Nicolo' Simonato AgrotechIn recent years, AI in agriculture has emerged as a powerful tool for analyzing datasets related to climate change and crop production. A new study published in Frontiers in Plant Science illustrates how remote sensing technologies, particularly the use of UAVs (uncrewed aerial vehicles), enable efficient plant phenotyping by gathering crucial data on crop characteristics. Researchers from Purdue University, including PhD candidate Claudia Aviles Toledo, showcase a deep learning model capable of predicting crop yield using a combination of environmental, genetic, and remote-sensing data. This innovative approach significantly reduces labor costs traditionally associated with measuring plant traits, such as height and biochemical analysis. The study highlights the role of UAV data acquisition and deep learning models in predicting complex traits in food crops like maize. By incorporating genomic data and environmental factors, researchers can develop algorithms that provide actionable insights for growers. This advancement not only streamlines the data analysis process but also enables plant breeders to make informed decisions about crop varieties tailored to specific conditions. As AI continues to evolve in the agricultural sector, it promises to enhance productivity and sustainability in food production.
Although far removed from that social spotlight, scientists with agricultural, natural, and technical backgrounds are also using AI in their work to analyze datasets to better understand and interpret climate change.
Purdue University Research Highlights AI’s Role in Predicting Crop Yields Through Remote Sensing
In a new paper published in Frontiers in Plant Science, Purdue University geomatics PhD candidate Claudia Aviles Toledo, working with her faculty advisors and co-authors, Melba Crawford and Mitch Tuinstra, demonstrated the capability of a recurrent neural network — a model that teaches computers to process data using long short-term memory — to predict produce yield from some remote sensing technologies and environmental and genetic data.
Plant phenotyping, where the plant characteristics are examined and characterized, can be a labor-intensive task. The labor-intensive and time-consuming tasks of measuring plant height by tape measure, using big handheld equipment to measure reflected light over a number of wavelengths, and pulling and drying specific plants for biochemical analysis are all labor-intensive and costly endeavors. Remote sensing, or gathering these data points from a distance using uncrewed aerial vehicles ( UAVs ) and satellites, is making such field and plant information more accessible.
Tuinstra, the Wickersham Chair of Excellence in Agricultural Research, professor of plant breeding and genetics in the department of agronomy and the science director for Purdue Institute for Plant Sciences, said,” This study highlights how advances in UAV-based data acquisition and processing coupled with deep-learning networks can contribute to prediction of complex traits in food crops like maize”.
Crawford, the Nancy Uridil and Francis Bossu Distinguished Professor in Civil Engineering and a professor of agronomy, gives credit to Aviles Toledo and others who collected genetic data in the field and with distant sensing. The world has seen isolated sensing-based phenotyping reduce labor costs and collect novel plant information that individual senses alone cannot discern through this collaboration and other studies.
Breakthroughs in UAV-Based Data and Deep Learning for Crop Trait Prediction
On robots and UAVs, spectral cameras can now be used to measure the reflectance of light wavelengths outside the apparent spectrum. Laser pulses are released by light detection and range ( LiDAR ) devices, which use them to create maps of the geometric structure of plants known as “point clouds” and measure the time when they reflect back to the sensor.
“Plants tell a story for themselves”, Crawford said. “They react if they are stressed. If they react, you can probably relate that to traits, environmental inputs, management practices such as fertilizer applications, irrigation or pests”.
Aviles Toledo and Crawford work as engineers to create algorithms that compile large datasets and analyze the patterns therein to predict the statistical likelihood of various outcomes, including yield of various hybrids created by plant breeders like Tuinstra. Before any farmer or scout can tell a difference, these algorithms categorize good and stressed crops and provide data on the effectiveness of various management techniques.
Tuinstra contributes to the study with a genetic mindset. Breeders of plants use data to identify genes that influence particular crop characteristics.
This is one of the primary AI systems to incorporate plant genetics into the production cycle in protracted, large-scale experiments, according to Tuinstra. Plant breeders can then determine how various traits respond to various circumstances, which will help them choose characteristics for upcoming more adaptable varieties. Growers can use this to determine which varieties will perform best in their area.
Advancements in Neural Networks thanks to AI in agriculture: Integrating Genomic, Weather, and Remote Sensing Data for Enhanced Crop Predictions
To create this neurological network, we combined genomic data from common corn varieties, economic data from weather stations, and remote-sensing hyperspectral and LiDAR data from corn. A subset of AI that learns from geographic and temporal patterns of data and makes forecasts for the future is represented by this deep-learning model. The network can be updated with limited training data in another geographic location or time previously trained in one location or time, thereby reducing the need for reference data.
Crawford said:”Before, we had used classic machine learning, focused on statistics and mathematics. We couldn’t really use neural networks because we didn’t have the computational power”
Neural networks have links connecting points that finally communicate with each other, similar to chicken wire. This model was modified by Aviles Toledo using much, short-term memory, which allows for both present and past data to be continuously in the forefront of the computer’s “mind” as it forecasts future outcomes. The long short-term memory model, augmented by attention mechanisms, also brings attention to medically crucial times in the growth cycle, including flowering.
Although Crawford claimed that the genomic data is still being processed to extract “aggregated quantitative features,” the weather and remote sensing data are also included in this new architecture. Working with Tuinstra, Crawford’s long-term objective is to add more complicated traits to their dataset and incorporate genetic markers more effectively into the neural network. This will lower labor costs while more properly enabling growers to get the information they need to make wiser decisions regarding their crops and land.
