SORTA: A YOLO-Based Computer Vision System for Automatic Identification and Grading of Chicken Egg Quality Using ESP32-CAM

Abstract

This paper presents SORTA (Sorting of Chicken Eggs via Real-Time Vision and Automation), an automated egg quality identification and sorting system that integrates YOLOv8-based object detection, Convolutional Neural Network (CNN) classification, and IoT hardware components. The system employs an ESP32-CAM module to capture real-time egg images, which are transmitted via WiFi to a Python-based inference server running YOLOv8n and a TFLite CNN model. A fusion strategy selects the prediction with the higher confidence score, while a load cell sensor provides weight data to finalize the egg grade according to the SNI 3926:2023 standard. The grading outcome (Grade A, B, C, or Unfit) controls a servo-driven robotic arm that physically sorts the eggs on a mini conveyor. Training on 4,005 augmented images over 15 epochs achieved a validation mAP@50 of 98.8%. In real-world tests at a commercial poultry farm using 100 eggs, the combined CNN+YOLO system achieved 70.0% overall accuracy, outperforming standalone YOLO (44.0%) by 26 percentage points. Weight sensing showed an error margin below 2.2%, and the monitoring website displayed all results in real time. The results demonstrate that the multi-modal fusion approach substantially improves sorting reliability compared to single-model baselines, offering a practical and low-cost solution for small-to-medium-scale poultry operations.