A Recognition Method of Ewe Estrus Crawling Behavior Based on Multi-Target Detection Layer Neural Network.

Area of Science:

• Agricultural Engineering
• Computer Vision
• Animal Science

Background:

• Estrus detection in sheep farming faces challenges with labor-intensive manual methods and stress-inducing contact sensors.
• Accurate estrus detection is crucial for efficient meat sheep production and animal welfare.

Purpose of the Study:

• To develop an automated, accurate, and non-invasive method for recognizing ewe estrus behavior in large-scale farming.
• To enhance the precision and efficiency of estrus detection using a novel neural network model.

Main Methods:

• Proposed a multi-objective detection layer neural network for ewe estrus crawling behavior recognition.
• Utilized K-means++ clustering for optimal anchor box sizing and added a 104 × 104 detection layer to improve small target recognition.
• Incorporated residual units to prevent feature loss and maintain aspect ratios for increased detection accuracy.

Main Results:

• Achieved 98.56% recognition precision, 98.04% recall, and 98% F1 score.
• Obtained a mean Average Precision (mAP) of 99.78% with a processing speed of 41 frames per second (FPS).
• The model size was 276 MB, demonstrating efficient deployment potential.

Conclusions:

• The proposed neural network model effectively addresses limitations in current estrus detection methods for sheep farming.
• The system provides accurate and real-time recognition of ewe estrus behavior, suitable for large-scale agricultural applications.
• This AI-driven solution offers a promising advancement for improving sheep farming management and animal welfare.