Sleep stage classification detection algorithm based on wearable physiological equipment

Abstract

The invention discloses a sleep stage classification detection algorithm based on wearable physiological equipment. The sleep stage classification detection algorithm comprises the following steps: acquiring a multi-mode electroencephalogram physiological signal; sequentially performing filtering noise removal and Z-score normalization processing on the multi-mode electroencephalogram physiological signals to obtain preprocessed signal data; the preprocessed signal data are optimized through a back propagation algorithm and gradient descent, a cross entropy loss function is used for measuring the difference between a generated signal and a real signal, and parameters are continuously adjusted to enable the generated signal to be closer to a real sleep signal; a discriminator with the same convolution unit form as the generator is constructed, the discriminator and the generator form a generative adversarial network model, adjustment parameters are automatically adjusted, and classification results of different sleep stages are output. By reasonably integrating EEG, EOG and other signals and the application of the GAN model, the model can learn the distribution of real sleep signals and perform accurate classification; and the classification accuracy and stability are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of sleep stage classification and detection, and particularly to a sleep stage classification and detection algorithm based on wearable physiological devices. Background Art

[0002] In the field of sleep stage classification, traditional methods mainly rely on extracting a single feature from electroencephalogram (EEG) signals to determine the sleep state. Generally, these methods focus on the time-domain or frequency-domain features of EEG signals, such as calculating the power spectral density of a specific frequency band, analyzing the occurrence frequency of specific waveforms, etc., and then matching them with a preset sleep stage feature template to determine the sleep stage. However, there are many drawbacks that cannot be ignored in this traditional way:

[0003] Limited information mining: Relying solely on the single signal feature of EEG cannot comprehensively cover the complex and diverse physiological information during sleep. During sleep, multiple physiological systems in the human body change in coordination. In addition to the electrical activity of the brain reflected by EEG, eye movements, muscle activities, etc. are also closely related to sleep stages, but traditional methods do not take these information into consideration. For example, in some special sleep conditions, such as when an individual is slightly disturbed by the outside world during sleep, the EEG signal may show abnormal fluctuations. Judging only based on the traditional single EEG feature, it is very easy to make misjudgments and cannot accurately identify the sleep stage.

[0004] Poor individual adaptability: There are natural differences in the frequency, amplitude, and waveform of EEG signals among different individuals. The traditional fixed feature classification standard is difficult to adapt to these variations from person to person. For example, the EEG frequency range of some individuals in a specific sleep stage may deviate from the general standard. Classifying according to the traditional method is likely to misjudge their sleep stage, seriously affecting the accurate assessment of individual sleep quality.

[0005] Weak anti-interference performance: In the actual sleep monitoring environment, there are various interference sources, such as electromyogram (EMG), electrocardiogram (ECG), and environmental electromagnetic interference. These interferences mix into the EEG signal and will seriously reduce the signal quality. Traditional methods lack an effective interference filtering mechanism. In the presence of interference, the extracted features may be distorted by the interference signal, resulting in a significant decrease in the accuracy of sleep stage classification. For example, when the EMG interference is strong, traditional methods may misjudge an EEG signal containing a large amount of EMG artifacts as a wakefulness state signal, causing an error in sleep stage classification.

[0006] Problem of data annotation: Traditional methods rely on manual annotation of EEG signals to determine sleep stages. This process not only consumes a large amount of time and manpower, but also has strong subjectivity. Differences in the experience and judgment criteria of different annotators make it difficult to guarantee the consistency and reliability of the annotation results. With the continuous increase in the amount of data, the problem of low efficiency of manual annotation has become increasingly prominent, severely restricting the construction of large-scale datasets and further limiting the development of sleep stage classification models based on big data. Summary of the Invention

[0007] To at least solve or partially solve the above problems, a sleep stage classification and detection algorithm based on wearable physiological devices is provided. By reasonably integrating signals such as EEG and EOG, the physiological information during sleep is fully utilized; the application of the GAN model, including the design of the generator and discriminator and the adversarial training mechanism between the two, enables the model to learn the distribution of real sleep signals and perform accurate classification; the preprocessing method for multi-modal signals ensures the data quality of the input model and improves the accuracy and stability of classification.

[0008] To achieve the above object, the present invention provides the following technical solutions:

[0009] A sleep stage classification and detection algorithm based on wearable physiological devices of the present invention includes the following steps:

[0010] A. Obtain multi-modal electroencephalogram physiological signals;

[0011] B. Preprocessing: Perform filtering to remove noise and Z-score normalization on the multi-modal electroencephalogram physiological signals in sequence to obtain preprocessed signal data;

[0012] C. Generator: Implement it on the preprocessed signal data through the backpropagation algorithm and gradient descent optimization. Measure the difference between the generated signal and the real signal with the cross-entropy loss function, and continuously adjust the parameters to make the generated signal closer to the real sleep signal;

[0013] D. Discriminator: Construct a discriminator with the same form of convolutional units as the generator. The discriminator and the generator form a generative adversarial network model, automatically adjust the adjustment parameters, and perform sleep stage classification on the multi-modal physiological signals through a series of convolutional, pooling, and fully connected operations, and output the classification results of different sleep stages.

[0014] As a preferred technical solution of the present invention, the discriminator discriminates between the sample signal data and the signal data generated by the generator. If the discriminator classifies the sample signal data and the signal data generated by the generator into the same category, adjust the adjustment parameters of the discriminator. If the discriminator classifies the sample signal data and the signal data generated by the generator into two categories, adjust the adjustment parameters of the generator and make the generator regenerate the signal data.

[0015] As a preferred technical solution of the present invention, the filtering for noise removal adopts a band-pass filter of 0.5 - 30 Hz.

[0016] As a preferred technical solution of the present invention, the multi-modal electroencephalogram physiological signals include electroencephalogram signals and electrooculogram signals.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0018] Advantages of multi-modal fusion: The present invention breaks through the limitations of traditional single signals and innovatively fuses multi-modal physiological signals, organically combining EEG signals and EOG signals. EEG signals reflect brain activities, and EOG signals provide eye movement information. The two complement each other and can comprehensively and accurately reflect the sleep state from multiple dimensions. Compared with traditional methods, the accuracy and reliability of sleep stage classification are greatly improved, and misjudgments caused by information loss are effectively avoided.

[0019] Advantages of the GAN model: Introducing the generative adversarial network (GAN) is a major highlight of the present invention. The generator in the GAN can learn the distribution law of real sleep signals, generate a large number of simulated sleep signals, expand the training data set, effectively alleviate the problems of insufficient and unbalanced data, enhance the generalization ability of the model, and enable it to better adapt to the sleep signal differences of different individuals. At the same time, the adversarial training process between the discriminator and the generator enables the model to deeply explore the internal features and patterns of sleep signals. Even in complex situations where the signals are noisy or incomplete, the sleep stages can be accurately identified, significantly improving the robustness of the model.

[0020] Automatic learning ability: The GAN model has a powerful automatic learning ability. Without manual extraction of complex features, it can automatically learn effective feature representations from multi-modal physiological signals. This not only improves the efficiency and accuracy of feature extraction but also avoids subjective biases and information loss that may occur in the process of manual feature extraction, enabling the model to better adapt to different individuals and complex sleep monitoring environments and providing more reliable data support for sleep research and clinical diagnosis. Description of the Drawings

[0021] The drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation to the present invention. In the drawings:

[0022] Figure 1 is the main architecture diagram of the generative adversarial network (GAN) algorithm for electroencephalogram (EEG) sleep stage classification of the present invention;

[0023] Figure 2It is the main architecture diagram of the discriminator of the present invention. Detailed implementation mode

[0024] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0025] In addition, if the detailed description of the known technology is unnecessary for showing the features of the present invention, it will be omitted.

[0026] Embodiment 1

[0027] As Figure 1-2 shown, the present invention provides a sleep stage classification detection algorithm based on a wearable physiological device. Multi-modal physiological signal acquisition and processing: Use a wearable device to collect multi-modal physiological signals including electroencephalogram (EEG) signals, electrooculogram (EOG) signals, etc. EEG signals reflect brain nerve activities, and different sleep stages have unique frequency, amplitude, and waveform patterns. For example, when awake, it is mainly dominated by high-frequency and low-amplitude β waves, and when in deep sleep, it is mainly dominated by low-frequency and high-amplitude δ waves; EOG signals can assist in judging the rapid eye movement (REM) sleep stage and provide supplementary information for sleep stage judgment through eye movement conditions. The collected signals are preprocessed, including filtering to remove noise (such as performing band-pass filtering on EEG signals from 0.5 to 30 Hz to remove power frequency interference and high-frequency myoelectric artifacts) and Z-score normalization processing to eliminate the amplitude differences between different signals and prepare for subsequent analysis.

[0028] Construction of a sleep stage classification model based on GAN: GAN consists of a generator and a discriminator. The task of the generator is to learn the distribution of real sleep signals and generate simulated sleep signals based on the input random noise. Its learning process is achieved through backpropagation algorithm and gradient descent optimization. The cross-entropy loss function is used to measure the difference between the generated signal and the real signal, and the parameters are continuously adjusted to make the generated signal closer to the real sleep signal. For example, it simulates the EEG signal characteristics of different sleep stages. The discriminator receives real sleep signals and simulated signals generated by the generator, performs feature extraction and classification through a series of convolutional, pooling, and fully connected operations, judges the authenticity of the signals, and identifies the sleep stages. The loss function of the discriminator is calculated based on its judgment results, and the parameters are updated through backpropagation and gradient descent to improve its discrimination ability. During the training process, the generator and the discriminator confront each other and co-evolve, making the signals generated by the generator more and more realistic and the classification ability of the discriminator continuously enhanced. Finally, the trained discriminator is used to classify the multi-modal physiological signals into different sleep stages such as awake, light sleep (N1, N2), deep sleep (N3), and REM sleep, and output the classification results.

[0029] The present invention rationally integrates signals such as EEG and EOG, and makes full use of the physiological information during sleep; the application of the GAN model, including the design of the generator and discriminator and the adversarial training mechanism between the two, enables the model to learn the distribution of real sleep signals and perform accurate classification; the preprocessing method for multi-modal signals ensures the data quality input into the model and improves the accuracy and stability of classification.

[0030] Finally, it should be noted that the above are only the preferred embodiments of the present invention and are not used to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A sleep stage classification and detection algorithm based on a wearable physiological device, characterized in that It includes the following steps: A. Obtain multi-modal electroencephalogram physiological signals; B. Preprocessing: successively perform filtering to remove noise and Z-score normalization on the multi-modal electroencephalogram physiological signals to obtain preprocessed signal data; C. Generator: implement on the preprocessed signal data through backpropagation algorithm and gradient descent optimization, measure the difference between the generated signal and the real signal with the cross-entropy loss function, and continuously adjust the parameters to make the generated signal closer to the real sleep signal; D. Discriminator: construct a discriminator with the same convolutional unit form as the generator, the discriminator and the generator form a generative adversarial network model, automatically adjust the adjustment parameters, and classify the multi-modal physiological signals through a series of convolutional, pooling, and fully connected operations to output the classification results of different sleep stages.

2. The sleep stage classification detection algorithm based on a wearable physiological device according to claim 1, characterized in that, The discriminator discriminates the sample signal data and the signal data generated by the generator. If the discriminator classifies the sample signal data and the signal data generated by the generator into one category, adjust the adjustment parameters of the discriminator. If the discriminator classifies the sample signal data and the signal data generated by the generator into two categories, adjust the adjustment parameters of the generator and let the generator regenerate the signal data.

3. The sleep stage classification detection algorithm based on a wearable physiological device according to claim 1, wherein, The filtering to remove noise uses a band-pass filter of 0.5 - 30 Hz.

4. The sleep stage classification detection algorithm based on a wearable physiological device according to claim 1, wherein The multi-modal electroencephalogram physiological signals include electroencephalogram signals and electrooculogram signals.

Images