BCG signal self-supervision feature extraction and sleep quality evaluation method based on SimCLR

Abstract

The invention discloses a SimCLR-based BCG signal self-supervision feature extraction and sleep quality evaluation method, and relates to the field of sleep quality evaluation, and the method comprises the following steps: S1, data acquisition: using professional sensor equipment, placing the professional sensor equipment at a position, close to the heart, of the chest of a subject, acquiring BCG signals in the whole night of a sleep period at a high sampling rate, and collecting a large number of BCG signals; s2, data preprocessing: preprocessing the collected BCG signal, then segmenting the signal, and cutting the whole-night continuous signal into a plurality of segments according to a preset time window; according to the method, SimCLR self-supervised learning only needs a small number of labeled samples to be used for fine tuning of the final evaluation model, potential information in BCG signals can be effectively mined by using massive unlabeled BCG data to mine potential features, the accuracy and efficiency of sleep quality evaluation are improved, dependence on massive labeled data can be reduced through a self-supervised learning mode, and the accuracy and efficiency of sleep quality evaluation are improved. And the data annotation cost and difficulty are reduced.

Description

Technical Field

[0001] The present invention relates to the field of sleep quality assessment, and in particular to a BCG signal self-supervised feature extraction and sleep quality assessment method based on SimCLR. Background Art

[0002] Sleep quality has a significant impact on people's physical and mental health and daily life. Accurately assessing sleep quality can help detect sleep disorders and other problems early and provide a basis for developing personalized intervention measures. The cardiogram signal is a biological signal that can reflect cardiac activity and the body's physiological state, and contains a wealth of sleep-related information. Traditional sleep quality assessment methods usually rely on polysomnography. Although this method is accurate, the equipment is complex, the cost is high, and it brings inconvenience to the subjects.

[0003] Therefore, it is necessary to invent a BCG signal self-supervised feature extraction and sleep quality assessment method based on SimCLR to solve the above problems. Summary of the Invention

[0004] The purpose of the present invention is to provide a BCG signal self-supervised feature extraction and sleep quality assessment method based on SimCLR to solve the problems raised in the above background technology.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a BCG signal self-supervised feature extraction and sleep quality assessment method based on SimCLR, comprising the following steps: S1: Data collection: Using specialized sensor equipment, place it on the subject's chest near the heart to collect BCG signals throughout the night at a high sampling rate, collecting a large amount of BCG signals. S2: Data preprocessing: The collected BCG signals are preprocessed and then segmented. The entire night's continuous signal is cut into multiple segments according to the preset time window. Each segment serves as the basic unit for subsequent model processing. Each segment is standardized so that the data of different segments have similar numerical ranges, improving the stability of model training. S3: Use SimCLR for self-supervised feature extraction: Implement multiple data augmentation strategies for each BCG signal segment. These transformed versions form positive sample pairs with the original segment, while different segments form negative sample pairs with each other. Select a suitable neural network structure as the feature extractor of SimCLR, input the enhanced BCG signal view into the neural network, calculate the cosine similarity between feature vectors based on SimCLR's contrastive learning principle, construct a contrastive loss function, and train the neural network using optimization algorithms such as stochastic gradient descent. Continuously adjust the network parameters so that the feature vectors of different views of the same BCG signal sample are close to each other in the feature space, and the feature vectors of views of different BCG signal samples are far away from each other. The trained SimCLR model can be used to extract the features of the BCG signal. Input the new BCG signal into the trained model to obtain the corresponding feature vector. S4: Sleep quality assessment: Select features related to sleep quality from the extracted BCG signal features. Methods such as feature importance analysis can be used. At the same time, the BCG signal features can be fused with the features of other related physiological signals. The extracted self-supervised features are used as input to build a classifier or regression model. The model is supervised and trained using BCG data samples with labeled sleep quality. Sleep quality is divided into different levels. The trained model can make accurate predictions and assessments of the sleep quality corresponding to the newly collected BCG signals.

[0006] Preferably, the loss function of the SimCLR unsupervised deep contrastive learning model is used to calculate the contrast loss

[0007] Preferably, in S2, the preprocessing of the collected BCG signal includes operations such as denoising, filtering, and normalization to remove noise interference and make the signal more stable and easier to process.

[0008] Preferably, the filtering can be a combination of low-pass and high-pass filters to remove high-frequency noise and low-frequency baseline drift and restore the pure waveform of the BCG signal.

[0009] Preferably, in S3, the BCG signal segment enhancement method includes random time stretching, random cropping, signal amplitude transformation and Gaussian noise addition, so as to increase data diversity.

[0010] Preferably, in S3, the neural network structure of the feature extractor suitable for SimCLR is a one-dimensional convolutional neural network.

[0011] Preferably, in S3, the feature vector contains key information related to sleep quality, such as sleep stage, breathing abnormalities, and heart rhythm changes.

[0012] Preferably, in S4, the constructed classifier or regression model can use support vector machine SVM, random forest and neural network, etc.

[0013] Preferably, the sleep quality assessment result verification includes the following methods: Dataset division: The collected BCG signal data is divided into training set, validation set, and test set. The training set is used to train the SimCLR model and sleep quality assessment model, the validation set is used to adjust model parameters and select the optimal model, and the test set is used to evaluate model performance. Model training and evaluation: The SimCLR model and sleep quality assessment model were trained using the training set data. The model parameters were adjusted using the validation set data. The model performance was evaluated on the test set. Evaluation metrics such as accuracy, recall, F1 value, and mean square error were used to measure the accuracy and reliability of the model in sleep quality assessment. Comparative experiment: A comparative experiment was conducted with the traditional BCG signal-based sleep quality assessment method to verify the advantages of the SimCLR-based method in feature extraction and sleep quality assessment.

[0014] Technical effects and advantages of the present invention: In the present invention, SimCLR self-supervised learning only requires a small number of labeled samples for fine-tuning the final evaluation model. It can use massive unlabeled BCG data to mine potential features, effectively mining the potential information in the BCG signal, and improving the accuracy and efficiency of sleep quality assessment. Through self-supervised learning, it can reduce the dependence on a large amount of labeled data, reduce the cost and difficulty of data labeling, and capture subtle changes and hidden patterns in the BCG signal through complex data enhancement and comparative learning. These deep features are better adaptable to complex situations such as differences in sleep characteristics of different individuals and the impact of environmental changes, providing a solid foundation for accurate sleep quality assessment, and have significant advantages over simple manual feature engineering methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0016] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0017] The present invention provides Figure 1The BCG signal self-supervised feature extraction and sleep quality assessment method based on SimCLR shown includes the following steps: S1: Data Collection: Use specialized sensor equipment, placed on the subject's chest near the heart, to collect BCG signals throughout the night at a high sampling rate, collecting a large amount of BCG signals. Other auxiliary sensors, such as accelerometers for detecting changes in body posture, can also be used to synchronously record data for subsequent multimodal analysis. The collection environment should be kept as quiet and comfortable as possible to minimize the impact of external interference on signal quality. S2: Data preprocessing: The collected BCG signals are preprocessed and then segmented. The segmentation time can be 30 seconds to 1 minute. The entire night's continuous signal is cut into multiple segments according to the preset time window. Each segment serves as the basic unit for subsequent model processing. Each segment is standardized so that the data of different segments have similar numerical ranges, improving the stability of model training. S3: Use SimCLR for self-supervised feature extraction: Implement multiple data augmentation strategies for each BCG signal segment. These transformed versions form positive sample pairs with the original segment, while different segments form negative sample pairs with each other. Select an appropriate neural network structure as the SimCLR feature extractor, input the enhanced BCG signal view into the neural network, and calculate the cosine similarity between feature vectors based on SimCLR's contrastive learning principle. Construct a contrastive loss function, and train the neural network using optimization algorithms such as stochastic gradient descent. Continuously adjust the network parameters so that the feature vectors of different views of the same BCG signal sample are close to each other in the feature space, while the feature vectors of views of different BCG signal samples are far apart. After multiple rounds of iterative training, the model can eventually automatically learn and extract discriminative and robust deep features from the BCG signal. The trained SimCLR model can be used to extract features from the BCG signal. New BCG signals are input into the trained model to obtain the corresponding feature vectors. S4: Sleep quality assessment: Select features related to sleep quality from the extracted BCG signal features. Methods such as feature importance analysis can be used. At the same time, the BCG signal features can be fused with features of other related physiological signals. The extracted self-supervised features are used as input to build a classifier or regression model. The model is supervised and trained using BCG data samples with labeled sleep quality. Sleep quality is divided into different levels. The trained model can make accurate predictions and assessments of the sleep quality corresponding to the newly collected BCG signals. Sleep quality can be divided into different levels, such as the proportion of deep sleep, the number of awakenings, the integrity of the sleep cycle, and other multi-dimensional comprehensive considerations, which can be divided into excellent, good, medium, and poor levels.

[0018] Calculate the contrast loss using the loss function of the SimCLR unsupervised deep contrastive learning model

[0019] In S2, the preprocessing of the collected BCG signal includes operations such as denoising, filtering, and normalization to remove noise interference and make the signal more stable and easier to process.

[0020] The filtering can be a combination of low-pass and high-pass filters to remove high-frequency noise and low-frequency baseline drift and restore the pure waveform of the BCG signal.

[0021] In S3, the BCG signal segment enhancement methods include random time stretching, random cropping, signal amplitude transformation, and Gaussian noise addition to increase data diversity.

[0022] In S3, it is verified that the neural network structure of SimCLR's feature extractor is a one-dimensional convolutional neural network.

[0023] In S3, the feature vector contains key information related to sleep quality, such as sleep stages, breathing abnormalities, and heart rhythm changes.

[0024] In S4, the constructed classifier or regression model can use support vector machine SVM, random forest and neural network, etc.

[0025] The verification of sleep quality assessment results includes the following methods: Dataset division: The collected BCG signal data is divided into training set, validation set, and test set. The training set is used to train the SimCLR model and sleep quality assessment model, the validation set is used to adjust model parameters and select the optimal model, and the test set is used to evaluate model performance. Model training and evaluation: The SimCLR model and sleep quality assessment model were trained using the training set data. The model parameters were adjusted using the validation set data. The model performance was evaluated on the test set. Evaluation metrics such as accuracy, recall, F1 value, and mean square error were used to measure the accuracy and reliability of the model in sleep quality assessment. Comparative experiment: A comparative experiment was conducted with the traditional BCG signal-based sleep quality assessment method to verify the advantages of the SimCLR-based method in feature extraction and sleep quality assessment.

[0026] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A self-supervised feature extraction and sleep quality assessment method for BCG signals based on SimCLR, characterized by: The steps include: S1: Data collection: Using specialized sensor equipment, place it on the subject's chest near the heart to collect BCG signals throughout the night at a high sampling rate, collecting a large amount of BCG signals. S2: Data preprocessing: The collected BCG signals are preprocessed and then segmented. The entire night's continuous signal is cut into multiple segments according to the preset time window. Each segment serves as the basic unit for subsequent model processing. Each segment is standardized so that the data of different segments have similar numerical ranges, improving the stability of model training. S3: Use SimCLR for self-supervised feature extraction: Implement multiple data augmentation strategies for each BCG signal segment. These transformed versions form positive sample pairs with the original segment, while different segments form negative sample pairs with each other. Select a suitable neural network structure as the feature extractor of SimCLR, input the enhanced BCG signal view into the neural network, calculate the cosine similarity between feature vectors based on SimCLR's contrastive learning principle, construct a contrastive loss function, and train the neural network using optimization algorithms such as stochastic gradient descent. Continuously adjust the network parameters so that the feature vectors of different views of the same BCG signal sample are close to each other in the feature space, and the feature vectors of views of different BCG signal samples are far away from each other. The trained SimCLR model can be used to extract the features of the BCG signal. Input the new BCG signal into the trained model to obtain the corresponding feature vector. S4: Sleep quality assessment: Select features related to sleep quality from the extracted BCG signal features. Methods such as feature importance analysis can be used. At the same time, the BCG signal features can be fused with the features of other related physiological signals. The extracted self-supervised features are used as input to build a classifier or regression model. The model is supervised and trained using BCG data samples with labeled sleep quality. Sleep quality is divided into different levels. The trained model can make accurate predictions and assessments of the sleep quality corresponding to the newly collected BCG signals.

2. The method for self-supervised feature extraction and sleep quality assessment of BCG signals based on SimCLR according to claim 1, characterized in that: Calculate the contrast loss using the loss function of the SimCLR unsupervised deep contrastive learning model 3. The method for self-supervised feature extraction and sleep quality assessment of BCG signals based on SimCLR according to claim 2, characterized in that: In S2, the preprocessing of the collected BCG signal includes operations such as denoising, filtering, and normalization to remove noise interference and make the signal more stable and easier to process.

4. The method for self-supervised feature extraction and sleep quality assessment of BCG signals based on SimCLR according to claim 3, characterized in that: The filtering can be a combination of low-pass and high-pass filters to remove high-frequency noise and low-frequency baseline drift and restore the pure waveform of the BCG signal.

5. The method for self-supervised feature extraction and sleep quality assessment of BCG signals based on SimCLR according to claim 4, characterized in that: In S3, the BCG signal segment enhancement methods include random time stretching, random cropping, signal amplitude transformation, and Gaussian noise addition to increase data diversity.

6. The method for self-supervised feature extraction and sleep quality assessment of BCG signals based on SimCLR according to claim 5, characterized in that: In S3, the neural network structure of the feature extractor suitable for SimCLR is a one-dimensional convolutional neural network.

7. The method for self-supervised feature extraction and sleep quality assessment of BCG signals based on SimCLR according to claim 6, characterized in that: In S3, the feature vector contains key information related to sleep quality, such as sleep stages, breathing abnormalities, and heart rhythm changes.

8. The method for self-supervised feature extraction and sleep quality assessment of BCG signals based on SimCLR according to claim 7, characterized in that: In S4, the constructed classifier or regression model can use support vector machine SVM, random forest and neural network, etc.

9. The method for self-supervised feature extraction and sleep quality assessment of BCG signals based on SimCLR according to claim 8, characterized in that: The verification of sleep quality assessment results includes the following methods: Dataset division: The collected BCG signal data is divided into training set, validation set, and test set. The training set is used to train the SimCLR model and sleep quality assessment model, the validation set is used to adjust model parameters and select the optimal model, and the test set is used to evaluate model performance. Model training and evaluation: The SimCLR model and sleep quality assessment model were trained using the training set data. The model parameters were adjusted using the validation set data. The model performance was evaluated on the test set. Evaluation metrics such as accuracy, recall, F1 value, and mean square error were used to measure the accuracy and reliability of the model in sleep quality assessment. Comparative experiment: A comparative experiment was conducted with the traditional BCG signal-based sleep quality assessment method to verify the advantages of the SimCLR-based method in feature extraction and sleep quality assessment.

Images