Automatic chick cough detection system based on improved audio spectrogram convolutional transformer neural network
Bowen Cai1,2, Bo Zhou2,3, Xiangshuai Kong3
1School of Environmental, Tsinghua University, Beijing, China.
Frontiers in Veterinary Science
|May 11, 2026
Summary
This study introduces an acoustic detection system for chicken coughing (ASCT-CC) to monitor poultry health. The system uses an improved audio spectrogram transformer (AST) for early detection of respiratory diseases, achieving over 92% accuracy.
Area of Science:
- Animal Science
- Computer Science
- Veterinary Medicine
Background:
- Respiratory diseases pose a significant threat to poultry health, especially in high-density farming environments.
- Rapid disease spread can lead to large-scale infections and economic losses.
- Early detection and intervention are crucial for preventing disease outbreaks in poultry farms.
Purpose of the Study:
- To develop an effective software system for monitoring and providing early warnings of respiratory diseases in chickens.
- To create an acoustic detection system for chicken coughing (ASCT-CC) suitable for real-world poultry farming conditions.
- To enhance poultry health and prevent disease spread through intelligent monitoring.
Main Methods:
- An improved audio spectrogram transformer (AST) architecture with a hybrid convolutional-transformer backbone was employed.
- Local multi-head attention replaced global attention for enhanced local acoustic information capture and noise robustness.
- A two-branch co-learning structure, focal loss, and a connectionist temporal classification (CTC) decoder were utilized for accurate cough event identification and localization.
- The system was deployed on edge computing devices using TensorRT acceleration and INT8 quantization for low-latency, real-time performance.
Main Results:
- The proposed ASCT-CC system achieved a mean average precision (mAP) of 92.86% during training.
- An independent test set yielded an identification rate of 92.11% for chicken cough events.
- The system demonstrated a low inference time of approximately 200 milliseconds, enabling 24/7 real-time monitoring.
Conclusions:
- The ASCT-CC system offers effective technical support for the early detection and intelligent control of respiratory diseases in poultry.
- The developed system provides multi-level early warning capabilities, crucial for timely disease management.
- The system's efficiency and accuracy contribute to enhancing chicken health and farm biosecurity.
