PEACH DEFECT DETECTION WITH MACHINE VISION

A laboratory machine vision system was developed to detect and identif y surface defects (scar, cuts, bruise, scale, wormhole, and brown rot) on fresh market peaches. Image analysis algorithms were developed for segmenting defect regions in the peach images, and a classifier ident ified the segmented regions as specific defect types. Experimental tes ts were conducted to determine system accuracy in estimating defect ar ea and identifying defect type. Sample correlation coefficients betwee n predicted and manually measured defect areas ranged from 0.56 for sc ale to 0.92 for brown rot. Classifier performance in identifying each segmented region as a member of one of eight classes (scar, stem cavit y, cut, bruise, scale, wormhole, brown rot, and noise) was 31% error r ate for the near-infrared system and 40% for the color system.