Monitoring cattle behavior using deep learning: An LSTM-based approach with accelerometer data

Abstract

Behavior data analysis is a crucial factor in the early detection of cow health issues, thereby optimizing farming processes and improving productivity in large-scale farms. Accelerometers, attached to the neck or legs of cows, collect movement data, providing a foundation for analyzing animal behavior. Previous studies have proposed cow behavior classification systems based on accelerometer data combined with machine learning algorithms. However, with the advancement of deep learning, the application of Long Short-Term Memory (LSTM) networks can significantly enhance classification performance. In this study, we utilize an LSTM network to classify four primary cow behaviors: Eating, Lying, Standing, and Walking. The LSTM model effectively processes time-series data by retaining essential information while filtering out unnecessary data. Experimental results demonstrate that the model achieves high classification performance, with an average accuracy of approximately 90% across all behaviors, outperforming traditional machine learning algorithms. This research can be implemented in smart farms, integrating with IoT technology to automate livestock monitoring and management efficiently.