Computer vision for automated inspection of potatoes

Inspection of fruits and vegetables by humans is labor intensive work. A machine vision system was developed to assist humans in automated inspection of potatoes. This system was designed to inspect potatoes for size, color, shape, and surface blemishes. A 24 bit color imaging system with multiple RGB color cameras and a 386-based computer was set up for capabilities of both color and monochrome vision. Machine vision methodologies and algorithms were developed for the inspection factors.; A method of using the hue-saturation-intensity (HSI) color scheme was studied for potato green detection. By using hue histogram features and multivariate discriminant analysis, the system achieved over 90% accuracy for green classification. Various vision and image processing techniques for shape descriptions were investigated for separation of a variety of shapes of potatoes. An effective method of using harmonics based on a Fourier radial boundary transform was developed for shape separation of potatoes. Based on experiments, the system achieved 83.3% human-machine agreement for shape grading of potatoes. Two dimensional invariant moments, one dimensional boundary moments, and other imaging techniques for shape descriptions showed inadequate power for shape separation of potatoes. Surface defects such as mechanical injury, diseases, scab and other blemishes were investigated for vision recognition toward potato grading. Based on the filtering, segmentation and statistical approaches, the degree of blemish was determined from the percentage of defect areas on the surface of a potato. A rule-supervised architecture was designed to control the modularized inspection process for combined grading. This allowed cooperative inspection operations for a fast and flexible inspection process.; Techniques and algorithms developed in this research provide a prototype for the implementation of automating an inspection system for produce grading.