When a Chinese farm owner wanted to set up sustainable practices that could scale to thousands of hectares of land, he turned to a creative tech startup for answers. That company, now TalentCloud, forged ahead with Microsoft Azure Machine Learning and Internet of Things (IoT) technology.
Since its 2012 founding, the company has parlayed those technologies into its Agro-Brain solution. Just as TalentCloud data scientists are more productive as they create the models that power the solution, Chinese farmers are also safely increasing their effectiveness, reducing pollution, and stepping up food safety.
Farmers and consumers have a new reason to be excited and reassured about the quality and safety of their food. While few shoppers may ever hear of Agro-Brain, the solution based on Microsoft Azure is a valuable tool for farmers. It provides full visibility into crop status and recommendations based on the latest agricultural science. That means less pollution and use of pesticides, higher quality and safety, and wins for everyone, from producer to consumer to the environment.
Agro-Brain began when two IT consultants wondered about the links between the most up-to-date agricultural science and the everyday farmer on the ground. Could it be that basing farming practice on data and insights could lead to wins for both farmers and consumers? And perhaps intelligent systems that could guide farmers and automate some farm functions would yield even greater benefits.
Convinced of the value of the idea, Xiaodong Wang (now Chief Executive Officer), and Guangliang Wei (Chief Technology Officer) formed TalentCloud. Resting on four pillars—data collection at the edge, plant science knowledge, farming best practices, and precision execution (circling back to the edge)—the solution combines a host of Azure technologies to promote sustainable agriculture and greater food safety.
Modernizing farming for people and the planet
Learning about traditional farming practices in China was an eye-opener for the founders of TalentCloud. “Lack of scientific knowledge is one of the biggest reasons for food safety crises and environmental pollution,” says Wang. “Traditionally, farmers have followed old ways without understanding the underlying science. China sends agricultural technicians to work with farmers, but there aren’t enough of them. And when in doubt, the farmers overuse chemicals.”
TalentCloud’s Agro-Brain solution helps farmers shift to data-centric practices. It tackles the problem on both the decision-making and operational levels. Agro-Brain uses cloud technology to gather millions of data points and synthesize them via sophisticated machine learning. It advises farmers and interacts with devices to control irrigation and other functions. Looking for a complete, end-to-end solution for Agro-Brain, TalentCloud chose Azure Machine Learning and Azure IoT services.
Harvesting food—and knowledge—from the fields
The Agro-Brain solution creates a closed-loop information system, combining real-time data collection in the field with plant science data that describes the entire life cycle of food crops, including related information about threats like pests, diseases, and growing conditions. The combination gives farmers access to up-to-date, customized crop management recommendations.
TalentCloud uses Microsoft products and services to optimize a four-part system that collects an array of data points—including soil conditions, air temperature, and humidity—from the field via IoT sensors. Camera images of crops and mobile device data supplement quantitative data. Sensors connect directly to Azure IoT Hub, which feeds data into Azure Machine Learning, rapidly training the models created by TalentCloud data scientists.
The automated machine learning capabilities in Azure Machine Learning accelerate model development, creating faster operational plans and forecasts for farmers. Other knowledge—for example, watering and fertilizer recommendations—returns back to the field as operational instructions, activating automated systems through Azure IoT Edge.
Creating better models, faster
The TalentCloud team faced limitations prior to adopting Azure solutions. “Previously, our back-end system could support only about 100,000 sensors simultaneously. After we deployed Azure IoT Hub, we were able to support millions of sensors,” explains Wang. More sensors mean bigger datasets, richer model training, improved model accuracy, and higher data scientist productivity. Even after increasing its sensors by an order of magnitude, TalentCloud didn’t have to add developers. “Using Azure IoT Hub gives our team time to focus on other things,” continues Wang. “We can easily support more sensor types with less lead time, and the system overall is more stable.”
TalentCloud improved productivity on several fronts with IoT Hub. Manually coding the solution would have required daunting effort. “Before, we had to consider scalability and data security,” says Wang. “Azure IoT Hub does so much for us. It takes care of peripheral but critical factors so that we don’t have to.”
The team develops more models faster with the automated machine learning capabilities in Azure. That was key to keeping data scientists productive.
“Agricultural models are very complicated. The algorithms are subject to so many influences. For example, temperature has many dimensions, such as the date and time it was recorded, average temperature, and changes between time periods. The automated machine learning capabilities in Azure Machine Learning save our data scientists from doing a lot of time-consuming work and debugging, which reduces our time to build models from several weeks to a few hours,” says Wang. “We’ve also reduced debugging time by 65 percent by tuning models with Machine Learning hyperparameter tuning capabilities.” Producing many models in a short time also means that Wang’s team can improve the quality of its models faster.
Using other Microsoft tools further enhances productivity and speed. The team uses Microsoft Visual Studio Code to build solutions and deploy them in the cloud. “Visual Studio Code works so well with Azure that it saves our developers a lot of time,” says Wang. “We like the ease and convenience it brings to our development process. We use Visual Studio Code because it is very friendly, very open, and highly interoperative across services.”
The TalentCloud data scientists use ONNX (Open Neural Network Exchange), an open-source format for machine learning models, to standardize their machine learning models and deploy them at the edge. This helps close the loop, sending the accumulated knowledge acquired from sensors and blended with phenological knowledge—plant life cycles and the climatic influence on those cycles—to devices at the edge and putting crop maintenance suggestions into actual practice.
TalentCloud uses the frameworks that work best for each scenario, including PyTorch, TensorFlow, and Caffe. “All our frameworks work well with Azure Machine Learning,” says Wang. “We can export a consistent standard that works well for us because so many tools work seamlessly with our Azure solutions.”