一种融合分组卷积和半监督的睡眠分期方法

已有的睡眠分期研究大部分采用监督学习的方法，其模型训练高度依赖于大量优质的标签数据，所提取的特征也较为粗糙。为此，提出了将半监督学习应用于分组卷积神经网络的睡眠分期方法。首先，采用分组残差卷积网络作为骨干网络，使不同分组学习的特征多样化，让整个网络关注来自不同子空间的信息，从而提取多角度特征； 其次，为减少标注技师的工作负担，采用半监督学习的方法，从大量未标注数据中提取特征与标注数据提取的特征进行对抗，以获得更多细粒度特征。实验结果表明，在Sleep-EDFx数据集上的睡眠分期准确率能够达到0.837±0.001，卡帕系数达到0.774±0.002，均优于对比算法。

A Sleep Staging Method Combining Grouping Convolution with Semi-Supervised Learning

Sleep is an important physiological activity of human body, and the quality of sleep affects physical and mental health. Most of the existing sleep staging is based on supervised learning, which is highly dependent on a large number of high-quality label data, and the extracted features are relatively rough. To save this issue, a sleep staging method combining grouped convolutional neural network and semi-supervised learning is proposed. First, the grouping residual convolution network is used as the backbone network to ensure the diversity of learning features and take the information from multiple subspaces into consideration, which extracts multi-angle features. Then, to reduce the workload of annotation technicians, semi-supervised learning method is adopted to extract features from a large number of unlabeled data and compete with those extracted from labeled data, which can obtain more fine-grained features. The experimental results show that the accuracy of sleep staging on sleep-EDFx can reach 0.837±0.001, and the Kappa coefficient reaches 0.774±0.002, which performs better than the baseline algorithm. The method presented has a good application prospect in the combination of medicine and industry.