MFSnet: a multi-scale feature screening network for chicken counting in dense environments

Area of Science:

• Computer Vision
• Artificial Intelligence
• Agricultural Technology

Background:

• Machine vision offers efficient chicken counting, but high-density scenarios cause overlapping chickens, hindering accuracy.
• Accurate chicken population counts are crucial for optimizing free-range poultry farming operations.

Purpose of the Study:

• To develop and validate a specialized chicken-counting network (MFSnet) for accurately counting chickens in densely populated free-range environments.
• To improve the efficiency and accuracy of automated chicken population monitoring in agricultural settings.

Main Methods:

• Proposed MFSnet utilizes multi-scale feature extraction and a Feature Screening Module (FSM) for enhanced chicken identification.
• A new dataset, Chicken2023, comprising 550 images with 49,747 labeled chickens was created for training and validation.
• The network's performance was evaluated against existing counting algorithms using metrics like Mean Absolute Error (MAE) and Root Mean Square Error (RMSE).

Main Results:

• MFSnet achieved superior counting performance on the Chicken2023 dataset, with an MAE of 2.7 and RMSE of 3.6.
• Compared to the top-performing existing network, MFSnet demonstrated a 6.25% reduction in MAE and a 6.26% reduction in RMSE.
• The proposed network accurately recognized chicken numbers even in challenging dense environments.

Conclusions:

• MFSnet effectively addresses the challenge of counting overlapping chickens in dense environments.
• The developed network model enhances poultry farming efficiency through accurate automated population monitoring.