Grape size detection and online gradation based on machine vision

This research investigated the size detecting and online grading of Red Globe grapes using images of entire cases, rather than individual grapes. Method of ellipse fitting based on iterative least median squares was proposed and the process of grape grading includes the following four steps: stem removal from the RGB and NIR images collected by the 2-CCD camera; edge extraction by multiple methods of edge detection, image binarization, morphological processing, et al.; size determination of individual grapes by using image segmentation and ellipse fitting to calculate short axis length; Finally, grading based on the 15% downgrade principle, this means that if the case contains more than 15% of multiple grades, then the case is re-evaluated. Thirty-eight cases of Red Globe grapes were graded using these methods and 35 cases were correctly graded with an accuracy rate reaching 92.1%. The results showed that the accuracy and speed meet the requirements of grape automatic online detection.