Detection of estrous ewes’ tail-wagging behavior in group-housed environments using Temporal-Boost 3D convolution

Abstract

In the early estrous stage, ewes exhibit characteristic behaviors, including frequent movement or repetitive tail-wagging. Accurately identifying tail-wagging behavior is essential for determining whether ewes are in estrus, which is crucial for optimizing breeding timing and enhancing productivity in sheep farming. However, detecting ewes’ tail-wagging behavior in group-housed environments remains challenging, because the movements of the sheep make it difficult to analyze the body parts of each ewe individually. This study aims to propose a method for detecting tail-wagging behavior of estrous ewes in group-housed environments. The proposed method consists of three main modules: keypoint detection of the sheep skeletal, localization of the tail regions, and detection of tail-wagging behavior using a Temporal-Boost 3D convolutional network. Firstly, YOLOv8-pose is employed to obtain tail skeleton keypoints of the ewes. Secondly, tolerance expansion techniques are used to determine the tail locations of all ewes. Finally, the Temporal-Boost 3D convolutional network extracts features from both RGB and optical flow sequences. To improve classification accuracy, dynamic weighted fusion is then applied to the softmax outputs from both the RGB and optical flow data streams, producing the final classification result. To evaluate the practical effectiveness of this method, a video was selected for tail-wagging behavior detection, which contained 39 actual tail-wagging segments. The proposed method successfully detected 40 continuous tail-wagging segments, capturing all actual segments and achieving an accuracy rate of 97.5%. These results indicate that the method can effectively detect tail-wagging behavior in ewes within group-housed environments, meeting the intelligent detection needs of sheep farms.