Estimating vineyard grape yield from images

Abstract

Agricultural yield estimation from natural images is a challenging problem to which machine learning can be applied. Convolutional Neural Networks have advanced the state of the art in many machine learning applications such as computer vision, speech recognition and natural language processing. The proposed research uses convolution neural networks to develop models that can estimate the weight of grapes on a vine using an image. Trained and tested with a dataset of 60 images of grape vines, the system manages to achieve a cross-validation yield estimation accuracy of 87%.