how to automatic segmentation signal

D = load('matlab.mat');
data = D.data;
Fs = D.fs;
t = linspace(0, 1, numel(data))/Fs;
cp = findchangepts(data, 'Statistic','std', 'MinDistance',2E+3, 'MaxNumChanges',7);
figure
plot(t, data)
hold on
YL = ylim;
plot([t(cp); t(cp)], YL(:)*ones(1,numel(cp)), '-r', 'LineWidth',1)
hold off

producing:

how to automatic segmentation signal - 2019 01 22.png

The ‘cp’ vector are the change points corresponding to the vertical red lines in the plot. The findchangepts function takes about 15 seconds to complete this (on my desktop, using tic and toc),

Experiment to get the result you want.

6 个评论
显示 4更早的评论隐藏 4更早的评论

Star Strider 2019-1-22

在 MATLAB Online 中打开

My pleasure.

After thinking about it later, tell findchangpts to create 8 change points, not 7. That will segment the first set as well, making the analysis easier:

cp = findchangepts(data, 'Statistic','std', 'MinDistance',2E+3, 'MaxNumChanges',8);

I assume 8 change points in my discussion of the indices and time calculations here.

The ‘cp’ output of findchangepts are indices into the argument vector, so to change them into time values, just subscript them such as I did in the second plot call. I would work with them as indices first, then convert them to time values. You will need to decide how to write your code to get the values you want. Consider using the diff (link) function for some of those, although on the itme values, not the index values.

So for example, to get the first set, use ‘cp(1)’ and ‘cp(2)’, the second, ‘cp(3)’ and ‘cp(4)’, and so on for all the intervals you want to analyse. The times would be for example ‘t(cp(1))’, and so for the others. Calculate the intervals from the time differences, not the index differences.

Star Strider 2019-1-22

编辑：Star Strider 2019-1-23

在 MATLAB Online 中打开

My pleasure.

Using this is significantly faster, and appears to be as accurate:

cp = findchangepts(data, 'Statistic','mean', 'MinDistance',4E+3, 'MaxNumChanges',8);

Calculate the absolute times as:

tv = [t(1) t(cp)]                                                   % Time Values (Segment Times)

and the intervals between successive change point segments as:

timediff = diff([0 tv])                                             % Serial Intervals (Segment Difference Times)

With your data, the results are:

Segment Times =    0.00000E+00    1.28867E-05    2.52191E-05    3.53367E-05    4.78518E-05    5.74961E-05    7.07605E-05    8.01827E-05
Segment Difference Times =    0.00000E+00    1.28867E-05    1.23324E-05    1.01175E-05    1.25151E-05    9.64426E-06    1.32645E-05    9.42215E-06

You then have to write the necessary code to find the other intervals you are interested in. Use the ‘tv’ (‘Segment Times’) vector for those calculations.

EDIT —

Added plot figure.

The plot: