An Intelligent System for Pigeon Egg Management: Integrating a Novel Lightweight YOLO Model and Multi-Frame Fusion...

Area of Science:

• Agricultural Technology
• Computer Vision
• Machine Learning

Background:

• Pigeon egg farming faces challenges with high breakage rates and labor costs.
• Current manual management methods are inefficient and prone to errors.

Purpose of the Study:

• To develop an intelligent system for automated pigeon egg recognition and positioning.
• To reduce breakage rates and labor costs in pigeon egg farming.
• To improve the efficiency and accuracy of pigeon egg management.

Main Methods:

• Utilized an improved YOLOv12n object detection algorithm for pigeon egg recognition.
• Employed OpenCV barcode recognition technology for positional information extraction.
• Collected real-time video data, annotated images, and trained a custom dataset (1500 training, 215 validation, 215 test images).
• Implemented data augmentation and a majority voting method to enhance detection accuracy.

Main Results:

• The YOLOv12n-pg model achieved a low computational load (4.9 GFLOPS), 1.56 M parameters, and a small model size (3.5 MB).
• Inference tests yielded high performance metrics: 99.4% mAP50 and 83.6% mAP50-95.
• The system successfully recorded "cage location-egg count" data, demonstrating reduced missed detection rates.

Conclusions:

• The developed system offers an effective solution for automated pigeon egg management.
• It significantly improves recognition performance, efficiency, and accuracy over manual methods.
• Provides a foundation for future advancements in intelligent poultry farming systems.