noise filter and pc filtering

14
NOISE FILTER AND PC FILTERING UFO GROUP LUISA MARIA GRAZIA SILVIA VITO

Upload: diamond

Post on 20-Jan-2016

40 views

Category:

Documents


0 download

DESCRIPTION

NOISE FILTER AND PC FILTERING. UFO GROUP. LUISAMARIA GRAZIASILVIAVITO. OUTLINE. Presence of noise. FFT filter. PCA filter. Conclusions. PRESENCE OF NOISE. Noise affecting data is due to several factors. Correlated noise (calibration, instrumental vibration) Detectors noise: - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: NOISE FILTER AND  PC FILTERING

NOISE FILTER AND PC FILTERING

UFO GROUP

LUISA MARIA GRAZIA SILVIA VITO

Page 2: NOISE FILTER AND  PC FILTERING

OUTLINEOUTLINE

Presence of noise.Presence of noise.

FFT filter.FFT filter.

PCA filter.PCA filter.

Conclusions.Conclusions.

Page 3: NOISE FILTER AND  PC FILTERING

PRESENCE OF NOISEPRESENCE OF NOISE

Noise affecting data is due to several Noise affecting data is due to several factors.factors.– Correlated noise (calibration, instrumental Correlated noise (calibration, instrumental

vibration)vibration)– Detectors noise:Detectors noise:

randomrandomgaussiangaussiantime and space decorrelationtime and space decorrelation

Page 4: NOISE FILTER AND  PC FILTERING

FFT FILTER ON AIRS DATAFFT FILTER ON AIRS DATA

Selection of two neighbor wavelengths with Selection of two neighbor wavelengths with

close Radiances.close Radiances. Difference of the selected wavelengths for Difference of the selected wavelengths for

the same image.the same image. FFT: noise corresponds to higher FFT: noise corresponds to higher

frequencies.frequencies.

Page 5: NOISE FILTER AND  PC FILTERING

FFT

R(1)-R(2)

It displays popping noise

Frequencies

Page 6: NOISE FILTER AND  PC FILTERING

FFT FILTER RESULTSFFT FILTER RESULTSThis filter doesn’t work!!

Page 7: NOISE FILTER AND  PC FILTERING

PCA – Principal Component PCA – Principal Component AnalysisAnalysis

It allows to reduce the It allows to reduce the dimensionality of the problem.dimensionality of the problem.

It finds the Principal It finds the Principal components (the eigenvectors components (the eigenvectors of the observation covariance of the observation covariance matrix).matrix).

Most of atmospheric of the Most of atmospheric of the observed variance is in the first observed variance is in the first PCs. PCs.

Right number of PCs allows for Right number of PCs allows for noise reduction and information noise reduction and information preservation.preservation.

Page 8: NOISE FILTER AND  PC FILTERING

UNFILTERED DATA101 AIRS SPECTRA

Page 9: NOISE FILTER AND  PC FILTERING

STANDARD NOISESTANDARD NOISE

RANGE CONSIDERED: 700-1100 cm-1

Page 10: NOISE FILTER AND  PC FILTERING

FILTER SENSITIVITY TO THE FILTER SENSITIVITY TO THE NUMBER OF PCsNUMBER OF PCs

# PCs =10 # PCs =175

X 5

MEAN OF NORMALIZED DIFFERENCEBETWEEN ESTIMATED AND STANDARD NOISE

Page 11: NOISE FILTER AND  PC FILTERING

RESULTS - # PCs = 10 RESULTS - # PCs = 10

Gaussian noise

Page 12: NOISE FILTER AND  PC FILTERING

RESULTS - # PCs = 175 RESULTS - # PCs = 175

Gaussian noise

Page 13: NOISE FILTER AND  PC FILTERING

FILTERED SPECTRAFILTERED SPECTRA

UNFILTERED SPECTRA

# PCs = 175 # PCs = 10

Page 14: NOISE FILTER AND  PC FILTERING

……that’s all folks!!that’s all folks!!

THANKS FOR YOUR THANKS FOR YOUR ATTENTION!!ATTENTION!!