It is to ask how to Extract features from a signal in matlab

Question

TESFAYE YIMAM MOHAMMAD il 14 Giu 2022

0
Link

Link diretto a questa domanda

https://it.mathworks.com/matlabcentral/answers/1740400-it-is-to-ask-how-to-extract-features-from-a-signal-in-matlab

Risposto: Yash Sharma il 20 Set 2023

I have a project to do. But I am a newbie in matlab.

***

Let me say somethig about the dataset.

Pulse Transit Time (PTT) is the time that pulse waves require to travel in blood vessels between two

sites. It is an important indicator for many medical properties and recent research focused on its

potential for blood pressure monitoring.

A total of 22 participants were involved in the data collection while performing three different activities (‘sit’, ‘walk’, ‘run’). The dataset contains 66 recordings. A recording contains 11 time series variables, each representing a physiological signal for more than 40,000

heartbeats.Each signal variable(column) has min 242K samples.

******

The idea is to extract features( like mean, min, max, std, median) from a signal. BUt the professor was telling us to extract those features with a sliding window with 50% of overlap.

1) Having such a big samples on every signal(column) what is the recommended window size?

2) what is the point of overlapping?

2) How is the sliding window implemented with a 50% of overlapping?

3) How are the extracted features are stored.

***

Anyone who can give me a starting point will be appreciated.

0 Commenti
Mostra -2 commenti meno recentiNascondi -2 commenti meno recenti

Accedi per commentare.

Accedi per rispondere a questa domanda.

Answer 1

Yash Sharma il 20 Set 2023

0
Link

Link diretto a questa risposta

https://it.mathworks.com/matlabcentral/answers/1740400-it-is-to-ask-how-to-extract-features-from-a-signal-in-matlab#answer_1314242

I understand that you have large amount of time series data and you want to extract features like mean, min, max, std, median using sliding window protocol.

The recommended window size for feature extraction depends on the signal's characteristics and analysis goals.
Choose a window size that captures meaningful patterns while remaining computationally feasible.
Start with a window size of several thousand samples (e.g., 5000 or 10000).
Adjust the window size based on analysis results and requirements.
Overlapping in a sliding window approach ensures that adjacent sections of the signal are included in the extracted features.
Overlapping allows for a smoother transition between windows and avoids potential information loss at the boundaries, it also captures temporal relationships that can improve accuracy.

Here’s an example code on how you can implement sliding window protocol with 50% overlap on timeseries data.

% Parameters
windowSize = 5000; % Window size
overlap = windowSize / 2; % Overlap size
% Generate example time series data
timeSeriesData = rand(1, 242000); % Replace with your actual time series data
% Initialize variables
numWindows = floor((length(timeSeriesData) - windowSize) / overlap) + 1;
features = zeros(numWindows, 5); % Assuming 5 features to be extracted
% Sliding window protocol
startIndex = 1;
for i = 1:numWindows
    endIndex = startIndex + windowSize - 1; 
    % Extract features from the current window   
    windowData = timeSeriesData(startIndex:endIndex);    
    feature1 = mean(windowData);    
    feature2 = min(windowData);    
    feature3 = max(windowData);    
    feature4 = std(windowData);    
    feature5 = median(windowData);    
    % Store the extracted features    
    features(i, :) = [feature1, feature2, feature3, feature4, feature5];    
    % Slide the window    
    startIndex = startIndex + overlap;    
end
% Display the extracted features
disp(features); 