Developing Edge AI Computer Vision for Smart Poultry Farms Using Deep Learning and HPC

Area of Science:

• Computer Vision
• Artificial Intelligence
• Agricultural Technology

Background:

• Modern poultry farming increasingly relies on digital platforms and IoT devices for operational efficiency.
• Integrating advanced analytics like AI can further enhance monitoring and decision-making processes in livestock management.
• Existing digital farming solutions can be augmented with specialized computer vision capabilities.

Purpose of the Study:

• To develop and deploy deep learning models for chicken detection and segmentation on edge AI devices for poultry farms.
• To enhance existing IoT farming platforms with computer vision functionalities for improved chicken monitoring.
• To enable automated tasks such as chicken counting, mortality detection, and growth assessment.

Main Methods:

• Utilized high-performance computing (HPC) for offline training of deep learning models.
• Employed Faster R-CNN architectures and AutoML for model selection and hyperparameter optimization.
• Deployed trained models on edge AI devices for real-time analysis in poultry farm environments.

Main Results:

• Achieved high accuracy in object detection (AP=85%) and instance segmentation (AP=90%) for chickens.
• Successfully integrated AI models into edge devices for deployment on actual poultry farms.
• Demonstrated the potential for automated monitoring functions including counting and health assessment.

Conclusions:

• The developed AI models show promise for enhancing digital poultry farm platforms.
• Edge AI deployment enables real-time monitoring and data collection for improved farm management.
• Further dataset development and model refinement are necessary for optimal performance and broader application.