way facial recognition technology operates. The program was shown thousands of pictures of weeds until it was able to recognize the difference between the weeds and crops. According to Blue River Technology, it would save up to 90% of what is currently used for herbicides.
As it is pulled through the fields, the cameras analyze each frame and produce a pixel-accurate map of where the crops and weeds are. Once the plants are all identified, each weed and crop is mapped to field locations, while the robot sprays only the weeds.
The program PyTorch is an open source deep learning framework which enables developers and researchers to perform research and seamlessly deploy in production. Facebook AI researcher, Soumith Chintala, was a co-creator of the platform, and while it remains an open-source framework with a number of developers and researchers contributing to its growth, Facebook has said they will be dedicated resources to be a core maintainer of the platform.
It isn’t the first time that AI has been involved in the weed-control business. In 2018, Swiss company EcoRobotix released a solar-powered robot which could work for up to 12 hours detecting and destroying weeds. The company says the robot uses 20 times less herbicide than traditional methods. Also in 2018, the Danish company Agrinavia began developing a system called RoboWeedMaPS that would also automatically recognize and detects weeds in the field.