Filtration, independent component, or wavelet? And why?
I have experimental data of skin temperature signal captured from a human subject when he was performing a task , I need to remove the artifact of this signal in order to do some statistical analysis based on the cleaned signal to know how this signal is significant for the task!
I found in literature that we could use:
1) low pass filter to remove high frequency noise.
2) band pass filter to remove high frequency noise and effect of breathing as this signal measured from human face!
3) independent component analysis to remove corrupted signal due to tracking process.
4) wavelet analysis.
Actually this signal results from tracking process of a video of IR camera, so I am afraid to build my statistical analysis on bad filtered signal, would you tell me what should i do and why?
Thanks in advance!
Before any filtering remove outliers. And before all try to avoid having the outliers by attaching the sensors well to the skin.
I deleted the outliers ,by deleting the values that are higher than mean + factor * standard deviation !
we measured the signal by tracking the Infrared Imaging Video ( it is a non-contact tool to measure the skin temperature)!
Your literature review is correct about the methods of filtration. However, I would go with wavelet analysis since it is a magnificent approach to remove the noise from high frequencies.
Dear Hassan Khaleel, actually i am a beginning student in digital signal processing , so would you recommend some references to study wavelet and would you give a small hint about the wavelet and independent component!
It very easy to find many references about wavelet as an approach and the way of composing the signal into different frequencies and remove the noise from the high frequency components. The best way to find and apply the functions of wavelet is in MATLAB where you can find them easily within the help of matlab and apply any function easily. Just go online and search about it and you will find many references and if you need further help with applying the wavelet just drop me your question and I will try to assist.
Dear Enas, you should use the Fourier transform to know the spectral information of your signals before any filtering or decomposition. As we know, the signals were detected by a non-contact IR camera, the interference or noise is also limited by your measurement, like the sampling rate. So I suggest that you have a clear undertand of your signal in time and frequency domain first of all.
Dear enas, I would add that you have to know the characteristics of the noise.
And before moving to the feature extraction and the classification parts, make sure
To evaluate your denoising methods by different evaluation criterias and check how much you have as PSR .. Gd luck!
Guessing for "PSR ..Gd "
PSR = Power Spectrum Residue (?)
Gd = Good, as @Sandy Rihana said "Gd luck" (Good luck)
Excuse me! sir Peter Husar,
I need to clarify what do you mean with signal properties?
sorry for that question, which could be at very beginning level of digital signal processing !
Dear Peter Husar,
what I know about my signal is:
1) sampling frequency is 25 HZ.
2) experiment paradigm: is to let subject a random sequence of 46 pictures with each picture takes 10 sec , what we are interested in is to study the significant of temperature signals regarding to different types of pictures the subject saw! , so from here it is important to study the temperature changes at least every 2 sec so 1/2=0.5 HZ .
3) the artifacts could be the breath rate which is within 18- 22 breathes per min for adult (1/(22/60)=2.7 HZ.
it seems to me that we are interested in frequency range from 0.5 HZ to 2 HZ
should i know any other signal properties?
thanks,
- Badges
- Science topic
- Similar topics
- Biological Science
- Physiology
- Biosignals