Grading of Vegetative Cuttings Using Computer Vision

This paper reports on recent progress in the development of system to group populations of vegetative cuttings. The system is required to assign a classification to cuttings such that they appear uniform after a growing period using single two-dimensional monochrome images. We have developed a fast segmentation technique that is able to measure plant features and a supervised learning scheme that learns a mapping from the features to a scalar classification. We report results based on segmentation of over 2000 geranium cuttings. The system is able to process images at 2 Hz and has an accuracy of over 90%. Both metrics exceed human performance.