Behavior detection of dairy goat based on YOLO11 and ELSlowFast-LSTM

Abstract

Monitoring dairy goat behavior can effectively assess their health status and welfare levels, ensuring both the yield and quality of goat milk. However, achieving accurate and rapid detection of dairy goat behaviors remains challenging. This study proposes a dairy goat behavior detection method based on YOLO11 and ELSlowFast-LSTM (3D-ELA-SlowFast-LSTM) to locate dairy goats and recognize five behaviors: standing, walking, lying down, climbing, and fighting. Firstly, the YOLO11 object detection module is used to pinpoint the locations of dairy goats. Next, the ELSlowFast-LSTM behavior recognition module is introduced to classify behaviors within the detected regions. This module utilizes the SlowFast network for spatiotemporal feature extraction, incorporating the 3D-Efficient Local (EL) attention mechanism specifically designed to enhance the extraction of behavior-related features. Additionally, the Long Short-Term Memory (LSTM) module is applied to model temporal sequence features. Finally, by combining the results of the two modules, the task of dairy goat behavior detection is accomplished. To evaluate the proposed method, we constructed the DairyGoat dataset. Experimental results show that our method achieved a $mAP$ value of 78.70%. Additionally, we compared the $mAP$ value of our proposed method with other behavior detection models, and the results demonstrate that our method achieves the best detection performance while maintaining a relatively low parameter count and computational load. In summary, this is an effective dairy goat behavior detection method that provides a new strategy for intelligent farming. The dataset and code are available at https://github.com/JunpengZZhang/ELSlowFast-LSTM.