Occlusion-Aware Caged Chicken Detection Based on Multi-Scale Edge Information Extractor and Context Fusion.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Agricultural Technology

Background:

• Caged chicken coop environments present challenges for accurate chicken detection due to uneven illumination and severe occlusion.
• Existing detection models often yield unsatisfactory accuracy in these complex agricultural settings.

Purpose of the Study:

• To develop an occlusion-aware caged chicken detection model that addresses poor illumination and severe occlusion.
• To improve the accuracy and robustness of chicken detection in real-world production environments.

Main Methods:

• Constructed a specialized image dataset using a head and neck co-annotation method and multi-stage co-enhancement.
• Proposed Chicken-YOLO, incorporating a multi-scale edge information extractor (MSEIExtractor), context-guided downsampling (CGDown), and a detection head with a multi-scale separation and enhancement attention module (DHMSEAM).

Main Results:

• Chicken-YOLO achieved superior detection performance, with 1.7% and 1.6% improvements in mAP50 and mAP50:95 over the YOLO11n baseline.
• The model demonstrated higher mAP50 than YOLO11s with significantly reduced parameters (58.8%) and computational cost (42.3%).
• Performance gains were particularly notable on specialized test sets for poor illumination (3.0% mAP50 increase) and occlusion (1.8% mAP50 increase).

Conclusions:

• Chicken-YOLO effectively enhances target capture under poor illumination and maintains contour continuity in occlusion cases.
• The model exhibits robustness against complex disturbances in caged chicken environments.
• The proposed methods offer a significant advancement for automated chicken monitoring in agriculture.