How to sort my signal based on the frequency domain? low to high

Here's a suggestion for enhancing the spectogram to pull out signals. More information about the feature signals would be helpful for extracting the energy at each timestep.

load('matlab.mat');
ICA_weights_right(ICA_weights_right<0) = 0;
t_s = 1:173;
f = 1:100;
% find the timesteps that have a strong signal - zero out the ones that do
% not have a strong signal
maskSpectrogram = ICA_weights_right;
zerolength = 20;
for ifeature = 1:100
    tmp = zeros(173,1);
    tmp(ICA_weights_right(:,ifeature)==0) = 1;
    tmpLabeled = bwlabel(tmp);
    stats = regionprops(tmpLabeled,'Area');
    idx = find([stats.Area] > zerolength);
    for k = 1:length(idx)
        maskSpectrogram(tmpLabeled == idx(k),ifeature) = 1;
    end
    tmp = zeros(173,1);
    tmp(ICA_weights_right(:,ifeature)>0) = 1;
    tmpLabeled = bwlabel(tmp);
    stats = regionprops(tmpLabeled,'Area');
    idx = find([stats.Area] > zerolength);
    for k = 1:length(idx)
        maskSpectrogram(tmpLabeled == idx(k),ifeature) = 1;
    end    
end
newSpectrogram = ICA_weights_right.*maskSpectrogram;
mask5x1 = ones(5,1);
newSpectrogram = imopen(newSpectrogram,mask5x1);
ICA_weights_right = ICA_weights_right'
newSpectrogram = newSpectrogram'
figure('color','white');
subplot(2,1,1)
imagesc(t_s,f,ICA_weights_right);
xlabel('time [s]');
ylabel('feature (frequency) index');
set(gca,'ydir','normal');
set(gca,'fontweight','bold','fontsize',14);
title('Spectrogram');
subplot(2,1,2)
imagesc(t_s,f,newSpectrogram);
xlabel('time [s]');
ylabel('feature (frequency) index');
set(gca,'ydir','normal');
set(gca,'fontweight','bold','fontsize',14);
title('Post-Processed Spectrogram');

4 个评论
显示 2更早的评论隐藏 2更早的评论

Meg Noah 2020-1-9

在 MATLAB Online 中打开

I made a mistake in interpretation. It's fixed above. You said there were 173 seconds of data. I had flipped the axes for frequency feature and time. It's now labeling the 173 seconds of data. What you have explained is the interpretation of the axis that has 100 samples, your 'feature' or 'frequency' axis. I don't know what frequencies correspond to the bins. Do you have more information? There should be some vector of 100 values for the '100 features from R-ICA' that presumably are given in Hz or rad/s. That is what is needed next. We can plot the significant features as functions of time. The cleaned up spectrogram is used to identify where the significan signals are.

totalAmplitude = sum(newSpectrogram');
idx = find(totalAmplitude > 2);
figure()
subplot(2,1,1)
for k = 1:length(idx)
    ifeature = idx(k);
    plot(t_s,ICA_weights_right(ifeature,:),'DisplayName',['Feature ' num2str(ifeature)]);
    hold on;
end
legend('location','best');
xlabel('time [s]');
ylabel('amplitude');
title('feature amplitude');
subplot(2,1,2)
for k = 1:length(idx)
    ifeature = idx(k);
    plot(t_s,newSpectrogram(ifeature,:),'DisplayName',['Feature ' num2str(ifeature)]);
    hold on;
end
legend('location','best');
xlabel('time [s]');
ylabel('amplitude');
title('feature amplitude after background noise removal');

CalebJones 2020-1-9

Your awesome.

Thank You

CalebJones 2020-2-7

Meg Noah

This approach was very helpful. And it has improved my research work by a big margin, however, is it possible to sort the entire feature space from low to high?

So that I can choose first 10 which is low freq and last 10 which is high freq.

What way the code can be tweaked to produce such an output?

请先登录，再进行评论。