Sensor-Based Cattle Behavior Classification Using Deep Learning Approaches

The usage of precision livestock has grown due to the need for higher efficiency and productivity in response to the high demand for food. To ensure sustainable development and quality control of the inputs required by the industry, it is essential to monitor and classify the behavior of cattle. Sensor-based monitoring systems provide accurate information by capturing raw data and identifying behavior through machine learning and deep learning algorithms. This approach has allowed farmers to better understand the individual needs of their animals. This study presents a deep residual neural network for the classification of cattle behavior. The performance of the ResNeXt model was evaluated using a public real-world dataset collected from sensors attached to the neck of six different Japanese black beef cows. The experimental results showed that the presented ResNeXt model achieved the highest average accuracy of 94.96% and the highest average F1-score of 93.66%. Compared to other baseline deep learning models and the current state-of-the-art model for cattle behavior classification, the presented model outperformed them and achieved better performance.