RAISIN GRADING BY MACHINE VISION

A machine vision system for grading raisins was developed, including an imaging test stand and image analysis algorithms. Raisin maturity is mainly based on visual features such as degree of wrinkles and shape. The features used in the image analysis were wrinkle edge density, average gradient magnitude, angularity, and elongation. A Bayes classifier was used to separate the raisins into three grades: B or better, C, and substandard. The machine vision system was evaluated by comparing the grading results with those of the industry standard airstream sorter, sight graders, and a panel of sight graders who graded by consensus. Four lots of raisins were used in the experimental tests: high quality, medium quality, low quality natural condition, and low quality reconditioned Panel sight grading was assumed to give the ''true'' grade. The sight grading results were the most accurate in terms of panel sight grading. Machine vision accuracy was comparable to the airstream sorter accuracy. Airstream sorting had the lowest variability in grading results, followed in order of increasing variability by machine vision, panel sight grading, and sight grading. The effect of reconditioning (a process used to improve the grade of low quality raisins by rehydrating and drying) was also examined for the low quality lots. Reconditioning had little effect on the grades assigned by human inspectors. However, the grades assigned by the airstream sorter and the machine vision system improved significantly.