Computer vision was used to detect early split lesions on the hull of
pistachio nuts. Gray scale intensity profiles were computed across the
width of the nut (perpendicular to the suture along the longitudinal
axis). If the profile crossed an early split lesion, a deep and narrow
valley on the profile at the early split location was observed. profi
les were computed every 0.5 mm along the longitudinal axis of the nut
and the number of adjacent profiles with deep and narrow valleys was r
ecorded. Early split nuts contained a significantly higher count of th
ese adjacent profiles than normal nuts. Combining unhulled nut cross-s
ectional area with the adjacent profile data, 100% of the early split
nuts and 99% of the normal nuts were correctly classified of the total
of 180 nuts tested Two devices were developed to convey and orient un
hulled pistachio nuts for presentation to a computer vision system. On
e device, which operated similarly to an ''air hockey'' table, correct
ly oriented 98% of the early split nuts and 99% of the normal nuts. Th
e other device, which used vibration to convey and orient the nut in a
''V'' trough, correctly oriented 97% of the early split nuts and 98%
of the normal nuts. A total sample size of 270 nuts was used to test e
ach orientation device.