Giesco
21st GiESCO International Meeting: ‘A Multidisciplinary Vision towards Sustainable Viticulture’ HYPERSPECTRAL IMAGING AND CNN FOR ON‐THE‐GO, NON‐DESTRUCTIVE ASSESSMENT OF GRAPE COMPOSITION IN THE VINEYARD 1 1 1
Salvador GUTIÉRREZ , Juan FERNÁNDEZ‐NOVALES , Javier TARDÁGUILA , Maria Paz DIAGO 1 1* Instituto de Ciencias de la Vid y del Vino (Universidad de La Rioja, CSIC, Gobierno de La Rioja) Finca La Grajera, Ctra. Burgos Km 6. (26007) Logroño, La Rioja, Spain. *Corresponding author:maria-paz.diago@unirioja.es
Abstract: Context and purpose of the study‐ Knowledge of the spatial‐temporal variation of the grape composition within a vineyard may assist decision making regarding sampling and vineyard management, especially if selective harvest is aimed. To have a truthful picture of the spatial‐temporal dynamics of grape composition evolution during ripening in a vineyard, a huge amount of measurements at different timings and spatial positions are required. Unfortunately, the quick in‐field measurement of a vast number of samples is very hard for simple variables such as total soluble solids (TSS), and impossible in the case of analyzing secondary metabolites, like anthocyanin concentrations. The goal of this study was the in‐field assessment and mapping of the TSS, acidity parameters and anthocyanin concentrations in a Tempranillo (Vitis vinifera L.) vineyard, using non‐destructive, on‐the‐go hyperspectral imaging (HSI). Material and methods‐ HSI of grapevine canopies was carried out using a line‐scan hyperspectral camera working in the Vis‐NIR range (400‐1000 nm) installed in all‐terrain‐vehicle, moving at 5 km/h in a commercial Tempranillo (Vitis vinifera L.) vineyard, under natural illumination conditions. Measurements were carried out at several dates during the ripening period over two consecutive seasons in 2017 and 2018. TSS, titratable acidity (TA), pH and anthocyanin concentrations analyses were also performed using gold standard, wet chemistry methods for model building and validation purposes. Convolutional neural networks (CNN) were applied for the development of regression models. The prediction results from the regression models were used for mapping (using GIS software) the evolution and distribution of grape composition in time–several datesand space–the vineyard plot. Results‐ Prediction models were generated for the different grape composition parameters, yielding 2 determination coefficients (R ) above 0.85 for TSS and TA and ~0.70 for pH and anthocyanin concentrations respectively. The built maps illustrated the seasonal dynamics of TSS and anthocyanin accumulation in the studied vineyard. The obtained results evidenced the potential of hyperspectral imaging acquired on‐the‐go for the non‐destructive, robust and massive assessment of TSS and total anthocyanin contents in grape berries in the vineyard. HIS may become a useful tool for decision‐ making on harvest selection and berry fate for winemaking.
Keywords: spatial‐temporal variability, total soluble solids, berry anthocyanins, Vis‐NIR spectral range, acidity parameters, prediction models.
Abstract: Context and purpose of the study‐ Knowledge of the spatial‐temporal variation of the grape composition within a vineyard may assist decision making regarding sampling and vineyard management, especially if selective harvest is aimed. To have a truthful picture of the spatial‐temporal dynamics of grape composition evolution during ripening in a vineyard, a huge amount of measurements at different timings and spatial positions are required. Unfortunately, the quick in‐field measurement of a vast number of samples is very hard for simple variables such as total soluble solids (TSS), and impossible in the case of analyzing secondary metabolites, like anthocyanin concentrations. The goal of this study was the in‐field assessment and mapping of the TSS, acidity parameters and anthocyanin concentrations in a Tempranillo (Vitis vinifera L.) vineyard, using non‐destructive, on‐the‐go hyperspectral imaging (HSI). Material and methods‐ HSI of grapevine canopies was carried out using a line‐scan hyperspectral camera working in the Vis‐NIR range (400‐1000 nm) installed in all‐terrain‐vehicle, moving at 5 km/h in a commercial Tempranillo (Vitis vinifera L.) vineyard, under natural illumination conditions. Measurements were carried out at several dates during the ripening period over two consecutive seasons in 2017 and 2018. TSS, titratable acidity (TA), pH and anthocyanin concentrations analyses were also performed using gold standard, wet chemistry methods for model building and validation purposes. Convolutional neural networks (CNN) were applied for the development of regression models. The prediction results from the regression models were used for mapping (using GIS software) the evolution and distribution of grape composition in time–several datesand space–the vineyard plot. Results‐ Prediction models were generated for the different grape composition parameters, yielding 2 determination coefficients (R ) above 0.85 for TSS and TA and ~0.70 for pH and anthocyanin concentrations respectively. The built maps illustrated the seasonal dynamics of TSS and anthocyanin accumulation in the studied vineyard. The obtained results evidenced the potential of hyperspectral imaging acquired on‐the‐go for the non‐destructive, robust and massive assessment of TSS and total anthocyanin contents in grape berries in the vineyard. HIS may become a useful tool for decision‐ making on harvest selection and berry fate for winemaking.
Keywords: spatial‐temporal variability, total soluble solids, berry anthocyanins, Vis‐NIR spectral range, acidity parameters, prediction models.