Recognition of Cattle's Feeding Behaviors Using Noseband Pressure Sensor With Machine Learning

Area of Science:

• Animal Science
• Biotechnology
• Machine Learning

Background:

• Automatic monitoring of cattle feeding behaviors (rumination, eating) is crucial for animal health, growth, and disease detection.
• Noseband pressure sensors accurately capture jaw movements indicative of chewing but face challenges with variable initial pressure.
• Consistent initial pressure is difficult to maintain, complicating the processing of feeding behavior data from sensors.

Purpose of the Study:

• To develop a machine learning approach to eliminate the influence of initial pressure on identifying rumination and eating behaviors in cattle.
• To enhance the accuracy of cattle behavior recognition using noseband pressure sensor data.
• To contribute to the standardized application and promotion of noseband pressure sensors in livestock management.

Main Methods:

• Utilized local slope analysis to capture local data variations from pressure sensor readings.
• Applied Fast Fourier Transform (FFT) to extract frequency-domain features from the sensor data.
• Employed the Extreme Gradient Boosting (XGB) algorithm for classifying rumination and eating behaviors based on extracted features.

Main Results:

• The proposed method, combining local slope and frequency-domain features, achieved an F1 score of 0.96.
• Recognition accuracy for both rumination and eating behaviors reached 0.966.
• The approach demonstrated improved behavior recognition accuracy compared to common data processing and time-domain methods.

Conclusions:

• The developed machine learning approach effectively identifies cattle feeding behaviors despite initial pressure variations.
• The combination of local slope and FFT-based frequency-domain features offers a robust method for behavior analysis.
• This work supports the wider adoption and reliable use of noseband pressure sensors for livestock monitoring.