Lightweight Sheep Head Detection and Dynamic Counting Method Based on Neural Network.

Area of Science:

• Computer Vision
• Machine Learning
• Artificial Intelligence

Background:

• Accurate target detection is crucial for precise counting in various applications.
• Existing methods may face challenges in speed and parameter efficiency for real-time animal monitoring.

Purpose of the Study:

• To develop an efficient and accurate deep learning model for sheep's head detection.
• To enhance target counting capabilities in intelligent animal husbandry systems.

Main Methods:

• Introduction of the Sheep's Head-Single Shot MultiBox Detector (SH-SSD) model.
• Integration of Triple Attention mechanism in MobileNetV3 backbone for parameter reduction and speed enhancement.
• Utilizing Spatial Pyramid Pooling and Triple Attention Bottleneck in the network's neck for improved feature extraction.
• Employing a Decoupled Head module in the network's head for optimized prediction.

Main Results:

• The SH-SSD model achieved an average detection accuracy of 96.11% for sheep's heads.
• Demonstrated significant improvements in detection metrics and a reduction in model parameters.
• Achieved high-precision quantitative statistics when combined with the DeepSort tracking algorithm.

Conclusions:

• The SH-SSD model offers a robust solution for sheep's head detection with high accuracy and efficiency.
• The model's performance and deployment simplicity provide valuable technical support for intelligent animal husbandry.
• SH-SSD contributes to advancements in automated livestock monitoring and management.