| ISHS Acta Horticulturae 1395: II International Symposium on Precision Management of Orchards and Vineyards
|
| Authors: | A.E. Abioye, E. Laroche-Pinel, B. Sams, B. Corales, K. Vasquez, V. Cianciola, L. Brillante |
| Keywords: | imaging spectroscopy, precision viticulture, Vitis vinifera L., grape composition, ground-based sensing |
| DOI: | 10.17660/ActaHortic.2024.1395.46 |
| Abstract:
Different parameters can affect berry composition, such as soil characteristics, water availability, or other environmental factors. Knowing the spatial variability of grape composition in a vineyard can help manage growing conditions or plan the harvest. An unmanned terrestrial vehicle (UTV) has been adapted especially for this study to help lift the canopy and expose fruits, and two hyperspectral cameras, using visible/near infrared (Vis/NIR) and near infrared/shortwave infrared (NIR/SWIR) domains, were mounted on the back with GPS systems and halogen lights for night imaging. With this system, we imaged one ‘Merlot’ vineyard located in Madera, California, four times during the growing season and sampled grapes for analysis in the laboratory. It is the first time a camera capturing images in the SWIR domain mounted on a UTV is used to assess grape composition. A total of about 650 samples were collected. To extract the grape signal from the sample plants, images were segmented, extracting the grapes signal with good precision using a classification based on the specific spectral signature of each class presented in the images (grape, leaves, background, mean error class of 2.2%). The reflectance of the grape’s signals was mapped with the grape composition using random forest and partial least square (PLS) regression model. These models’ performance was assessed using 5-fold cross-validation, root mean square error (RMSE), and coefficient of determination (R2). The prediction of grape composition using the reflectance exhibited promising results using a random forest regression model. Prediction metrics were: R2 of 0.82 and RMSE of 0.34 °Brix for total soluble solids, R2 of 0.81 and RMSE of 0.08 for pH, R2 of 0.78 and RMSE of 0.89 g L‑1 for titratable acidity, and R2 of 0.91 and RMSE of 1.29 mg g‑1 berry fresh mass for total anthocyanins. This experiment represented the first time a SWIR camera mounted on a UTV was used to analyze grape composition. Our results show that SWIR images are viable to accurately extract the grape signal for classification purposes as well as assessment of grape composition. | |
Download Adobe Acrobat Reader (free software to read PDF files) | |
