Tomato fruit maturity estimation from RGB images

We introduce an approach to estimate tomato fruit maturity from RGB images based on color and size of the tomato fruit. We show that estimating the fruit maturity from the color alone is inaccurate during the first half of the tomato maturing period. We have trained a neural network to estimate the fruit maturity level based on a set of more than one thousand tomato images annotated with position, color and size labels. We use an unsupervised instance segmentation approach which detects individual tomatoes.