Classifying Ingestive Behavior of Dairy Cows via Automatic Sound Recognition

Area of Science:

• Animal Science
• Machine Learning
• Acoustic Analysis

Background:

• Accurate assessment of dairy cow ingestive behaviors is crucial for monitoring productivity and health.
• Understanding the relationship between forage characteristics and ingestive sounds can enhance monitoring techniques.

Purpose of the Study:

• To establish correlations between forage species/heights and acoustic characteristics of dairy cow ingestive behaviors (bites, chews, chew-bites).
• To compare the efficacy of three deep learning models and optimization strategies for classifying these behaviors.
• To evaluate the performance of deep learning in classifying ingestive behaviors across diverse forage conditions.

Main Methods:

• Development of acoustic analysis to capture sound characteristics of ingestive behaviors.
• Comparative evaluation of three deep learning models, including a long short-term memory (LSTM) network.
• Testing classification accuracy under varying forage species and heights.

Main Results:

• Sound characteristics (amplitude, duration) of ingestive behaviors were generally larger for taller forages (tall fescue, alfalfa).
• The LSTM network demonstrated superior performance, especially with a filtered dataset balancing duration and audio files.
• Peak classification accuracy exceeded 0.93, with performance variations of 0.4-0.5 observed across different forage conditions.

Conclusions:

• Deep learning effectively classifies dairy cow ingestive behaviors using acoustic signals.
• Classification accuracy can be influenced by forage characteristics, limiting differentiation in some scenarios.
• The developed acoustic-based tool shows potential for precision dairy management but requires further refinement.