MACHINE VISION METHODS FOR DEFECT SORTING STONEFRUIT

A machine vision system consisting of an illumination chamber, monochr omatic camera with a near infrared band-pass filter, frame grabber, an d microcomputer was developed. Defect segmentation and image processin g methods were developed using histograms of peach and defect pixels a long the rows and columns. The feature extraction techniques were simp le, fast, and well suited for pipeline image processing hardware becau se they used raster scans. Tests were conducted to study the performan ce of the machine vision system at detecting and identifying major def ects. The overall classification error in identifying peaches with maj or defects (cut, scar, bruise, and wormhole) was 28.6%. Errors were pr imarily due to natural variability in the features. Correlation coeffi cients between machine predicted and manually measured defect areas we re 0.75 and 0.72 for bruise and scar, respectively. Surface curvature and manual measurement errors were primary causes of variability.