Early yield prediction in different grapevine varieties using computer vision and machine learning

• DOI: 10.1007/s11119-022-09950-y
• Publication Date: 2022-08-08T19:02:42Z
• Authors: Fernando Palacios; Maria P. Diago; Pedro Melo-Pinto; Javier Tardaguila
• Keywords: Vineyard; Vine; Machine Vision; Veraison; Pruning

Abstract Yield assessment is a highly relevant task for the wine industry. The goal of this work was to develop new algorithm early yield prediction in different grapevine varieties using computer vision and machine learning. Vines from six ( Vitis vinifera L.) were photographed mobile platform commercial vineyard at pea-size berry stage. A SegNet architecture employed detect visible berries canopy features. All features used train support vector regression (SVR) models predicting number actual yield. Regarding berries’ detection step, F1-score average 0.72 coefficients determination (R 2 ) above 0.92 achieved all between estimated berries. method yielded values root mean squared error (RMSE) 195 berries, normalized RMSE (NRMSE) 23.83% R 0.79 per vine leave-one-out cross validation method. In terms forecast, correlation its value 0.54 0.87 among NRMSE 16.47% 39.17% while global model (including varieties) had equal 0.83 29.77%. can be predicted up 60 days prior harvest several algorithm.