Some say artificial intelligence (AI) is the next agricultural revolution.
Scientists like Dr. Won Suk “Daniel” Lee work to create the innovations that will make this revolution a reality. Lee is a Professor of Precision Agriculture at the University of Florida (UF) in the Department of Agricultural and Biological Engineering. Lee conducted his PhD. research at the University of California-Davis on the automated spraying of tomato fields, but he noted that his interest in the subjects of math, science, and robotics started “many years ago.”
Solving Problems With Science
Farmers today face a multitude of problems, including crop diseases, weeds, and pests that can result in low crop yields. Farm workers may also face long hours of difficult and repetitive labor to plant and harvest crops, and supplies and equipment can be expensive. For Lee, finding solutions to these problems is all in a day’s work.
“You need to define the problem, or the objective based on growers’ input, and then you need to work on how you can achieve those objectives,” Lee said.
In fact, he described problem-solving as his favorite part of his work.
Defining AI
Lee’s research uses AI to address these problems. He defined artificial intelligence, broadly, as automation.
“Our primary purpose is to automate those processes, so that you can do a better estimation of the crop and soil properties,” he said.
Lee uses a form of AI in his research called machine learning, which includes a sub-type called deep learning.
“Artificial intelligence is the broad category. Inside artificial intelligence, there's machine learning. That means that the machine learns by itself based on the data in the previous results. Inside machine learning, there's an area called deep learning. Deep learning mimics human brain function,” he said.
Agricultural scientists use machine learning methods to identify fruit, flowers, pests, and more. They use images of specific objects to train the device to recognize them. The deep learning models allow the device to distinguish between the target object and other items in the surrounding area in an image. The device learns to identify the desired objects under examination through repetition, much like a human.
“Our primary purpose is to automate those processes, so that you can do a better estimation of the crop and soil properties."
A New Kind of Pest Control
AI offers new ways to combat two-spotted spider mites, which are an eight-legged nuisance to farmers and gardeners alike, Lee and his team developed a smartphone app and an independent imaging device for detecting mites and their eggs on strawberry plants.
“[The device is] like a small box that has a camera and a light inside,” Lee explained.
Farmers can use their smartphones to scan a leaf and create an image. The AI in the app can then scan the photo to distinguish between mites, eggs, and other objects in the image. The AI can also count the eggs and mites. Using a Global Positioning System (GPS), the app creates a distribution map of the mites’ locations. The team also kept cost in mind when developing the independent imaging software, with the cost of materials totaling less than $300. Lee noted that an agricultural technology company has already expressed interest in commercializing this technology.
The Color of Progress
Color imaging is another area where Lee hopes to make technology more affordable and easier to use. Scientists currently monitor crops using hyperspectral imaging, a technique that uses sensors to collect information from across a wide spectrum of light. Existing hyperspectral imaging devices can be quite expensive, so Lee’s aim is to create a cheaper and more convenient alternative using standard RGB (red, green, blue) cameras. An image typically contains three different RGB bands. Hyperspectral imaging devices take images over hundreds of different wavelength bands, and different disease symptoms react to different wavelengths. Scientists can then use these wavelengths to identify diseases in crops.
Lee also uses color imaging to detect wetness in strawberry plants. He believes this technique is more convenient than using commercially available wetness sensors, which he said can be “inconvenient to use and calibrate.” The color imaging device consists of a solar panel as a power source and a small computer called a Raspberry Pi. A wetness-detection algorithm programmed into the Raspberry Pi analyzes images taken by RGB cameras to identify wet and dry conditions. Lee’s team uses this information to estimate the probability of strawberry diseases in the near future in the Strawberry Advisory System (SAS).
"You need to define the problem, or the objective based on growers' input, and then you need to work on how you can achieve those objectives."
The possibilities for AI application in agriculture may seem endless, but limitations do exist. “I think a misconception is that artificial intelligence can do anything,” Lee cautioned.
He also noted that AI is still a developing field. Lee said that commercialization of agricultural AI technology may take “a few more years.” Scientists also do not fully understand how deep learning works, though Lee noted that another emerging field called explainable AI exists to provide logical explanations for these deep learning models.
Despite limitations, AI remains a versatile and powerful tool. Lee’s web page provides additional information about his research program. For more information about UF’s research in agricultural applications for AI, visit the UF/IFAS AI website and the Department of Agricultural and Biological Engineering website.
About the Author
Heather Gavigan is a first-year master's student studying Agricultural Education & Communication, with a Communications focus. She also received her bachelor's degree in English from UF. She is currently employed by the UF Department of Agricultural & Biological Engineering as the department's fiscal assistant.