Physiological Signal Normalization Using Cepstral Features
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for normalizing physiological signals, such as min-max normalization and Z-Score normalization, are limited by the need to assume specific data distributions and are unstable when data ranges change, restricting their applicability and effectiveness.
Innovation Solution
A method involving frequency domain feature extraction, cepstral feature extraction, and cepstral mean normalization is employed to normalize physiological signals, allowing for removal of individual differences without restricting data distribution, using a cepstral mean normalization model trained on multiple subjects' data to standardize cepstral features based on mean and standard deviation parameters.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If min-max normalization is used to normalize physiological signals, then the normalization is simple to implement, but the method is unstable when data ranges change and requires redefinition of max and min values
Solution Approach 1:
The patent performs preliminary feature extraction (frequency domain features and cepstral features) before normalization to transform the raw physiological signals into a more stable feature space. This preliminary processing allows subsequent normalization to operate on features with more consistent statistical properties, reducing the need for frequent redefinition of normalization parameters when new data is added.
Solution Approach 2:
The patent changes the parameter space by transforming physiological signals from time domain to frequency domain and then to cepstral domain. This parameter transformation converts the original signal characteristics into features that are more suitable for normalization, making the normalization process more stable and less sensitive to data range changes.
2Reliability
If Z-Score normalization is used to normalize physiological signals, then the normalization handles data with varying ranges better, but the method requires assuming normal distribution of data
Solution Approach 1:
The patent transforms physiological signals into cepstral features through frequency domain analysis and cepstral transformation. This parameter change creates a new feature space where the data distribution becomes more suitable for normalization without requiring strict normal distribution assumptions, thereby improving both reliability and adaptability.
Solution Approach 2:
The patent introduces frequency domain features and cepstral features as intermediary representations between raw physiological signals and the final normalized data. These intermediate features serve as a mediator that transforms the original data into a form that is more amenable to normalization while preserving important physiological information.
3Productivity
If traditional normalization methods are used, then the processing is computationally efficient, but the methods restrict data distribution and have limited generalization ability
Solution Approach 1:
The patent applies parameter changes by transforming signals through frequency domain and cepstral domain representations. This creates features that capture essential physiological patterns while being more robust to individual differences and distribution variations, thereby improving generalization ability while maintaining computational efficiency through efficient FFT-based implementations.
Solution Approach 2:
The patent develops a normalization approach based on cepstral features that can be universally applied to different types of physiological signals and different subject populations. The method does not require retraining or parameter adjustment for different data distributions, providing multi-functional applicability across various physiological signal processing tasks.
Data Source
AI summary
Provided are a method and an apparatus for removing individual differences in physiological signals, an edge computing device, and a storage medium. The method includes: acquiring physiological signals of a plurality of subjects, where the physiological signals are acquired by a physiological signal acquisition device when the plurality of subjects perform a same task; extracting frequency domain features based on the physiological signals, where the frequency domain features are configured to characterize frequency characteristics of the physiological signals; extracting cepstral features based on the frequency domain features; and normalizing the cepstral features to obtain cepstral features of the physiological signals with at least part of individual differences among the plurality of subjects removed. The method according to the present disclosure is beneficial to normalizing physiological signals, and removing individual differences in the physiological signals without restricting the data distribution of physiological signals, thereby having a wider range of applications.