scene class recognition using high resolution sar/insar spectral decomposition methods

22
SCENE CLASS RECOGNITION USING HIGH RESOLUTION SAR/INSAR SPECTRAL DECOMPOSITION METHODS Anca Popescu, Inge Gavat University Politehnica Bucharest (UPB) Mihai Datcu German Aerospace Center (DLR) IGARSS 2011 24-29 July 2011, Vancouver, Canada University POLITEHNICA Bucharest Faculty of Electronics, Telecommunications and Information Technology

Upload: duer

Post on 16-Jan-2016

50 views

Category:

Documents


0 download

DESCRIPTION

University POLITEHNICA Bucharest Facult y of Electronics, Telecommunication s and Information Technology. SCENE CLASS RECOGNITION USING HIGH RESOLUTION SAR/INSAR SPECTRAL DECOMPOSITION METHODS. Anca Popescu, Inge Gavat University Politehnica Bucharest (UPB) Mihai Datcu - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: SCENE CLASS RECOGNITION USING HIGH RESOLUTION SAR/INSAR SPECTRAL DECOMPOSITION METHODS

SCENE CLASS RECOGNITION USINGHIGH RESOLUTION SAR/INSAR SPECTRAL

DECOMPOSITION METHODS

Anca Popescu, Inge GavatUniversity Politehnica Bucharest (UPB)

Mihai DatcuGerman Aerospace Center (DLR)

IGARSS 2011

24-29 July 2011, Vancouver, Canada

University POLITEHNICA BucharestFaculty of Electronics, Telecommunications and Information Technology

Page 2: SCENE CLASS RECOGNITION USING HIGH RESOLUTION SAR/INSAR SPECTRAL DECOMPOSITION METHODS

Motivation: High resolution scene category indexing, large number of structures visible in urban sites

Non – parametric feature extraction methods

Page 3: SCENE CLASS RECOGNITION USING HIGH RESOLUTION SAR/INSAR SPECTRAL DECOMPOSITION METHODS

Summary

• Introduction

• Data descriptors

• Spectral features

• Spectral components

• Experimental data

• Classification results

• Conclusions

Page 4: SCENE CLASS RECOGNITION USING HIGH RESOLUTION SAR/INSAR SPECTRAL DECOMPOSITION METHODS

Introduction

Study of value adding processing methods for SLC and InSAR data, for scene class recognition

Page 5: SCENE CLASS RECOGNITION USING HIGH RESOLUTION SAR/INSAR SPECTRAL DECOMPOSITION METHODS

Direct Estimation from image spectra

Regularization based estimation +

Bayesian Model order selection

Consistency validation

Introduction

Study of value adding processing methods for SLC and InSAR data, for scene class recognition

Concept: Make use of the information contained in the phase of the SAR signal

Page 6: SCENE CLASS RECOGNITION USING HIGH RESOLUTION SAR/INSAR SPECTRAL DECOMPOSITION METHODS

Data descriptors – spectral features

Direct estimation from spectra based on spectral differences

Page 7: SCENE CLASS RECOGNITION USING HIGH RESOLUTION SAR/INSAR SPECTRAL DECOMPOSITION METHODS

Data modelThe 2-D signal model:

),()}(2exp{1

, nnenfnfjyK

kkkknn

Where kkk jba

kk ff ,

),( nne

complex amplitude of the kth sinusoid

unknown frequencies of the kth sinusoid

2-D noise

Data descriptors spectral components estimation

,k kk ff ,

Problem: Estimate the parameters of the sinusoidal signals:

Page 8: SCENE CLASS RECOGNITION USING HIGH RESOLUTION SAR/INSAR SPECTRAL DECOMPOSITION METHODS

J. Li, P. Stoica: “Efficient Mixed-Spectrum Estimation with Applications to Target Feature Extraction”, IEEE Transactions on Signal Processing, 44, 1996, 281-295

2

12 ||)}(2exp{),(|| F

K

kkkk nfnfjnnyC

Model choice: minimize NLS criterion:

Peak of the 2-D periodogram

221

0

2, |),(|maxarg)ˆ,ˆ( nkfj

N

n

nkfjkkfkfkk eennyff

1

0

22),(1 N

n

nkfjnkfjkk eenny

NN

1

0

2|}2exp{}2exp{),(|N

nkkkkk nfjnfjnnyg

Algorithm preparations: if αk and fk are known, minimize cost function for the kth sinusoid is:

Height of the peak (complex)

Data descriptors spectral components estimation

Page 9: SCENE CLASS RECOGNITION USING HIGH RESOLUTION SAR/INSAR SPECTRAL DECOMPOSITION METHODS

Y1 ampl 1 freq 1

…………….

Y2 ampl 1 freq 1

…………….

Y3 ampl 1 freq 1

ampl 2 freq 2

ampl 3 freq 3

…………….

ampl 2 freq 2

STEP 1STEP 2STEP 3

…………...

................

)'(cos.9

),(ˆ,ˆ;ˆ,ˆ,ˆ,ˆ),(.8

),(ˆ,ˆ;ˆ,ˆ,ˆ,ˆ),(.7

),(ˆ,ˆ;ˆ,ˆ,ˆ,ˆ),(3.6

)'(cos.5

ˆ,ˆ),(.4

),(ˆ,ˆ.3

ˆ,ˆ),(2.2

),(ˆ,ˆ1.1

22211332

11122331

33322113

221

222

112

11

tlysignificanchangetdoesnfunctionteconvergencpracticaluntilIterate

nnyfromobtainedareffffromobtainedisnny

nnyfromobtainedareffffromobtainedisnny

nnyfromobtainedareffffromobtainedisnnyK

tlysignificanchangetdoesnfunctionteconvergencpracticaluntilIterate

ffromobtainedisnny

nnyfromobtainedaref

ffromobtainedisnnyK

nnyfromobtainedarefK

Data descriptors spectral components estimation

Page 10: SCENE CLASS RECOGNITION USING HIGH RESOLUTION SAR/INSAR SPECTRAL DECOMPOSITION METHODS

AKAIKE Information Criterion – model order selection

Estimates the expected Kullback-Leibler information between the model generating the data and a candidate model

)))](ˆ|([log(max yxgEE xyg

kyL ))|ˆ(log(

kyLAIC 2))|ˆ(log(2

Model selection: = parameter to be estimated from empirical data y

Best Model: Minimum AIC value

y = generated from f(x), X is a random variableLog likelihood function for model selection:

Kullback-leibler Information(distance between models)

Maximized log-likelihood(parameter estimation)

Data descriptors spectral components estimation

Page 11: SCENE CLASS RECOGNITION USING HIGH RESOLUTION SAR/INSAR SPECTRAL DECOMPOSITION METHODS

Goal: asses parameter’s capability to discriminate scene classes

Evaluation: Accuracy = (TP+TN) / (TP+TN+FP+FN)

Methodology for scene class indexing

Page 12: SCENE CLASS RECOGNITION USING HIGH RESOLUTION SAR/INSAR SPECTRAL DECOMPOSITION METHODS

Test Site – Bucharest, Romania

TerraSAR-X High Resolution Spotlight: LAN 130

Page 13: SCENE CLASS RECOGNITION USING HIGH RESOLUTION SAR/INSAR SPECTRAL DECOMPOSITION METHODS

SLCProduct type: HS SSC Master: HS SSC Slave: HS SSC

Acquisition date: 30.09.2008 Acquisition date: 11.10.2008 Acquisition date: 30.09.2008Ground range resolution: 0.8905 m Ground range resolution: 0.8905 m Ground range resolution: 0.8905 m

Azimuth resolution: 1.1000 m Azimuth resolution: 1.1000 m Azimuth resolution: 1.1000 mNr. of looks: 1 Nr. of looks: 1 Nr. of looks: 1

Interferometric pair

Frequency of classes in database: dominant classes: tall blocks, green areas, urban fabric, commercial and industrial sites

Experimental data – TSX LAN-130Test Site – Bucharest, Romania

Page 14: SCENE CLASS RECOGNITION USING HIGH RESOLUTION SAR/INSAR SPECTRAL DECOMPOSITION METHODS

1650 SLC patches, 200 x 200 m

Water course/ water bodyStadion

Very tall buildingTall Block

Industrial Site

Test Site – Patch database formation

Interferometric data

Page 15: SCENE CLASS RECOGNITION USING HIGH RESOLUTION SAR/INSAR SPECTRAL DECOMPOSITION METHODS

Spectral Centroid, Flux,and Rolloff (azimuth and range)

Mean and variance of most significant 3 cepstral coefficients

Results – Spectral and cepstral parameters

Page 16: SCENE CLASS RECOGNITION USING HIGH RESOLUTION SAR/INSAR SPECTRAL DECOMPOSITION METHODS

Relax parameter αk (modulus representation), selection of six random components from the estimated stack

Results – Spectral components

Dense urban area, mostly tall blocks

Green area, small vegetation

Page 17: SCENE CLASS RECOGNITION USING HIGH RESOLUTION SAR/INSAR SPECTRAL DECOMPOSITION METHODS

Reconstructed data from estimated spectral components and original SAR patch

Results – Spectral Components

Page 18: SCENE CLASS RECOGNITION USING HIGH RESOLUTION SAR/INSAR SPECTRAL DECOMPOSITION METHODS

0 5 10 15 20 250.94

0.95

0.96

0.97

0.98

0.99

1

Class

TNR Spectral features SLC

Results – Scene Class RecognitionSLC/InSAR

True Negative Rate indicator

0 5 10 150.94

0.95

0.96

0.97

0.98

0.99

1

Class

TNR InSAR Spectral features

23 scene classes discoverable with SLC database

15 scene classes discoverable with InSAR database

Page 19: SCENE CLASS RECOGNITION USING HIGH RESOLUTION SAR/INSAR SPECTRAL DECOMPOSITION METHODS

Results – Scene Class RecognitionSLC – Spectral Features and Spectral

ComponentsAccuracy indicator

0 5 10 15 20 250.88

0.9

0.92

0.94

0.96

0.98

1

Accuracy Spectral features SLC

Accuracy Spectral Components SLC

Class

Spectral components

Spectral features

Page 20: SCENE CLASS RECOGNITION USING HIGH RESOLUTION SAR/INSAR SPECTRAL DECOMPOSITION METHODS

Influence of interferogram spectral features on classfication accuracy

Results – Scene Class RecognitionSLC / InSAR

0 5 10 15 20 250.9

0.91

0.92

0.93

0.94

0.95

0.96

0.97

0.98

0.99

1

Class

Accuracy InSAR Spectral Features

Accuracy SLC Spectral Features

Large roadGreen

urban area

Sparsely vegetated areas

Very tall building

Tall BlockInSAR

SLC

Page 21: SCENE CLASS RECOGNITION USING HIGH RESOLUTION SAR/INSAR SPECTRAL DECOMPOSITION METHODS

• Evaluation of the capability of spectral parameters to discriminate scene classes for complex HR TerraSAR-X data

• Spectral estimation method for high resolution data characterization and reconstruction, based on RELAX algorithm.

• Evaluation of the capability of spectral components to discriminate scene classes for complex HR TerraSAR-X data

• Accuracy of recognition better than 80% for main class training

• Acknowledgements to DLR team for providing the interferometric data: Nico Adam, Christian Minet, Nestor Yague-Martinez , Helko Breit.

Conclusions

Page 22: SCENE CLASS RECOGNITION USING HIGH RESOLUTION SAR/INSAR SPECTRAL DECOMPOSITION METHODS

Thank you for your attention!

[email protected]