A Video Mosaicing-Based Sensing Method for Chicken Behavior Recognition on Edge Computing Devices

Area of Science:

• Agricultural Science
• Computer Science
• Animal Science

Background:

• Accurate chicken behavior recognition is vital for animal welfare, health monitoring, and efficient farm management.
• Existing methods often require significant computational resources, limiting their application on edge devices.
• Developing efficient, on-device behavior analysis systems is crucial for real-time insights in poultry farming.

Purpose of the Study:

• To introduce an effective technique for chicken behavior recognition on edge computing devices.
• To leverage video sensing mosaicing and deep learning for accurate, real-time behavior identification.
• To demonstrate the feasibility of on-device poultry behavior analysis for practical applications.

Main Methods:

• A novel approach combining video sensing mosaicing with deep learning models (specifically MobileNetV2).
• Implementation on edge computing devices for localized video data processing.
• Training and validation using video datasets of chickens exhibiting distinct behaviors.

Main Results:

• The proposed method achieved a classification accuracy of 79.61% for three distinct chicken behaviors.
• Successful deployment and operation on edge computing devices were demonstrated.
• The system shows high potential for real-time chicken behavior analysis.

Conclusions:

• The video sensing mosaicing and deep learning technique is effective for chicken behavior recognition on edge devices.
• This approach offers a promising solution for enhancing poultry welfare, health, and farm management.
• Further research into identifying a wider range of behaviors will improve the comprehensiveness of poultry behavior analysis.