Grapevine yield estimation using image analysis for the variety Syrah

Mestrado em Engenharia de Viticultura e Enologia (Double Degree) / Instituto Superior de Agronomia. Universidade de Lisboa / Faculdade de Ciências. Universidade do Porto

Yield estimation in recent years is identified as one of more important topics in viticulture because it can lead to more efficiently managed vineyards producing wines of highly quality. Recently, to improve the efficiency of yield estimation, image analysis is becoming an important tool to collect detailed information from the vines regarding the yield. New technologies were developed for yield estimation using a new ground platform, such as VINBOT, using image analysis. This work was done in a vineyard of the “Instituto Superior de Agronomia”, with the aim to estimate the final yield, during the growing cycle 2019 of the variety “Syrah”, using images collected by the VINBOT robot. The images were captured with the RGB-D camera placed on the VINBOT robot in the vineyard and in addition, we obtained laboratory images using an RGB-D manual camera. In this work, the correlation of yield components between ground truth data and images data was evaluated. In addition, it was evaluate the projected bunches area in the images and the percentage of visible bunches not occluded by leaves and by other bunches. It was found a growth factor of bunches on the periods from pea-size to harvest. The efficacy to estimate bunch weight from the projected area was higher at maturation. The relationship between canopy porosity and exposed bunches showed for all the stages high and significant R2 indicating that we can use it to estimate bunches covered by leaves through image analysis. The percentage of visible bunches without the leaves occlusion and bunch occlusion was 29% at pea-size, 21% at veraison and 45% at maturation. It was estimated the final yield at pea-size, with an MA%E of 54%, at veraison and maturation were observed values of MA%E of 7% and 5%, respectively. Our results enable to conclude that the image analysis is an alternative to the traditional way to estimate the yield

N/A