GADF-enhanced global-local feature network for wearable sensor-based cow behavior recognition

Area of Science:

• Animal Science
• Biotechnology
• Machine Learning

Background:

• Continuous monitoring of dairy cow behavior is crucial for modern herd management.
• Wearable inertial sensors offer a practical alternative to camera systems for behavior tracking.
• Representing complex motion patterns from raw inertial sensor data presents a significant challenge.

Purpose of the Study:

• To develop an effective cow behavior recognition system using Inertial Measurement Units (IMUs).
• To enhance motion pattern representation by utilizing nose-mounted sensors.
• To improve the accuracy and robustness of automated behavior recognition in dairy cows.

Main Methods:

• Utilized acceleration data from nose-attached IMUs to capture head movement patterns.
• Transformed one-dimensional time-series inertial signals into two-dimensional Gramian Angular Difference Field (GADF) matrices.
• Developed a multi-scale feature modeling network with cross-attention for integrating motion cues.

Main Results:

• Achieved a high overall accuracy of 90.12% for cow behavior recognition.
• Demonstrated robust performance across various behavioral categories.
• Validated the effectiveness of GADF transformation and multi-scale feature modeling.

Conclusions:

• The proposed IMU-based framework provides an effective and robust solution for automated cow behavior recognition.
• Transforming inertial signals into GADF images enhances the representation of temporal correlations.
• Multi-scale feature modeling effectively integrates complementary motion information for improved recognition.