AI for Sustainable Poultry: Chicken Detection Through Machine Learning

Background

Modern livestock farming faces immense challenges. With 60 billion chickens raised globally each year, 3 billion fail to reach maturity, and 1.6 billion are rejected due to welfare issues like illness and injuries. This results in significant economic and ethical concerns. Large flock sizes, driven by tight margins and the pursuit of sustainability, leave animals vulnerable to neglect, disease, and poor welfare.

Automating chicken surveillance addresses these issues by providing continuous monitoring to quickly detect and meet animals’ needs. By reducing mortality and optimizing feed usage, such advancements could save 3 million tons of feed annually while promoting ethical farming practices.

Faromatics, a pioneer in Precision Livestock Farming and creator of ChickenBoy—the first ceiling-suspended robot for monitoring broiler chickens—collaborated with Omdena for an innovative Chicken Detection Project. This two-month initiative aimed to enhance ChickenBoy’s capabilities by leveraging machine learning to detect and track individual chickens using video data. This advancement enables farmers to improve both animal welfare and farm productivity.

Objective

The Chicken Detection Project focused on creating a machine learning-based system to detect individual chickens and track their movements frame-by-frame. This system would operate in real-time, enhancing ChickenBoy’s anomaly detection capabilities for chicken behavior.

Approach

The project leveraged cutting-edge computer vision algorithms such as YOLO (You Only Look Once) and R-CNN for chicken detection. The team used videos recorded by ChickenBoy’s onboard 800×600 RGB camera and trained models on labeled image datasets. The developed solution was optimized to run on a Raspberry Pi 4 equipped with a Google Coral Edge-TPU, ensuring real-time performance at a minimum of 5 frames per second (FPS).

Key deliverables included:

• Python code for chicken detection and tracking.
• Trained model parameters and labeled datasets.
• Performance metrics to validate model accuracy and tracking reliability.

Metrics used for evaluation:

1. Detection Quality:
   • Intersection over Union (IoU): Minimum of 0.75.
   • Recall: Minimum of 0.75.
   • Precision: Minimum of 0.90.
2. Tracking Accuracy:
   • Multiple Object Tracking Precision: Error within 5% of image diagonal pixels.

Results and Impact

The Chicken Detection Project delivered a robust detection and tracking system with high precision, recall, and IoU, meeting all performance benchmarks. The solution significantly enhances ChickenBoy’s ability to monitor chicken health and behavior in real-time, empowering farmers to reduce welfare issues, lower mortality rates, and improve productivity. By automating labor-intensive tasks, it supports sustainable and ethical farming practices.

Future Implications

The success of the Chicken Detection Project sets the stage for broader applications in precision farming. Future developments could include integrating health diagnostics, optimizing feeding schedules, and adapting the technology to other livestock. This project not only transforms poultry farming but also paves the way for sustainable food production practices worldwide.