Researchers at the University of Szeged are using camera-based monitoring technologies to train artificial intelligence systems to identify early signs of plant stress, as part of a project exploring plant resilience. At SZTE’s 13th Innovation Day, Dr. Melinda Tar, senior research fellow at the Institute of Plant Sciences and Environmental Protection at the Faculty of Agriculture, and her research team received Proof of Concept funding for their innovative project.
“Our goal is to detect plant stress before symptoms become visible to the naked eye,” said Dr. Melinda Tar, whose team – including Dr. Vilmos Bilicki, Dr. Edit Mikó, Dr. Péter Jakab, Ingrid Melinda Gyalai, and Flórián Kovács – received Proof of Concept funding for the project.
The team plans to use a range of advanced cameras – including thermal cameras – to capture early plant responses to stress, with the system designed to detect subtle indicators such as the first signs of drought-related wilting.
“We also plan to monitor crops using drones above open fields, while in greenhouse environments the cameras can be mounted, for example, on the carts that move along the rows to perform plant care tasks – allowing these devices to function as fixed yet mobile cameras,” explained Dr. Melinda Tar.
The researcher also described how, at the start of the project, the team constructed plant-growing tents where stress conditions such as drought are deliberately induced. Within this controlled environment, cameras capture images several times a day at predefined intervals, continuously monitoring the crops. These recordings will serve as the basis for training the artificial intelligence system, a process led by Dr. Vilmos Bilicki and his colleagues at SZTE’s Institute of Informatics. Once trained, the system will be able to predict when a plant is under stress, enabling timely intervention and helping to safeguard crop yields.
Dr. Melinda Tar, senior research fellow at the Institute of Plant Sciences and Environmental Protection, Faculty of Agriculture, University of Szeged. Photo: Karina Bartha
At present, the research involves cultivating plants in the controlled tents, where their stress responses are monitored using cameras.
“We have begun testing drought stress in cucumber and maize plants to understand how they respond,” said Dr. Melinda Tar. “In particular, we aim to determine when changes in leaf color, leaf orientation, and other sensor-detectable properties begin to occur before visible stress symptoms appear.” She added that these two plant species are particularly well suited for training artificial intelligence models, as cucumber exhibits rapid and pronounced stress responses, whereas maize shows more gradual and complex reactions, for instance under drought stress or nutrient deficiency.
“These two plant species make it possible to induce symptoms within a relatively short time frame, while also allowing the temporal patterns of stress development to be detected rapidly. By combining data from different cameras and sensors, we can retrospectively assess the condition of plants before visible symptoms emerge. These results provide valuable training data for artificial intelligence models, enabling the predictive identification of developing stress even before it becomes apparent to the naked eye, and supporting earlier and more targeted interventions to prevent yield loss,” the researcher explained. She also noted that the study will later be extended to other plant species and that the method will also be tested under open-field conditions. A further objective is to incorporate plant diseases into the analysis, allowing artificial intelligence models to predict the presence of pathogens or pests in both greenhouse and field environments.
“A clear sign of stress in maize is when the leaves begin to curl upwards – a phenomenon often referred to as ‘leaf rolling.’ If this condition persists, the plant will respond with yield loss, which eventually becomes irreversible. However, this visible symptom is preceded by an earlier phase: as the process begins, the leaf’s internal water pressure changes, the plant starts to wilt, gas exchange is reduced as the stomata close, and the leaves gradually shift from their normal position to a slightly upward orientation. These early signals are not detectable to the human eye, but they can be captured by cameras, allowing artificial intelligence to indicate that stress is developing and that intervention is needed,” explained Dr. Melinda Tar, illustrating the method with a practical example.
The research is particularly promising in that it builds on established camera-based plant monitoring while extending it through the integration of artificial intelligence, which has so far been used in only a limited number of studies. Through this integration, AI can process large volumes of data, learn through models, and thereby enhance the effectiveness of the predictive system.
In the longer term, the researchers plan to make the method available to input suppliers, developers, and plant breeding companies seeking to create new varieties and assess their stress tolerance. “Our goal is to ensure that access is not limited to large-scale farms but extends to medium-sized and smaller producers as well, for example through a dedicated platform or a rental-based model,” said Dr. Melinda Tar.
Original Hungarian article by Helga Balog
Photos by Karina Bartha
