19
Remote Sensing Laboratory Dept. of Information Engineering and Computer Science University of Trento Via Sommarive, 14, I-38123 Povo, Trento, Italy Lorenzo Bruzzone Lorenzo Bruzzone Francesca Bovolo Francesca Bovolo A SEMANTIC-BASED MULTILEVEL APPROACH TO CHANGE DETECTION IN VERY HIGH GEOMETRICAL RESOLUTION MULTITEMPORAL IMAGES E-mail: [email protected] E-mail: [email protected] Web page: http://rslab.disi.unitn.it Web page: http://rslab.disi.unitn.it

Lorenzo Bruzzone Francesca Bovolo

  • Upload
    liv

  • View
    73

  • Download
    0

Embed Size (px)

DESCRIPTION

A SEMANTIC-BASED MULTILEVEL APPROACH TO CHANGE DETECTION IN VERY HIGH GEOMETRICAL RESOLUTION MULTITEMPORAL IMAGES. Lorenzo Bruzzone Francesca Bovolo. E-mail: [email protected] Web page: http://rslab.disi.unitn.it. Outline. Introduction on change detection in VHR images. 1. - PowerPoint PPT Presentation

Citation preview

Page 1: Lorenzo  Bruzzone Francesca  Bovolo

Remote Sensing LaboratoryDept. of Information Engineering and Computer Science

University of TrentoVia Sommarive, 14, I-38123 Povo, Trento, Italy

Remote Sensing LaboratoryDept. of Information Engineering and Computer Science

University of TrentoVia Sommarive, 14, I-38123 Povo, Trento, Italy

Lorenzo BruzzoneLorenzo BruzzoneFrancesca BovoloFrancesca Bovolo

A SEMANTIC-BASED MULTILEVEL APPROACH TO CHANGE DETECTION IN VERY HIGH GEOMETRICAL RESOLUTION

MULTITEMPORAL IMAGES

A SEMANTIC-BASED MULTILEVEL APPROACH TO CHANGE DETECTION IN VERY HIGH GEOMETRICAL RESOLUTION

MULTITEMPORAL IMAGES

E-mail: [email protected]: [email protected] page: http://rslab.disi.unitn.itWeb page: http://rslab.disi.unitn.it

Page 2: Lorenzo  Bruzzone Francesca  Bovolo

University of Trento, Italy

Outline

2Lorenzo Bruzzone, Francesca Bovolo

Introduction on change detection in VHR images

General approach to change detection in VHR images

Experimental results

1

Conclusion

Illustration on the use of the approach for the solution of a specific change detection problem

2

3

4

5

Page 3: Lorenzo  Bruzzone Francesca  Bovolo

University of Trento, Italy

Main assumption: unsupervised change-detection techniques generally assume that multitemporal images are similar to each other except for the presence of changes occurred on the ground.

Problems: This assumption is seldom satisfied in VHR images due to:

the complexity of the objects present in the scene (which may show different spectral behaviors at two different dates even if their semantic meaning does not change);

the differences in the acquisition conditions (e.g., sensor acquisition geometry, atmospheric and sunlight conditions, etc.).

Introduction: Change Detection in VHR Images

3Lorenzo Bruzzone, Francesca Bovolo

Page 4: Lorenzo  Bruzzone Francesca  Bovolo

University of Trento, Italy

July 2006 October 2005

Quickbird images acquired on a portion of the city of Trento (Italy)

4Lorenzo Bruzzone, Francesca Bovolo

Introduction: Change Detection in VHR Images

Page 5: Lorenzo  Bruzzone Francesca  Bovolo

University of Trento, Italy

Aim of the Work

5Lorenzo Bruzzone, Francesca Bovolo

We propose a general top-down approach to the definition of the architecture of change detection methods for multitemporal VHR images.

The proposed approach:

explicitly models the presence of different radiometric changes on the basis of the properties of the considered images

extracts the semantic meaning of changes;

identifies changes of interest with strategies designed on the basis of the specific application;

exploits the intrinsic multiscale properties of the objects and the high spatial correlation between pixels in a neighborhood.

Page 6: Lorenzo  Bruzzone Francesca  Bovolo

University of Trento, Italy 6Lorenzo Bruzzone, Francesca Bovolo

Proposed Approach: Architecture Design

Multitemporal data set

Identification of the tree ofradiometric changes

Direct extraction ofchanges of interest

Refined detection of the radiometricchange of interest

Change detection map

Differential extraction of changesof interest by cancellation

Selection of thestrategy for detectingchanges of interest

Auxiliaryinformation

Detection of allradiometric changes

Detection of the changes of interest

Page 7: Lorenzo  Bruzzone Francesca  Bovolo

University of Trento, Italy 7Lorenzo Bruzzone, Francesca Bovolo

Changes due to acquisition

conditions (Acq)

Differences in atmospheric

conditions (Atm)

Differences in acquisition

system (Sys)

Changes occurred on the ground (Grd)

Vegetation Phenology (veg)

Anthropic activity (Ant)

Natural disasters (Dis)

Environmental conditions (Env)

Radiometric Changes(rad)

Sensor view angle

Sensor acquisition

mode

Type of sensor

Seasonal effects

Identification of the Tree of Radiometric Changes

Page 8: Lorenzo  Bruzzone Francesca  Bovolo

University of Trento, Italy 8Lorenzo Bruzzone, Francesca Bovolo

Proposed Approach: Architecture Design

Multitemporal data set

Identification of the treeof radiometric changes

Direct extraction ofchanges of interest

Refined detection of the radiometricchange of interest

Change detection map

Differential extraction of changesof interest by cancellation

Selection of thestrategy for detectingchanges of interest

Auxiliaryinformation

Detection of allradiometric changes

Detection of the changes of interest

Change Vector Analysis, Context-sensitive techniques, etc.

Page 9: Lorenzo  Bruzzone Francesca  Bovolo

University of Trento, Italy 9Lorenzo Bruzzone, Francesca Bovolo

Detection of Changes of Interest

Refined detection of theradiometric change of interest

Non-relevantchange 1

Detection of radiometric changes

Non-relevantchange 2

Non-relevantchange N

-+

X1 X2

Direct detection of changes of interest Differential detection by cancellation

Detection ofchange of interest 1

Detection ofchange of interest K

X1 X2

- -+ +

+ +

Map of changes Map of changes

Page 10: Lorenzo  Bruzzone Francesca  Bovolo

University of Trento, Italy 10Lorenzo Bruzzone, Francesca Bovolo

O1 O2

P1 P2

X1 X2

Meta-levelsfusion

Map of a specific Radiometric change

Pixel radiometry

Geometric or statistic primitives

Classification map, object map,…

Multilevel Architecture: Semantic of Changes

Pixel Meta-level (px)

Primitive Meta-level (p)

Object Meta-level (o)

j=1,…,Jpx

j=1,…,Jp

j=1,…,Jo O

P

Page 11: Lorenzo  Bruzzone Francesca  Bovolo

University of Trento, Italy

October 2004 July 2006 Reference Map

Data Set Description

Study area: South part of Trento (Italy).

Multitemporal data set: portion (380×430 pixels) of two images acquired by the Quickbird satellite in October 2004 and July 2006.

Causes of Change: changes on the ground, seasonal changes, registration noise.

Page 12: Lorenzo  Bruzzone Francesca  Bovolo

University of Trento, Italy 12Lorenzo Bruzzone, Francesca Bovolo

Proposed Approach: Architecture Design

Multitemporal data set

Identification of the tree ofradiometric changes

Direct extraction ofchanges of interest

Refined detection of the radiometricchange of interest

Change detection map

Differential extraction of changesof interest by cancellation

Selection of thestrategy for detectingchanges of interest

Auxiliaryinformation

Detection of allradiometric changes

Detection of the changes of interest

Change Vector Analysis, Context-sensitive techniques, etc.

Page 13: Lorenzo  Bruzzone Francesca  Bovolo

University of Trento, Italy

Identification of the Tree of Radiometric Changes

13Lorenzo Bruzzone, Francesca Bovolo

Rad

sh rn

Sys Grd

Veg Ant

atgl b

Grassland Newbuildings

Shadowchanges

Appletrees

Registrationnoise

Page 14: Lorenzo  Bruzzone Francesca  Bovolo

University of Trento, Italy

Changes Tree and Detection Strategy

14Lorenzo Bruzzone, Francesca Bovolo

Rad

sh rn

Sys Grd

Shadowchanges

Registrationnoise

Identification of the tree of radiometric changes

Refined detection of Grd

Detectionof sh

Detection of radiometricChanges (CVA)

Detectionof rn

-+

X1 X2

-+

Differential detection by cancellation

Map of changes

Page 15: Lorenzo  Bruzzone Francesca  Bovolo

University of Trento, Italy

Multilevel Representation of Radiometric Changes

15Lorenzo Bruzzone, Francesca Bovolo

X1 X2

Pix

el M

eta

-leve

l (px

)P

rimiti

ve M

eta-

leve

l (p)

Magnitude of multispectralchange vectors

Shadow changeindex

Parcel map

Registrationnoise map

Image radiometry

Shadow Index

Segmentation map

S. Marchesi, F. Bovolo, L. Bruzzone, “A Context-Sensitive Technique Robust to Registration Noise for Change Detection in VHR Multispectral Images”, IEEE Transactions on Image Processing, Vol. 19, pp. 1877-1889, 2010.

F. Bovolo, “A Multilevel Parcel-Based Approach to Change Detection in Very High Resolution Multitemporal Images,” IEEE Geoscience and Remote Sensing Letters, Vol. 6, No. 1, pp. 33-37, January 2009.

L. Bruzzone and D. Fernández-Prieto, "Automatic Analysis of the Difference Image for Unsupervised Change detection," IEEE Trans. Geosci. Rem. Sens., vol. 38, pp. 1170-1182, 2000.

V. J. D. Tsai, "A comparative study on shadow compensation of color aerial images in invariant color models," IEEE Trans. Geosci. Remote Sens., vol. 44, pp. 1661-1671, 2006.

Page 16: Lorenzo  Bruzzone Francesca  Bovolo

University of Trento, Italy 16Lorenzo Bruzzone, Francesca Bovolo

Proposed Approach: Block Scheme

X1

X2

Shadowdetection

Parceldetection

Multiscale analysisfor rn detection

CVA

Comparisonsh

detection

rad

detection

={nc, Grd}

Change-detectionmap

Magnitude of multispectral

change vectors

Shadow changeindex

Shadow index

- -+

Page 17: Lorenzo  Bruzzone Francesca  Bovolo

University of Trento, Italy 17Marzo 2011Silvia Demetri

TechniqueFalse

AlarmsMissed Alarms

Total Errors

Overall accuracy (%)

CVA pixel-based 5005 9924 14929 90.86

CVA parcel-based 3537 10261 13798 91.56

Proposed method 1470 8480 9950 93.91

Experimental Results

95

90

85

80

Overall change detection accuracy (%)

90.8691.56

93.91

CVAPixel-based

CVA parcel-based

Proposedmethod

Page 18: Lorenzo  Bruzzone Francesca  Bovolo

University of Trento, Italy 18Marzo 2011Silvia Demetri

Reference Map

Change Detection mapCVA parcel based

Change detection mapProposed approach

October 2005 July 2006

Experimental Results

Page 19: Lorenzo  Bruzzone Francesca  Bovolo

University of Trento, Italy

We presented a general top-down approach to the definition of the architecture of change detection methods for multitemporal VHR images.

The main concepts exploited for the definition of the change detection architecture are: Modeling the types of radiometric changes expected between images; Extracting the semantic meaning from radiometric changes.

The approach proposed includes: Direct detection of changes of interest or differential cancellation of

uninteresting radiometric changes; Multilevel and context-sensitive techniques; Iterative strategy.

The approach has been successfully applied to the definition of aneffective architecture for change detection between Quickbird images in different application scenarios.

Conclusion

19Lorenzo Bruzzone, Francesca Bovolo