Edge-intelligent lightweight vision system for perspective-corrected egg-cage matching in cage-reared ducks

Area of Science:

• Agricultural Engineering
• Computer Vision
• Animal Science

Background:

• Individual egg production monitoring is crucial for optimizing breeding management in cage-reared ducks.
• Existing multi-sensor systems are costly, and fixed cameras have limited coverage, necessitating innovative solutions.

Purpose of the Study:

• To develop a cost-effective, mobile camera-based system for automated individual egg production monitoring in laying ducks.
• To enhance the accuracy and efficiency of duck egg detection, cage identification, and egg-cage matching.

Main Methods:

• Developed a Lightweight Duck Egg and QR Code Detection (LDEQ-OD) model based on an improved YOLOv11 framework with a Dual Detection Head (DH) and C3-DDF module.
• Employed OC-SORT for multi-object tracking and a Cascade Robust QR Code Decoding (CRQD) algorithm for enhanced QR code recognition.
• Implemented a Minimum Aspect Ratio Deviation (MARD) dynamic matching strategy to compensate for geometric distortions.

Main Results:

• The LDEQ-OD model achieved high accuracy (99.6% precision, 99.3% recall, 95.3% mAP@0.5:0.95) with fast inference (59.2 ms) on edge devices.
• CRQD significantly improved the Code Identification Rate from 72.7% to 99.3% compared to traditional decoders.
• The MARD strategy resulted in high Egg-Cage Matching Accuracy (ECMA) of 98.3% with a low Mean Absolute Error (MAE) of 0.017 eggs per cage.

Conclusions:

• The proposed mobile camera monitoring system enables intelligent, real-time tracking of individual egg production in cage-reared ducks.
• This technology supports precision breeding management by providing accurate and efficient data acquisition.
• The system demonstrates robust performance under various conditions, including motion blur, uneven illumination, and complex geometric perspectives.