Fruit sizing in-field using a mobile app

A mobile device based application was developed for manual in-field fruit sizing of mango (Mangifera indica) using machine vision. Fruit on tree were imaged against a hand-held A4 blue background with a scale included, with segmentation of fruit and scale using the b* value of CIE L*a*b* colour space. To correct for depth of field difference between the background and the fruit, the distance between camera and background was estimated using a pin-hole camera model, and the half thickness of the fruit was estimated from allometrics with fruit length and width. RMSE of 3.7 mm on length and 3.6 mm on width was achieved. Fruit allometrics were also used to convert lineal estimates to fruit weight, with a RMSE of 65.9 g achieved. © 2019 International Society for Horticultural Science. All rights reserved.