IT giant takes steps to develop technologies for use in agricultural production, bringing data together
St. Louis, Mo. — Microsoft is throwing its hat into the agriculture ring in a big way as it moves toward the commercialization of FarmBeats.
“FarmBeats is a way to get a lot of data from the farm and merge all of that data with the different data that exists in the cloud to come up with unique insights for agriculture,” said Ranveer Chandra, chief scientist of Microsoft Azure.
“We’re trying to make it very inexpensive to get large amounts of data to the cloud.”
The company vision for agriculture devices known as IoT, or Internet of Things devices, which includes such things as soil sensors, drones and telematics from farm equipment, sends information with little or no farmer intervention to Microsoft’s goliath cloud-based artificial-intelligence powered platform, Azure IoT Edge.
FarmBeats combines and analyzes the streams of information to produce actionable insights to farmers, as well as enables them to share their data with the agricultural analytics companies of their choosing.
The research Microsoft is conducting helps solve one of the biggest issues growers using precision agriculture practices face — connectivity — through its TV white space initiative.
TV white spaces are set to connect devices to broadband from across a farm, similar to how a wi-fi router connects devices across a house.
In rural areas when you switch through channels, most of the channels will be static, and it is these channels that will be used to transmit data.
The TV white space system developed by Microsoft scans the TV spectrum to find unused channels and then starts transmitting in those channels.
“While it is transmitting in those spectral, it will keep listening to see if a TV transmission came back. If a TV transmission comes back, it’s going to move to a different TV channel. And if needed, it’s going to adjust its power level and its bandwidth,” Chandra said during his keynote presentation at the InfoAg conference in St. Louis.
He said in tests the system covers a 12-kilometre diameter area and is less prone to interference from trees and crop canopies compared to wi-fi.
“Each channel is six megahertz wide TV channel, if you go outside the U.S. in some places, it’s eight megahertz wide,” Chandra said.
“The way we convert megahertz to megabytes per second is we usually multiply by 3.6. So at six megahertz we’re talking 20 megabytes per second on each channel.”
So if there are 20 channels free at a farm, a user would have access to 400 mbps of unused capacity.
“At 400 mbps, we are not just talking about connecting sensors, we can be connecting drones, cameras, streaming a lot of information you could get from the middle of the farm,” Chandra said.
He said the problem of creating accurate and up-to-date precision maps, such as soil moisture maps, is the need to have many sensors in the field.
To address this problem, Microsoft is researching the use of drones to monitor fields.
“The key thing we built using artificial intelligence techniques, we built a way to combine UAV imagery with ground sensor data,” Chandra said.
“If two parts of the farm look similar not just in RGB but in multispectral and hyperspectral imagery, they are likely to have similar values.”
With only a few soil sensors combined with drone imagery, predictions can be made with a high degree of accuracy, with the use of computer vision algorithms, to construct a 3D point cloud of the farm.
The point cloud is then used to construct digital perspectives of the farm.
“FarmBeats can use drones to expand the sparse sensor data to create summaries for the farm. Then combine them with the ground sensor data to predict what the values are going to be in the rest of the farm,” Chandra said.
“We showed that using this technique is three times more accurate than using existing schemes that don’t use any imagery to start making these predictions.”
As satellites produce higher resolution imagery, their data will also be used with this technique.
Predictions with a high degree of accuracy have been made for soil temperature, moisture and pH by FarmBeats with this method.
A further problem addressed by FarmBeats is poor broadband connection at the farmhouse.
“Most farmers pay for broadband but all they get is one to three mbps of data. To put in perspective every time you fly a drone you could be generating a gigabyte of data in 15 minutes. Sending a gigabyte of data with one to three megabytes of data is going to take forever,” he said.
To solve this problem, FarmBeats uses household computers that use Windows 10, which runs Azure.
“This PC sits in the farmer’s office, does a lot of computing and then only sends the summaries, which we then merge with other data streams, for instance satellite data, soil data and so on,” Chandra said.
“Not all the data is sent to the cloud. The high res version stays on the PC. We also have a web server so when there is no internet connectivity the farmer can log on with the PC and get all the data. All the insights that he can access are on the PC itself.”
He said Azure IoT Edge is composed of isolated components so if other companies want to use this architecture all that is needed is an executable that can work with the different APIs to insert their Ag solution.
FarmBeats is also researching how to extend the battery life of IOT devices used to collect agronomic data.
For instance it connected a Raspberry Pie to solar-powered field sensors that consults with the weather service to see the weather forecast.
“If it’s going to be sunny then everything runs full throttle. If it’s going to be cloudy it slows down the system. Rather than taking an image from the camera multiple times per second, you could take an image every few seconds,” Chandra said.
With this technique, solar-powered field sensors were able to last 30 cloudy days instead of four cloudy days in Seattle.
To extend the battery life of drones, FarmBeats has figured out how to use the wind to help accelerate and decelerate.
“When wind was between five and 15 m.p.h. we got up to 30 percent improvement in drone battery life,” Chandra said.
Research that will be published this fall will show how radio frequencies can be used to power sensors in fields.
“Because these farms have TV white space antennas, which are emitting power, we are looking at can we use that power to actually power those sensors rather than have a battery?”
Another intelligence recently added to FarmBeats is microclimate predictions of soil temperature and soil moisture over five days across a farm.
“We got data from 50 weather stations across Washington state over the last seven years at 15 minute intervals. We use that data to train a machine learning model to start making very local predictions,” Chandra said.
“If you have low-cost sensor at the farm, what we look at is can we use those weather predictions to customize the prediction more accurately to the farm itself?”
A key goal of FarmBeats is to reduce the cost of producing, transmitting, and analyzing farm data.
For instance, the company is developing a method of using a smartphone’s wifi to sense soil moisture, which even small-scale farmers in developing countries will be able to use.
“If you can measure the time of flight of a wi-fi signal, the amount of time it takes for the signal to go from your phone to an antenna in the soil, you can use that to measure soil moisture and soil EC. The reason is that the waves move slower in wet soil,” Chandra said.
“We compared to other expensive commercial sensors and we showed that with both soil moisture and soil EC the results are very accurate.”
Growers will be able to mount such a device on a vehicle to take readings as it crosses a field.
The first versions being developed will require an antenna in the soil. The second version of the wi-fi-based soil moisture sensors won’t require an antenna. It will rely on backscatter technology to get readings.