AI revives classic microscopy for on-farm soil health testing

The conventional microscope obtains a modern turn – researchers are developing a microscope system powered by AI that could make healthy, cheaper and more accessible soil health tests for farmers and land managers around the world.

Researchers at the University of Texas de San Antonio, United States, have successfully combined low-cost optical microscopy with automatic learning to measure the presence and quantity of fungi in soil samples. Their technology of proof of concept at an early stage is presented at the Goldschmidt conference in Prague on Wednesday, July 9.

The determination of the abundance and diversity of soil fungi can provide valuable information on the health and fertility of the soil, as fungi play an essential role in the biogeochemical nutrient cycle, water retention and plant growth. With this knowledge, farmers can optimize the production and sustainability of crops by making informed decisions on soil management, in particular the application of fertilizers, irrigation and soil.

Optical microscopes are the oldest design of the microscope and have long been used to discover and identify tiny organisms in the soil. Other forms of soil tests use techniques such as phospholipid fatty acid tests and DNA analysis to detect organisms, or to measure the presence of chemicals such as nitrogen, phosphorus and potassium. Although powerful, these modern methods tend to be expensive or simply emphasizing the chemical composition, often overlooking all the biological complexity of soil ecosystems.

Alec Graves from the University of Texas to the San Antonio College of Sciences, the United States, presents research at the Goldschmidt conference this week. He said: “The current forms of biological analysis of the soils are limited, which requires costly laboratory equipment to measure molecular composition or an expert to identify organisms by view using laboratory microscopes. Complete soil tests are not widely accessible to farmers and land directors, which must understand how agricultural practices have an impact on soil health.

“Using automatic learning algorithms and an optical microscope, we create a low-cost solution for soil tests that reduces the required workforce and expertise, while providing a more complete image of soil biology.”

In their design at an early stage, researchers have built and tested an automatic learning algorithm to detect fungal biomass in floor samples, incorporating this into personalized software to label the microscope images. This was created using a data set of several thousand images of mushrooms from South Center-South floors. The software works with only a total magnification of 100x and 400x microscope, available in many microscopes in the affordable terrain, including those found in school laboratories.

“Our technique analyzes a video of a floor sample, dividing this into images and uses a neural network to identify and quantify fungi,” explains Graves. “Our proof of concept can already detect fungal strands in diluted samples and estimate fungal biomass.”

The team is now working to integrate their technique into a mobile robotic platform to detect mushrooms in the ground. The system will combine the collection of samples, microphotography and analysis in a single device. They aim to have a fully developed deployable device ready for tests in the next two years.

Research is led by Professor Saugata Datta, director of the Institute of Water Research Sustainability and Policy of the UTSA. Details of the automatic learning algorithm should be published in a newspaper evaluated by peers later this year.