Contextualized Small Target Detection Network for Small Target Goat Face Detection.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Livestock Management

Background:

• Deep learning is increasingly vital for modern livestock management.
• Accurate goat face detection is fundamental for recognition and management systems.
• Existing methods face challenges with low resolution, small targets, and indistinct features.

Purpose of the Study:

• To propose a novel neural network for goat face object detection.
• To address challenges including low image resolution, small targets, and indistinct features.
• To improve the accuracy and effectiveness of goat recognition in livestock settings.

Main Methods:

• Developed a novel neural network architecture specifically for goat face detection.
• Incorporated contextual information and feature-fusion complementation techniques.
• Compared the proposed network against existing object detection models using F1-Score, precision, recall, and average precision.

Main Results:

• Achieved improvements of 8.07% in average precision (AP), 0.06% in precision (P), and 6.8% in recall (R).
• The proposed network demonstrated superior performance in detecting small and indistinct goat faces.
• Effectively mitigated the impact of small targets, enhancing overall detection accuracy.

Conclusions:

• The novel object detection network significantly improves goat face detection capabilities.
• This advancement provides a robust foundation for developing intelligent livestock management systems.
• Highlights the potential of specialized deep learning models in precision agriculture.