Web-Based Quality Classification System for Domestic Chicken Eggs Using a Hybrid You Only Look Once (YOLO) Method

Abstract

This study develops EggVision AI, a web application for classifying chicken egg quality by integrating YOLOv8 and a MobileNetV2-based Convolutional Neural Network (CNN). YOLOv8 detects egg locations and possible shell cracks, while MobileNetV2 classifies cropped egg images into Good, Fair, and Damaged categories. The classification dataset contains 2,344 images, comprising 1,640 training images and 704 validation images. MobileNetV2 achieved 94.74% accuracy, 94.80% precision, 94.74% recall, and 94.72% F1-score on the validation set. After fine-tuning, YOLOv8 achieved 99.62% precision, 100% recall, 99.50% mAP@0.5, and 77.26% mAP@0.5:0.95 on an evaluation set of 45 images containing 159 objects. These results represent different tasks: quality classification and object detection, respectively. Both models were integrated into a web application that allows users to upload images and view the analysis results for each detected egg.