Table 1 Parameter settings.

Number of Convolutional Kernels per Layer

64

The number of convolution kernels in each layer of graph convolution, used to capture the local spatial features of skeletal nodes

Activation Function

ReLU

The ReLU activation function is used to introduce non-linearity, enhancing the model’s ability to fit complex features

Optimization Algorithm

Adam

The Adam optimizer is employed for fast convergence, while the adaptive learning rate handles the sparse gradient issue

Learning Rate

0.001

The learning rate is set to 0.001, and a cosine annealing schedule is adopted to gradually reduce the learning rate, improving the model’s convergence performance

Batch Size

32

The number of samples per training iteration is set to balance training speed and model performance, helping to prevent overfitting

Number of Training Epochs

80

The total number of training epochs is set to 80 based on experimental validation. It is verified that this number achieves convergence while avoiding both overfitting and underfitting

Key Point Detection Threshold

0.5

The minimum confidence threshold for skeletal keypoints is defined, and keypoints below this threshold are filtered to improve data quality

Data Normalization Range

[− 1, 1]

Data are normalized to the range of [-1, 1] to ensure consistent scaling of different features, and enhance the model’s stability and convergence speed