Tg. Crowe et Mj. Delwiche, REAL-TIME DEFECT DETECTION IN FRUIT .1. DESIGN CONCEPTS AND DEVELOPMENT OF PROTOTYPE HARDWARE, Transactions of the ASAE, 39(6), 1996, pp. 2299-2308
Three cameras sensing reflectance in the visible region and narrow ban
ds in the near infrared (NIR) region were proposed for simultaneous co
lor evaluation and defect detection. Information from the visible regi
on was intended for use in color grading. A narrow band centered at 78
0 nm allowed concavity identification with structured illumination, an
d a second band centered at 750 nm was used for detection of dark spot
s under diffuse illumination. Based on these concepts, hardware for fr
uit handling and image acquisition was developed. The system included
a single-lane roller conveyor interface electronics, cameras, lamps, a
nd laser line generators. As a first attempt, the system was configure
d to acquire two NIR images of each fruit and rotate the fruit 180 deg
rees between each image. By modeling fruit as spheres, and considering
the hardware configuration, some system parameter guidelines were dev
eloped. For a given fruit and concavity size, theoretical analyses pro
vided estimates for the required number of images and lines of structu
red illumination per fruit. Rotational errors due to variation in frui
t size were also estimated. Considering fruit 74 mm in diameter and ci
rcular concavity regions 5 mm in diameter, five images (with 72 degree
s rotation between images) of each fruit and a total of 8 lines of ill
umination per fruit are required. Selecting roller geometry based on f
ruit size would reduce the effect of fruit size variation on the desir
ed amount of roll.