People often think about the internet of things in an industrial or commercial setting, providing sensors and data to a well-defined process or activity. But there’s one fundamental part of our society that can benefit immensely from IoT technologies, one that’s being left behind because it’s not traditionally seen as a technological activity: agriculture and, more specifically, farming.
AI for Earth
Microsoft’s AI for Earth initiative supports a wide selection of academic and nongovernment organization projects with Azure resources and access to its Cognitive Services APIs. The six-month-old program has already showing interesting results, using machine learning to identify areas affected by natural disasters with satellite imagery and using game theory to predict where to find poachers before they can kill endangered animals.
Microsoft has also worked with the Snow Leopard Trust to improve its image-recognition services, training its neural networks on images and streams of highly camouflaged big cats, helping to deliver a comprehensive survey of endangered animal numbers. It’s research that’s led to new thinking about how to process imagery and how to work with historical data.
FarmBeats: IoT and machine learning for agriculture
Those same machine learning techniques and tools are being combined with IoT hardware and research wireless networks to provide precision metrics and near-real-time information to farmers around the world, as part of Microsoft Research’s FarmBeats project. It builds on many of the key Azure IoT technologies I’ve been writing about, along with new partnerships like support of DJI’s drones in Azure’s IoT SDKs.
I have a personal interest in technologies that support farming. Family members were hill farmers in Northern Ireland and, early on in my engineering career, I worked on a project designing sensor technologies for use in managing grain storage silos. So, FarmBeats quickly grabbed my attention, with its aim of instrumenting small and mid-size farms, as well as supporting large-scale industrial agriculture.
The problems facing farmers of all kinds are complex and often specific to their land. What’s important for a vineyard in Northern California is different from a grain farm in Nebraska, or from a sheep farm in England’s Lake District, or from a coffee plantation in Kenya. Each has different experiences of soil, of topography, of climate—and that’s before you even start thinking about their different farming methodologies or the different types of agricultural products they create.
Data-driven agriculture
Much agricultural research has focused on deep instrumentation of farms, providing massive amounts of real-time data that’s hard to process and use. It’s also often disconnected, with information often displayed at the point of collection, where it’s hard to collate and use as part of an overall management strategy or as an input to advisory machine learning services.
FarmBeats’ principal researcher Ranveer Chandra has approached instrumenting farms from the idea of “end-to-end data-driven agriculture.” It’s a concept where farmers are given maps that let them precisely deliver water and nutrients to get the optimum yield, without stressing the land.
One of the problems facing anyone trying to deliver IoT to farms is bandwidth. Farms are big, and even the smallest can stress the most powerful Wi-Fi networking equipment. Although FarmBeats has experimented with LoRa wireless networks, it’s currently using white spaces technologies, taking advantage of abandoned TV transmission bands. These support significant amounts of data over long ranges, even in challenging wireless environments like orchards and hilly areas. With 180MHz of available spectrum in the continental US, there’s capacity for 500Mbps across most farms, more than enough to handle the many data sources FarmBeats uses. Chandra suggests that most farms will generate around 5Gb of data a day, more than enough to saturate traditional connectivity models.
While connectivity on the farm can be solved with novel wireless solutions, things are more complex when connecting a farm to the cloud, even in developed economies where rural broadband provision significantly lags urban capacity. The average US farm has less than 3Mbps of broadband connection, so some form of local processing is going to be essential to handle the workloads to manage data from in-field sensors, drones, and static balloons. Processing aerial imagery is key to much of FarmBeats’s operation, because multispectral images can deliver a lot of information about plant growth and coverage, as well as helping pinpoint concentrations of roaming herds.
Architecting an intelligent edge
FarmBeats takes advantage of several Azure technologies, to reduce the system’s bandwidth requirements and work within a farm’s restricted network capabilities. By running IoT Edge on a local PC, much of Azure’s IoT functionality can be brought down from the cloud to the farm. The resulting IoT gateway handles sensor interfaces, using MQTT to deliver messages from low-bandwidth, relatively simple devices, FTP to bring in images from drones, and a video processor for live camera images. Local processing then works with the data from various sources to generate maps and other video and image data.
Because Azure IoT Edge is a container that encapsulates elements of Azure functionality, it can also host machine learning models that have been trained in the cloud Azure ML service. A web server in the FarmBeats gateway displays results from the various services, indicating when and where to deliver fertilizer, or detailing irrigation plans for the next few days based on weather data and on soil moisture readings. It’s also able to work with Azure IoT’s new drone APIs to plan flight paths and download imagery from drones in-flight.
The complete system isn’t completely isolated, though it can easily run offline. Some elements, like drone flight plans and device configurations,can be synchronized with Azure and managed via IoT Hub. Chandra notes that “there’s unique data here on the edge that is never shipped to the cloud; we’ve flipped the cloud architecture.”
Containerizing the digital farm
What’s fascinating about FarmBeats is how complex an IoT infrastructure it implements, and how little it costs to deploy. White spaces radio is a well-known technology that’s easy to implement with software-defined radios, while the gateway hardware can be based around a standard Windows 10 PC running easy-to-deploy containers for the various IoT Edge components and services. Using containers also makes it easier to deploy custom software to each farm, with different modules that can be mixed as required.
FarmBeats may still be a research project, but it goes a long way to show how valuable IoT technologies can be to a wide selection of industries and services, well beyond its easy automation wins. There’s a tradition of farming as a craft, and a new generation of hacker farmers that are building and deploying their own IoT systems. But they’ve been doing it on their own.
FarmBeats is aiming to provide digital transformation for an entire industry, across the entire planet.