Bloom intensity corresponds to the number of flowers present in orchards during the early growing season. In fruit orchards, bloom intensity and climate are crucial to guide processes of thinning and pruning, which directly affect fruit load, size, coloration, and taste. Despite its importance, bloom intensity is still typically estimated by means of visually inspecting a random sample of trees within an orchard. Such process is very imprecise and time-consuming, which contribute to increased labor costs. Several automated computer vision systems have been proposed to estimate bloom intensity. However, existing methods are mostly species-specific and present overall performance still far from satisfactory even in relatively controlled environments (e.g., at night with artificial illumination).
This project aims to estimate blooming intensity by applying novel deep learning image understanding methods combined with recently developed multi-target tracking algorithms to analyze sequences of images of apple trees collected by a robotic platform at a USDA orchard under natural conditions.