Behavior Classification and Analysis of Grazing Sheep on Pasture with Different Sward Surface Heights Using Machine...

Area of Science:

• Animal Science
• Agricultural Engineering
• Machine Learning

Background:

• Sheep health and productivity monitoring benefits from automated behavior classification.
• Wearable Inertial Measurement Unit (IMU) sensors offer a promising approach for sheep behavior recognition.
• Current methods often lack consensus on data processing and classification, with limited application to continuous behavior data.

Purpose of the Study:

• To evaluate multiple combinations of algorithms, time windows, and sensor data for classifying continuous grazing sheep behavior.
• To identify the optimal configuration for accurate sheep behavior recognition using IMU sensor data.
• To assess the impact of sward surface height (SSH) on sheep behavior patterns.

Main Methods:

• Compared extreme learning machine (ELM), AdaBoost, and stacking algorithms.
• Tested time windows of 3, 5, and 11 seconds.
• Utilized three-axis accelerometer (T-acc), three-axis gyroscope (T-gyr), and combined T-acc and T-gyr data for classification.
• Applied the best-performing model to continuous behavior classification of grazing sheep over 67.5 hours across varying SSH.

Main Results:

• The optimal combination was a stacking model with a 3-second time window using both T-acc and T-gyr data, achieving 87.8% accuracy and a Kappa value of 0.836.
• Continuous monitoring revealed distinct behavioral patterns related to SSH: sheep spent the most time walking on short grass, grazing on medium grass, and resting on tall grass.
• The study successfully applied the classification model to real-world grazing scenarios.

Conclusions:

• A stacking model utilizing IMU sensor data and a 3-second window provides a highly accurate method for continuous sheep behavior classification.
• Sheep behavior, including walking, grazing, and resting, is significantly influenced by sward surface height.
• These findings support improved grazing sheep management and performance evaluation through automated behavior monitoring.