Development of a real-time size classification and defect detection machine vision software for mango (Mangifera indica L. ev.'Carabao')

Date

6-2016

Degree

Bachelor of Science in Agricultural and Biosystems Engineering

Major Course

Major in Agricultural Power and Machinery Engineering

College

College of Engineering and Agro-Industrial Technology (CEAT)

Adviser/Committee Chair

Delfin C. Suministrado

Restrictions

Restricted: Not available to the general public. Access is available only after consultation with author/thesis adviser and only to those bound by the confidentiality agreement.

Abstract

'Carabao' mangoes are among the most important exports of the Philippines, ranking third only behind pineapple and banana. However, in spite of its importance, it suffers significant losses during the post-harvest stage due to manual quality control and inconsistencies with human visual inspection. In this study, machine vision was used to classify mangoes according to its size and defects based on PNS/BAFPS 13:2004 and modified exporter guidelines. A calibration run was performed using mango samples harvested 120 days after flower induction, and grayscale coefficients determined for detection of detects from the peel are -0.7, 0.95, and 0 for red, blue, and green respectively. Weight model determined during the calibration run is based on the mean projected area with R² of 95%. During validation, the model had an average percent error of 1.70% and undersized the mangoes by an average of 1.34%. Size classification according to weight had a 92.85% success rate. Market classification according to defects has a 76.19% success rate. Overall success rate of the software is 69.05%. Analysis time for a mango side is 310 ms.

Language

English

Location

UPLB College of Engineering and Agro-Industrial Technology (CEAT)

Call Number

LG 993.5 2016 A2 /J35

Document Type

Thesis

This document is currently not available here.

Share

COinS