融合多头自注意力机制的无接触心率估计模型

Contactless heart rate estimation model with multi-head self-attention mechanism

为了在光照变化和头部运动条件下实现准确稳定的无接触心率估计,基于U-Net模型提出一种融合多头自注意力机制的端到端心率估计模型rPPG-UNet。该模型通过使用U型编码器—解码器网络结构实现对生理特征的提取与重建,并使用Skip Connection连接编码器与解码器实现浅层时间特征的复用。该模型还融合多头自注意力机制来捕获生理特征的时间依赖性。最后,该模型采用多任务学习策略以提高心率估计的准确度,加速网络训练。在公开数据集上的实验结果表明,rPPG-UNet的性能优于其他基线模型,可以实现更准确的无接触心率估计。

To achieve accurate and stable contactless heart rate estimation under lighting changes and head motion conditions,this paper proposed an end-to-end heart rate estimation model called rPPG-UNet,which was based on the U-Net model and incorporated a multi-head self-attention mechanism.The model realized the extraction and reconstruction of physiological features by using the U-shaped encoder-decoder network structure,and used Skip Connection to connect the encoder and decoder to realize the multiplexing of shallow temporal features.The model also incorporated a multi-head self-attention mechanism to capture the temporal dependencies of physiological features.Finally,the model adopted a multi-task learning strategy to improve the accuracy of heart rate estimation and accelerate network training.Experimental results on public datasets show that rPPG-UNet outperforms other baseline models and can achieve more accurate contactless heart rate estimation.