Home / News
Researchers from LEAF/TERRA, in collaboration with the Instituto Superior Técnico, propose an artificial intelligence model based on U-Net++ for the accurate segmentation of vegetation cover in thermal images of vines captured at an angle.
In the context of precision viticulture, efficient monitoring of the vineyard’s water status is essential to promote sustainable irrigation practices and optimize production quality. In this context, researchers from LEAF – Linking Landscape, Environment, Agriculture And Food, a research unit of Instituto Superior de Agronomia (ISA), School of Agriculture, and members of the TERRA Associate Laboratory, in collaboration with the Instituto Superior Técnico, have developed an innovative solution based on computer vision and deep learning techniques for the advanced analysis of thermal images of grapevines.
LEAF/TERRA researchers Gonçalo Victorino, J. Miguel Costa, and Carlos M. Lopes helped develop a U-Net++ model adapted for segmenting vegetation cover in thermal images of vines captured at an angle, ensuring effective separation of leaves from other visual elements such as trunks, posts, or the sky. This approach allowed for an extremely accurate estimate of canopy temperature, with a root mean square error (RMSE) of only 0.14 °C relative to the manual reference method.
Graphical abstract
The integration of this model enables autonomous, real-time analysis of thermal images, eliminating the need for manual processing or subsequent filtering. In this way, the methodology applied by LEAF/TERRA researchers paves the way for continuous monitoring systems capable of providing reliable data in the field and under variable operating conditions.
Based on the estimated canopy temperature, a machine learning model was trained to assess the water status of the vine. This model achieved a test R² of 0.618, a value very close to that of methods based on soil moisture probes (R² of 0.654), demonstrating that image-based digital solutions can be a robust and scalable alternative for water management in viticulture.
This study is an example of the strategic role of TERRA members in producing applied scientific knowledge, with direct potential to transform agricultural practices and influence public policy.
Share the news:
SPONSOR
DOI 10.54499/LA/P/0092/2020
Copyright Labterra 2024 © By Ação360
DOI 10.54499/LA/P/0092/2020