Development of a Computer Vision-Based System for Multiple Detection and Quality Grading of Carabao Mangoes

Abstract

Manual grading of Carabao mangoes remains a significant challenge in postharvest operations because it is labor-intensive, time-consuming, and prone to inconsistencies caused by human judgment. However, traditional grading methods rely heavily on manual inspection, which is labor-intensive, subjective, and prone to inconsistencies. This study developed a computer vision-based system for the multiple detection and quality grading of Carabao mangoes aligned with the Philippine National Standard (PNS/BAFS 13:2024). The developed system integrates a controlled imaging chamber, a three-camera image-acquisition setup, and a YOLOv11 object detection model to evaluate mango ripeness, size, and surface defects automatically. A dataset consisting of 1,650 original annotated images was prepared from locally acquired Carabao mango images and collected from public datasets, and further expanded through augmentation to produce a total dataset of 3,148 images for model development and evaluation. Experimental results demonstrated a mean Average Precision (mAP@0.50) of 96.8% and an optimal F1-score of 0.92. Feature extraction modules achieved classification accuracies of 94.0%, 92.0%, and 94.0% for ripeness, size estimation, and surface defect detection, respectively. Comparison with expert grading yielded a 92.0% agreement rate and a Cohen’s Kappa coefficient of 0.889. The findings indicate that the system provides a practical, accurate, and scalable solution for automated mango quality assessment in postharvest operations.