CN-121997032-A - EEG signal feature extraction method, system, storage medium and terminal

CN121997032ACN 121997032 ACN121997032 ACN 121997032ACN-121997032-A

Abstract

The invention provides an EEG signal characteristic extraction method, a system, a storage medium and a terminal, wherein the method comprises the following steps of preprocessing an EEG signal; constructing a multi-scale reference window and a test window, calculating a multi-scale spectrum structure change score of the preprocessed EEG signal, detecting candidate segment boundary points of the preprocessed EEG signal based on the multi-scale spectrum structure change score, acquiring EEG self-adaptive segments based on the candidate segment boundary points, performing Hermite function decomposition on the EEG signal in the EEG self-adaptive segments to acquire EEG segment level features, and constructing EEG feature vectors based on the EEG segment level features and EEG global statistics. The EEG signal feature extraction method, the system, the storage medium and the terminal can extract the EEG signal features with the characteristics of strong event sensitivity, high self-adaption capability, excellent stability and the like.

Inventors

XU TIANHENG
FAN YANHONG
WANG ZHENYU
PENG PEI
CHEN XIANFU
OUYANG YULING
HU HONGLIN

Assignees

中国科学院上海高等研究院

Dates

Publication Date: 20260508
Application Date: 20260407

Claims (10)

1. A method of EEG signal feature extraction, the method comprising the steps of: preprocessing the EEG signal; Constructing a multi-scale reference window and a test window, and calculating a multi-scale spectrum structure change score of the preprocessed EEG signal; Detecting candidate segment boundary points of the preprocessed EEG signal based on the multi-scale spectral structure variation score; Acquiring an EEG adaptive segment based on the candidate segment boundary points; performing Hermite function decomposition on the EEG signals in the EEG self-adaptive segmentation to obtain EEG segment level characteristics; An EEG feature vector is constructed based on the EEG segment level features and EEG global statistics.
2. The EEG signal feature extraction method according to claim 1, wherein preprocessing the EEG signal comprises the steps of: High-pass filtering the EEG signal to remove baseline wander; performing low-pass filtering on the EEG signal after high-pass filtering to inhibit myoelectric artifacts and high-frequency noise; and carrying out power frequency notch filtering on the EEG signal after the low-pass filtering to eliminate power supply interference.
3. The EEG signal feature extraction method according to claim 1, wherein constructing a multi-scale reference window and test window, and calculating a multi-scale spectral structure change score of the preprocessed EEG signal comprises the steps of: constructing a multi-scale reference window and a test window, wherein the length of the reference window is larger than that of the test window under each scale; calculating a power spectral density of the preprocessed EEG signal based on the reference window and the test window, and normalizing the power spectral density; for the same scale s, calculating the Jensen-Shannon divergence between the reference window and the test window based on the normalized power spectral density Spectral entropy difference And respectively carrying out normalization treatment; Calculating a spectral structure change score at a corresponding scale based on the normalized Jensen-Shannon divergence and spectral entropy difference Wherein And Respectively represents normalized Jensen-Shannon divergence and normalized spectral entropy difference, And All represent weight coefficients and satisfy , Representing a kth time position; mapping the spectrum structure change scores under each scale to a unified time grid, and then performing multi-scale fusion to obtain the multi-scale spectrum structure change scores Wherein Representing the spectral structure change scores mapped to a uniform time grid at scale s, Representing multi-scale fusion weights and satisfying 。
4. The EEG signal feature extraction method according to claim 1, wherein detecting candidate segment boundary points of the pre-processed EEG signal based on the multi-scale spectral structure variation score comprises the steps of: short-time monitoring is carried out on the high-frequency energy of the preset frequency band of the preprocessed EEG signal, and an emergency area is identified; Enhancing the multi-scale spectrum structure change score of the emergency region to obtain an updated multi-scale spectrum structure change score Wherein Representing the multi-scale spectral structure change score, Representing an emergency indication function, The preset parameters are indicated to be the same as the preset parameters, Representing a kth time position; Constructing an adaptive threshold based on the median and absolute median of the updated multi-scale spectral structure variation scores Wherein med (-) represents the median, MAD (-) represents the absolute median, Representing preset parameters; and marking the corresponding EEG signal as a candidate segment boundary point of the EEG signal when the updated multi-scale spectrum structure change score is larger than the adaptive threshold.
5. The EEG signal feature extraction method according to claim 1, wherein obtaining an EEG adaptive segment based on the candidate segment boundary points comprises the steps of: When the time interval between two adjacent candidate segment boundary points is smaller than the minimum interval threshold value, merging the two candidate segment boundary points into a single boundary point, and reserving a position with a larger updated multi-scale spectrum structure change score as a boundary point after fusion; Acquiring the segment duration of the candidate segment generated by the candidate segment boundary point after neighborhood fusion; When the segment duration is smaller than a lower threshold, the corresponding candidate segment is integrated into the adjacent segment where the adjacent boundary point with smaller multi-scale spectrum structure change score is located after updating; when the updated multi-scale spectrum structure change scores of the boundary points at the left side and the right side are the same, merging one side adjacent segment with longer segment duration; And when the segment time length is larger than the upper limit threshold value, recursively segmenting the corresponding candidate segment according to the peak position or the midpoint position of the multi-scale spectrum structure change score updated in the segment.
6. The EEG signal feature extraction method according to claim 1, wherein performing Hermite function decomposition on the EEG signals in the EEG adaptive segmentation, obtaining EEG segment level features comprises the steps of: Acquiring decomposition parameters of Hermite function decomposition based on intra-segment signal features of the EEG adaptive segmentation, wherein the decomposition parameters comprise decomposition orders And scaling the scale Wherein, to The decomposition order is obtained by performing rounding and then performing truncation processing according to the first upper limit and the first lower limit For a pair of Performing truncation processing according to a second upper limit and a second lower limit to obtain the scaling scale Alpha, gamma, beta, c are mapping coefficients, The information capacity is represented by the information capacity, Representing the centroid of the spectrum, ε representing the minimum constant that avoids denominator zero; And performing Hermite function decomposition on the EEG signals based on the decomposition parameters to acquire the EEG segment level features.
7. The EEG signal feature extraction method according to claim 1, wherein constructing an EEG feature vector based on the EEG segment level features and EEG global statistics comprises the steps of: taking the segment length of the EEG adaptive segment as a weight; calculating a weighted mean value and a weighted standard deviation of the segment level features; combining the weighted mean, the weighted standard deviation, and the EEG global statistic to form the EEG feature vector.
8. An EEG signal characteristic extraction system is characterized by comprising a preprocessing module, a construction module, a detection module, an acquisition module, a decomposition module and a construction module; the preprocessing module is used for preprocessing the EEG signals; The construction module is used for constructing a multi-scale reference window and a test window and calculating a multi-scale spectrum structure change score of the preprocessed EEG signal; The detection module is used for detecting candidate segment boundary points of the preprocessed EEG signal based on the multi-scale spectrum structure change score; The acquisition module is used for acquiring an EEG self-adaptive segment based on the candidate segment boundary points; The decomposition module is used for performing Hermite function decomposition on the EEG signals in the EEG self-adaptive segmentation to acquire EEG segment level characteristics; The construction module is configured to construct an EEG feature vector based on the EEG segment level features and EEG global statistics.
9. A terminal is characterized by comprising a processor and a memory; the memory is used for storing a computer program; The processor is configured to execute the computer program stored by the memory to cause the terminal to perform the EEG signal feature extraction method according to any one of claims 1 to 7.
10. A computer readable storage medium having stored thereon a computer program, characterized in that the program, when executed by a terminal, implements the EEG signal feature extraction method of any one of claims 1 to 7.

Description

EEG signal feature extraction method, system, storage medium and terminal Technical Field The invention belongs to the technical field of biomedical signal processing, and relates to an EEG signal characteristic extraction method, an EEG signal characteristic extraction system, a storage medium and a terminal. Background Sleep apnea is one of the most common sleep disordered breathing disorders, and its central features include interruption of breathing, hypoventilation-induced blood oxygen decline, and concomitant arousal. Long-term non-intervention significantly increases the risk of cardiovascular disease, metabolic syndrome and cognitive impairment, and therefore a need exists for a convenient and effective early screening means. Current clinical diagnosis relies mainly on polysomnography (Polysomnography, PSG). The method is accurate, but has complicated equipment, high requirement on monitoring environment and high cost, and is not suitable for large-scale general investigation. With the popularity of wearable devices, automated sleep apnea detection based on single-lead electroencephalograms (Electroencephalography, EEG) is becoming a research hotspot. However, EEG signals are highly non-stationary and event-scale diverse, exposing significant limitations to conventional feature extraction methods. In the prior art, EEG feature extraction typically relies on fixed window time domain, frequency domain or time-frequency domain analysis. Such fixed window methods are difficult to adaptively cope with the dynamic changes of sleep EEG, and especially to capture short-term emergencies such as arousal, sympathetic activation and high frequency energy bursts that are common in the course of apneas. These events are short in duration and rapid in energy change, and easily cross the analysis window, so that the features are diluted, the event boundaries are misplaced, and the physiological process cannot be accurately reflected. Even with more advanced time-frequency analysis methods, such as short-time fourier transform, wavelet transform, empirical mode decomposition (EMPIRICAL MODE DECOMPOSITION, EMD), and variational mode decomposition (Variational Mode Decomposition, VMD), there are problems of cured decomposition scale, limited sensitivity to sudden events, insufficient robustness in fast-changing scenarios, etc. Existing segmentation strategies rely on single-scale spectral entropy or energy mutation detection, but EEG signals inherently have a multi-scale structure. Apnea-related events exhibit different patterns on different time scales, such as respiratory phase rhythms, arousal bursts, frequency band energy changes, etc., across different time scales, single-scale features have difficulty fully describing these multi-level spectral structure changes. Furthermore, existing methods lack specialized recognition mechanisms for sudden events (such as high frequency bursts caused by arousal), which are often important physiological signals for judging apneas. Lack of such specialized mechanisms can impair the sensitivity of event detection, making automatic screening difficult to improve. In summary, the prior art has the following general problems that the EEG signal cannot be effectively subjected to self-adaptive segmentation based on physiological event boundaries, a joint characterization mechanism of a multi-scale spectrum structure and an emergency is lacked, a stable aggregation frame is lacked in segment-level characteristics, and finally, insufficient characteristic expression capability and poor robustness are caused, so that the application requirements of automatic screening of clinical sleep apnea are difficult to meet. Disclosure of Invention The invention aims to provide an EEG signal characteristic extraction method, an EEG signal characteristic extraction system, a storage medium and a terminal, which can extract EEG signal characteristics with the characteristics of strong event sensitivity, high self-adaption capability, excellent stability and the like. In a first aspect, the invention provides an EEG signal feature extraction method, which comprises the steps of preprocessing an EEG signal, constructing a multi-scale reference window and a test window, calculating a multi-scale spectrum structure change score of the preprocessed EEG signal, detecting candidate segment boundary points of the preprocessed EEG signal based on the multi-scale spectrum structure change score, acquiring EEG self-adaptive segments based on the candidate segment boundary points, performing Hermite function decomposition on the EEG signal in the EEG self-adaptive segments, acquiring EEG segment level features, and constructing an EEG feature vector based on the EEG segment level features and EEG global statistics. In one implementation of the first aspect, preprocessing the EEG signal comprises the steps of: High-pass filtering the EEG signal to remove baseline wander; performing low-pass filtering on the EEG