Research on Cow Behavior Recognition Based on Improved SlowFast with 3DCBAM

Abstract

Accurate and fast recognition of dairy cow behavior is crucial for the intelligent perception of cow health status and disease prevention. Traditional cow behavior detection consumes a lot of staffing and resources, while wearable sensors can cause stress responses in cows. This paper proposes an improved SlowFast-based cow behavior recognition algorithm to identify cow behaviors such as standing, lying down, walking, drinking water, and eating grass and improves this spatiotemporal action detection model with the CBAM (Convolutional Block Attention Module) attention mechanism, which can adaptively weight each feature map in each path to enhance the ability to capture essential information in videos, thereby improving the network's accuracy. Experiments show that the improved SlowFast model proposed in this paper can achieve an accuracy of 97.3% for cow behavior recognition, which is 3.1% higher than the basic model.