DFCCNet: A Dense Flock of Chickens Counting Network Based on Density Map Regression.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Agricultural Technology

Background:

• Automated chicken counting is crucial for modern poultry management.
• Existing methods struggle with challenges like poor lighting, irregular sizes, and dense flocks, leading to inaccurate and unstable counts.
• There is a need for robust automated counting solutions in dense farming environments.

Purpose of the Study:

• To propose a novel deep learning network, the dense flock of chickens counting network (DFCCNet), for accurate and stable automated chicken counting.
• To address the limitations of existing methods in handling dense flocks and challenging environmental conditions.
• To provide a benchmark dataset for dense chicken flock counting research.

Main Methods:

• Developed DFCCNet based on density map regression, incorporating feature fusion from different levels to enhance chicken-background distinction.
• Implemented multi-scaling techniques to detect and count chickens across various sizes, improving accuracy and performance.
• Utilized feature convolution kernels to extract precise target information, mitigating occlusion effects for reliable counting.

Main Results:

• The DFCCNet achieved robust performance across three density levels with mean absolute errors of 4.26, 9.85, and 19.17.
• The method demonstrated a processing speed of 16.15 frames per second (FPS).
• A new benchmark dataset, Dense-Chicken, comprising 600 images with 99,916 labeled chickens, was created and made available.

Conclusions:

• DFCCNet offers an automatic, fast, and accurate solution for counting chickens in dense agricultural settings.
• The network's ability to handle challenging conditions and its high processing speed make it suitable for real-world applications.
• DFCCNet can be integrated into handheld devices, facilitating practical application in agricultural engineering and poultry management.