Pig behavior recognition serves as a crucial indicator for monitoring health and environmental conditions. However, traditional pig behavior recognition methods are limited in their processing capabilities, struggling to accurately extract image features and dynamically analyze sequence data. This paper introduces a novel ST_TransNeXt model, which ingeniously integrates the TransNeXt module with the sLSTM, enabling a profound understanding of the dynamic characteristics of pig group behaviors. Specifically, the Bio-inspired Aggregated Attention within the TransNeXt module is inspired by biological vision system to efficiently fuses local and global image features, sLSTM processes multi-frame data to capture temporal dependencies, the combination of these two techniques constructs a powerful temporal feature vector, enhancing model performance. Experimental results demonstrate that the proposed ST_TransNeXt model achieves an accuracy rate of 93.98%, outperforming existing models by more than 1.1% and achieving a maximum reduction in loss value of 0.4737.