Automation for Enterprise Scale Farming
With a strong focus on exploring emerging technologies set in real-world
use cases, the Concepts Team at Myplanet has been working for most of 2017 on new ways to tackle workplace challenges. Our latest adventure in innovation looked at bringing AI-driven automation to agricultural management. Farming has always been a labour-heavy industry, but automating key aspects of it could vastly improve crop yields, field health, and make the industry as a whole less subject to the high-risk variables of weather and pests, among others.
Are fields receiving too much water? Is there enough nitrogen in the soil? Are there signs of root-worm in this cornfield? The ongoing growth of the global population combined with the shrinking availability of resources has meant increased pressure for accurate predictions in agricultural yield. Multispectral images taken via satellite technology have served the farming industry well in investigating crop health. However, drone-captured images are also demonstrating merit in detecting local issues with a more flexible frequency via highly-detailed photographs.
But the freedom in readily capturing data also comes with its own challenges. Processing and analyzing thousands of images manually can be a time-consuming activity. With the aid of machine learning, how might we envision an intelligent interface system to automate agricultural health inspections, enabled by drone technology?
Concept Overview: Roomba for Farms
Our latest concept looks at how an agricultural business may program a drone—or a fleet of drones—to autonomously inspect a field, learn from data over time, and flag areas of risk based on both historical data and current findings. As an imagined future, this concept examines how IBM micro-services like visual recognition, trade-off analytics and predictive analytics could be the driving technologies changing the way we conduct precision agriculture. Basically, we wanted to explore the idea of an intelligent interface for agricultural automation, a.k.a. create the equivalent of a Roomba for farms.
1. Starting with a training scan
Imagine a scenario where an agricultural corporation is managing a multitude of farms, each with fields that routinely undergo crop rotation. The laborious process of manually labelling each field within the system is replaced by an initial scan that pre-identifies the crops via IBM Watson visual recognition service.
2. Setting up an automated flight schedule
The next step is to create an automated drone schedule, starting with setting an intention for the flight in the form of alert types. Alert types vary from monitoring for irrigation and nitrogen levels, to detecting vegetation-specific pests.
Determining the time and frequency of drone flights requires the consideration of various data streams such as type of crop, variability of weather, seasonality, respecting flight regulations across different regions, and managing between various drone schedules. With such a complex decision matrix of variables to consider, machine learning in the form of tradeoff analytics is used to make optimized system recommendations on when and how often a drone should be sent out.
System recommendations are also made on the best drone model as a match to the task at hand. As an example, a land-based drone might be recommended for more granular, on the ground pest examinations, while an aerial one may be more appropriate to gauge an overview of nitrogen deficiencies.
Once scheduled, the drone in flight can leverage Watson visual recognition to quickly and efficiently begin the process of gathering relevant data.
We had a chance to work with Vishnu Hari, IBM Watson Technical Consultant, to understand how a technology like IBM Watson might be further utilized to improve our approach. In particular, he focused on the potential of a land-based drone used for similar tasks.
“Furthering the vision of minimizing human interaction, a Jetson TX2 chipset attached to the drone can perform neural network computations to support collision avoidance and trail detection. With training over time, the goal is for autonomous inspections to be conducted with minimal human interference, creating significant cost savings. IBM is confident that supplementing the neural network with Watson’s Image Recognition can accelerate the process towards a truly intelligent, trail-sensitive drone navigation.”
-Vishnu Hari, IBM Watson Technical Consultant
3. Flagging potential crop risks
The true value of automation in this scenario is in the insights generated from the data. The arduous process of examining thousands of images is replaced by an expert system that effectively learns from previously collected data to recognize and flag common risks.
“Companies can find value in connecting this system to IBM’s enterprise asset management system Maximo or building maintenance management system Tririga , giving them a comprehensive view over which assets need to be reviewed, and triggering maintenance based on severity.”
-Vishnu Hari, IBM Watson Technical Consultant
Beyond Precision Agriculture
While the focus of this concept has been on the agricultural sector, the applications of drone-enabled inspections combined with AI are far ranging. Drones are finding their place in a multitude of applications— such as the oil and gas industry in examining high temperature flare stacks, or the telco sector in detecting power line deficiencies of cell towers, or in industrial settings in determining roof leakage on commercial buildings. Combined with the addition of cognitive technologies, one can quickly see the benefits of this emerging tech in the form of reduced operational costs, increased safety of workers, and providing time-saving analytics for employees.
Be sure to 👏 and share the post. Interested in the other innovative work we have been doing with Watson? Fascinated by what the shifting landscape of big data can do? You can reach us here to find out more about how we can apply the latest tech to improve your business.