AI and ML in Smart Farming

There are currently more than 8.1 billion people in the world. It is estimated that by 2050 there will be 9.7 billion. According to the United Nations Food and Agriculture Organization (FAO), feeding that population will require a 70% rise in agricultural production.

Simply scaling recent production techniques to meet future demands is not an option as it is also worth noting that the agricultural industry is the fifth largest consumer of energy and is a significant contributor to greenhouse gas emissions. Let's take a look at the role of AI and ML in smart farming.

Smart Farming

The practice of smart farming is now an industry hot topic. We see the use of new technologies in agriculture and livestock production to increase both quantity and quality. As for the technologies, they include GPS, sensors (that are increasingly smart), the internet of things in agriculture (IoTAg), cloud computing, automation, driverless vehicles, artificial intelligence (AI), and machine learning (ML) in smart farming.

These technologies can work together to create a highly optimized end-to-end system, resulting in increasing levels of autonomy.

A major aspect of smart farming is precision agriculture (PA). It improves crop yield through automated production methods and was first theorized in the 1980s. However, John Deere was the first to put theory into practice by launching its GreenStar Precision Farming System in 1996, introducing GPS guidance and automated steering.

In the early days of PA, experts recognized the importance of data, and the GreenStar brochure carried the tagline “Information is your new crop!” Since then, PA has advanced significantly and is now considered central to smart farming, which focuses on accessing and utilizing precise, real-time data to enhance crop quality and quantity, optimize human labor, and, of course, increase agribusiness profits.

Better data enables faster, more confident decision-making, and it also allows for automating much of the decision-making process, leading to immediate action.

Sensors

Agronomy, the science of soil management and crop production, is key to producing higher yields. For example, a basic indicator of a crop’s health (and growth stage) is its color, including some spectral properties not visible to the human eye.

Satellite imagery can be used to create a variety of spectral indices. Useful ones in crop production include normalized difference vegetation index (NDVI, which compares levels of near infrared [NIR] and visible red light), leaf area index (LAI), and moisture stress index (MSI).

Recent years have seen the increased use of multirotor and fixed-wing UAVs fitted with standard vision and hyperspectral cameras and thermal sensors for monitoring vs. satellites. Spectral properties can also be an indicator of soil health, and useful information comes from electrochemical sensors (measuring pH and nutrient levels) and gamma radiation sensors.

Combined with bigger picture data – such as air and dew point temperatures, wind speed and direction, relative humidity, air pressure, and solar radiation – this information can feed into a connected agriculture ecosystem.

Farmers can use data on crop health to create a prescription map (PM) that details where to apply inputs such as seeds, fertilizers, pesticides, and water. Additionally, weather forecasts, input costs, and the cost and availability of machinery can help guide the top-level decision on when to apply the inputs.

It is important to control input quantities as they have a direct bearing on agribusiness profitability, and several environmental issues as well.

VRTs

Variable rate technologies (VRTs) apply seeds, fertilizers, water and pesticides in optimum quantities and in locations where they are most needed. There are generally two types of VRT, map-based and sensor-based.

Map-based VRT adjusts your product application based on a pre-generated map of your field. Sensor-based VRT doesn’t use a map at all but mounted sensors that measure soil properties or crop characteristics in real-time.

For example, during seeding, the machinery adjusts the seeding rate based on the PM. Farmers must fit a texture-soil-compaction sensing system to machinery to adjust tillage depth, as the map overlooks compaction.

As mentioned, IoTAg is very much part of the smart farming picture. IoTAg-enabled wireless devices will measure conditions in abundance. These devices must be rugged, as they will be exposed to the elements in fields, farm machinery, and livestock monitoring.

Many will also need to be battery-powered as they will be in remote locations. Devices can last over a year in sleep mode, or several years with practical PV cell top-up, using low-power MCUs.

We must address cybersecurity, as IoTAg devices function as nodes on the farm’s network. While the device's data may not be sensitive, it connects to a network with valuable information and automated machinery control.

AI & ML

VRT, powered by data and GNSS guidance, boosts automation with technologies like planter shutoffs and sprayer boom control.

The bigger potential though comes with the addition of artificial intelligence (AI) and machine learning (ML) in smart farming; and the market for AI in agriculture is projected to grow from its current $1.7 billion in 2023 to $4.7 billion by 2028, a CAGR over 23 percent.

Real-time soil compaction measurement requires a simple closed-loop control system with armatures and a strain or displacement measurement method.

Real-time crop and weed distinction requires a computer vision system with ML algorithms to decide on herbicide application. And, if it is a crop, what is its health? Curled leaves and wilting are often an indication of disease.

An ML-enabled vision-based system will be able to detect traces of insects and decide which plants require pesticides. Decisions will also consider factors like soil moisture, as symptoms may not uniquely indicate disease or infestation. Lack of water could also cause wilting, so the ML model must accept different types of input data.

Microchip

As mentioned, low-power MCUs are already used extensively in IoT devices and therefore can be used in IoTAg devices too. AI and ML can be implemented on MCUs, thanks to the Tiny Machine Learning (tinyML) movement. By implementing ML algorithms on MCUs it is possible to provide the edge-processing and decision-making required for many VRT applications.

AI & ML: Smart Farming Data

Smart farming uses data to enhance yield, with AI and ML supporting practices like VRT for improved efficiency and automation.

However, it is the addition of AI and ML into the smart farming agribusiness ecosystem that promises to provide the greatest yield by making on-location decisions and making optimum use of inputs.