We describe a novel system for grading oranges into three quality band
s, according to their surface characteristics. The system is designed
to process fruit with a wide range of size (55-100 mm), shape (spheric
al to highly eccentric), surface coloration and defect markings. This
application requires both high throughput (5-10 oranges per second) an
d complex pattern recognition. The grading is achieved by simultaneous
ly imaging each item of fruit from six orthogonal directions as it is
propelled through an inspection chamber. In order to achieve the requi
red throughput, the system contains state-of-the-art processing hardwa
re, a novel mechanical design, and three separate algorithmic componen
ts. One of the key improvements in this system is a method for recogni
sing the point of stem attachment (the calyx) so that it can be distin
guished from defects. A neural network classifier on rotation invarian
t transformations (Zernike moments) is used to recognise the radial co
lour variation that is shown to be a reliable signature of the stem re
gion. The succession of oranges processed by the machine constitute a
pipeline, so time saved in the processing of defect free oranges is us
ed to provide additional time for other oranges. Initial results are p
resented from a performance analysis of this system.