Removing spikes with standard deviation

Hey.

I'm trying to remove some spikes in a water current data using nanmean and nanstd.

To do it, i created this script:

Data3=Data2(:,:,5:6); %Data2 is the real data matrix. To don't take risks, i created a copy and work on it.

it=2;   

[~,col,page]=size(Data3);

for p=1:page
    for c=1:col
        for i=it+1:length(Data3)-it;
            x=nanmean(Data3(i-it:i+it,c,p)); 
            y=nanstd(Data3(i-it:i+it,c,p));
            if Data3(i,c,p)>x+3*y || Data3(i,c,p)<x-3*y; 
            Data3(i,c,p)=NaN;
            end
        end
    end
end
%clear p c r x y z k row col page it 

After running it, i did a plot to check if the original data(Data2) are now diferent from the one with no spikes(Data3)

plot(Data2(:,5,5))
hold
plot(Data3(:,5,1))

Despite this, i realized that they still equivalents and perfectly equals.

Why my code isn't working?

I appreciate any kind of help on it. Thanks to you all. This community is one of the best in the world!

Answer 1

[Star Strider]

Without the data, it is not possible to determine the reason the code does not work.

See if the rmoutliers function (introduced in R2018b) will do what you want.

[Gabriel Luca Pugliese Borges]

My matlab version is R2016b and i preferred to don't atualize because of my old computer. Is there a way to you copy and put in here the script of this function so i can try out?

[Star Strider]

That is prohibited, and for good reason.

If you have the Signal Processing Toolbox, see if the hampel function (introduced in R2015b) will do what you want.

It is not possible to determine the reason the code is not working without having the data it is used with. It could be that the ‘spikes’ are less than so nothing is replaced.

[Gabriel Luca Pugliese Borges]

Oops, thanks for warning me that!

I'll try the hampe1 function.

Besides that, you don't need the exactly same data that i used to test the command.

i created this matrix to test and it didn't work as well

[Star Strider]

I do not understand the indexing. If you want to create moving windows for the mean and standard deviation, the movmean and movstd functions (R2016a) would be easier. Also see the documentation section on Data Smoothing and Outlier Detection.

[Gabriel Luca Pugliese Borges]

Alright, Strider. Thank you so much for all helping and advices. I will test all the options that you gave me.

Wish you a good day.

[Star Strider]

My pleasure!

You, too!