Guangxi Mandarin Orange Classification Processes Using Machine Vision

Abstract

This thesis presents a methodology for classifying Mandarin orange grades according to Chinese criteria with a computer vision system that combines hardware and software elements. A mechanical roller-flipping apparatus alters the Mandarin orange's orientation in multiple positions. A machine vision system subsequently captures thirty images of mandarin orange skin from multiple angles and utilizes various processing techniques, including image collection, blob analysis, preprocessing, segmentation, and feature extraction. The classification of oranges entails utilizing techniques such as morphology, median filtering, and the Fourier transform to find and analyze pixels indicative of surface flaws. For classification and grading purposes, we convert the defective pixels into their diameter and area. The experiment demonstrates the application of diameter and rectangular areas in the classification of Mandarin oranges into three distinct categories: Special Grade, Grade 1, and Grade 2. The grade 3 classification can be established by measuring the diameter and calculating the percentage of the defective area in the orange peel.

-