Concepts for real-time grading of fruits and vegetables were developed. Three cameras sensing reflectance in the visible region and narrow bands centered at 750 nm and 780 nm were proposed for color evaluation and defect detection. Information from the visible region was intended for use in color grading. The narrow band centered at 780 nm allowed concavity identification with structured illumination, and the region around 750 nm was used for detection of dark spots.

Hardware for fruit handling and image acquisition was developed. The system included a single-lane roller conveyor, interface electronics, cameras, lamps, and laser line generators. All hardware was configured to acquire two near infrared (NIR) images of each fruit and rotate the fruit 180° between each image.

A pipeline image processing algorithm was developed to acquire and analyze the NIR images in real-time. Information from the structured illumination image was used to distinguish between defects and concavities which both appeared as dark spots in the 750 nm image. The total projected area of defects on each fruit was determined, and subsequent classification was based on the defect pixel total.

Apples and peaches were used to evaluate system performance. By adjusting the defect pixel threshold to achieve a 10% error rate on good apples, classification errors for bruise, crack, and cut classes were 72%, 61%, and 66%, respectively. Comparable results for bruise, scar, and cut peach classes were 73%, 36%, and 80%, respectively. Specular reflectance was the major source of error in the apple data.

Acquiring more than two images of each fruit and using more than six lines of structured illumination per fruit would reduce sorting errors. After removing fruit from the original data set in which a defect was not presented to the camera or the concavity was between consecutive lines of illumination, potential sorting efficiencies were determined. W ith a 10% sorting error rate for good classes, the classification error rates for bruise, crack, and cut apple classes were reduced to 60%, 57%, and 56%, respectively. Similarly, error rates for the bruise, scar, and cut peach classes were reduced to 26%, 3%, and 54%, respectively.