Abstract
The efficient and precise recognition of diarrhea-related behaviors in Jinnan calves is crucial for ensuring their healthy development. Nevertheless, conventional behavior recognition techniques are often limited by a notable decline in performance when distinguishing between similar behavioral patterns. This paper proposes a novel behavior recognition model for Jinnan calf diarrhea, named S_T_Mamba (Sequence Tree Mamba). Specifically, S_T_Mamba incorporates a sequence processing strategy and a tree state space module (TreeSSM). The sequence processing strategy utilizes sequence as inputs to capture the temporal dependencies underlying the video. Additionally, the tree state space module is designed to extract and aggregate long-range pixel association features from video frames, enabling the effective recognition of subtle distinctions between similar behaviors. Therefore, the proposed model significantly enhances the performance of calf diarrhea behavior recognition. Experimental results indicate that the S_T_Mamba model achieves state-of-the-art performance in Jinnan calf diarrhea behavior recognition. Specifically, S_T_Mamba achieves 99.78% accuracy, outperforming existing popular models by 0.59% to 1.99%.