Classification of Shape of Bell Pepper by Machine Vision System

Abstract

A machine vision method was developed for the objective and rapid classification of the shape of bell (sweet) peppers. Based on the local grading standards, bell peppers are classified in two grades (A and B) based on external color, shape and incidence of bruises or disease. This paper reports on two experiments. In the first experiment, significant features to represent shape properties were selected and a classification test was conducted using these features. As a result, five features (Rt, Cmin, Cr, Hl₂m and Hl₃m) calculated from three images of each sample were deemed to be relevant. The classified results by a neural network classifier showed that the classification accuracy for shape A, B and all samples was 95.12%, 100% and 95.7%, respectively. In the second experiment, the number of side views and the angle between cameras were optimized using a simulation. Simulated results indicated that the best angle was 90°, when only two views were available. And the classification accuracy was going up following the increasing of number of views from 2 to 18. Relatively better results could be achieved when the number of views was four, the classification accuracy for shape B was 80%, and for all samples was 91%. When the number of views was eight, a satisfied classification accuracy could be obtained, 92 % for shape B, 100 % for shape A and 96% for all the samples.