基于迁移学习的深层卷积神经网络心电信号疲劳分类

传统的心电疲劳分类方法虽然能有效地识别疲劳状态，但需要采集较长时间的信号，不能达到疲劳状态的实时监测。本文设计一种深层卷积神经网络模型用于评估操作员疲劳状态，对操作员的短时心电信号进行疲劳状态的自动分类。首先，提出一种将心电信号转化为图像的方法，将采集到的心电信号转化成二维图像，即将心电信号直接映射到二维空间转换成时域图片信息。然后，将图片送入深层卷积神经网络模型中去训练，实现对操作员疲劳状态的分类。本文方法降低了模型的复杂性，减少了模型的参数，同时训练的数据不需要经过类似噪声滤波、特征提取等任何预处理步骤。结果表明该模型能自动从心电信号中提取有效特征，实现对操作员非疲劳和疲劳两种状态的正确分类，分类准确率达到97.36%。

Classification of fatigue state using ECG signals and DCNN with transfer learning

Abstract: Although the traditional fatigue classification method based on electrocardiogram (ECG) signals can effectively identify fatigue state, it needs to acquire long-term ECG signals and cannot achieve the real-time monitoring of the fatigue state. A deep convolutional neural network is designed for evaluating the fatigue state of the operator, and automatically classifying the fatigue state based on the short-term ECG signals. A method of converting the ECG signals into an images are firstly proposed for converting the acquired ECG signals into a two-dimensional image, which means that the ECG signals are directly mapped to the two-dimensional space and converted into time-domain image information. Then the images are sent to deep convolutional neural network model for training, thereby realizing the classification of fatigue state. The proposed method decreases the complexity of the model and reduces the parameters of the model, and meanwhile, the training data does not need any preprocessing such as noise filtering and feature extraction steps. The results show that the proposed model can automatically extract effective features from the ECG signals, and realize the correct classification of the non-fatigue and fatigue states of the operator, achieving a classification accuracy up to 97.36%.