Advances in the Processing of VHR Optical Imagery in Support of Safeguards Verification

Size: px

Start display at page:

Download "Advances in the Processing of VHR Optical Imagery in Support of Safeguards Verification"

Mark Anthony
6 years ago
Views:

Member of the Helmholtz Association Symposium on International Safeguards: Linking Strategy, Implementation and People IAEA-CN220, Vienna, Oct 20-24, 2014 Session: New Trends in Commercial Satellite

1 Member of the Helmholtz Association Symposium on International Safeguards: Linking Strategy, Implementation and People IAEA-CN220, Vienna, Oct 20-24, 2014 Session: New Trends in Commercial Satellite Imagery Advances in the Processing of VHR Optical Imagery in Support of Safeguards Verification October 21, 2014 Irmgard Niemeyer, Clemens Listner, Mort Canty International Safeguards Group Institute of Energy and Climate Research IEK-6: Nuclear Waste Management and Reactor Safety Forschungszentrum Jülich GmbH, Germany

2 Advances in the Processing of VHR Optical Imagery in Support of Safeguards Verification 1 Introduction 2 3 Methods Examples 4 Summary & Outlook Slide 2

3 Advances in the Processing of VHR Optical Imagery in Support of Safeguards Verification 1 Introduction 2 3 Methods Examples 4 Summary & Outlook Slide 3

4 Very high spatial resolution optical imaging sensors (< 1m) from 2000 Slide 4

5 Very high spatial resolution optical imaging sensors (<1m): WorldView-2 DigitalGlobe Slide 5

6 Very high spatial resolution optical imaging sensors (<1m): WorldView-2 DigitalGlobe Slide 6

Satellite Imagery in Support of Safeguards Verification Commercial high resolution imagery is routinely

planning; Verify the accuracy and completeness of information supplied by Member States; Detect changes

7 Satellite Imagery in Support of Safeguards Verification Commercial high resolution imagery is routinely used within the IAEA s safeguards system as a reference source to: Aid in in-field and inspection planning; Verify the accuracy and completeness of information supplied by Member States; Detect changes and monitor activities at nuclear sites; Investigate undeclared activities; Provide analytical input to the State Evaluation process. Geoinformation technologies further advance the use of satellite imagery to generate site plans and store and manage critical information related to sites and facilities. [Steinmaus et al. 2013] Slide 7

8 Processing of Satellite Imagery in Support of Safeguards Verification Optical / Multispectral (VIS, NIR, MIR) Hyperspectral Sensors (VIS- MIR) Thermal Infrared Sensors (TIR) Synthetical Aperture Radar Sensors (SAR) Pre- processing Feature Extraction Classification Change Detection Anomaly Detection 3D Modelling 3D Modelling Slide 8

9 Advances in the Processing of VHR Optical Imagery in Support of Safeguards Verification 1 Introduction 2 3 Methods Examples 4 Summary & Outlook Slide 9

10 Object- based change detection Comparison of two satellite imagery datasets Object-based approach Adapted segmentation algorithm for bitemporal datasets PCA and IR-MAD for transformation of the feature space Threshold selection and unsupervised classification Plugins for ecognition Preprocessing Segmentation Feature Extraction Transformation Classification Postprocessing Slide 10

11 Problems of previous approaches Segmentation stability and workflow integration Use of shape features Correlated object features Detection of relevant changes Segmentation of two images that are identical except for Gaussian noise (μ=0,σ=0.1) Slide 11

12 Multiresolution Segmentation (MRS)* Region-oriented approach Start with chessboard segmentation Select segment X (seed) and merge it with neighbor Y, if dxy (, ) = min ( dxz (, )), Z N( X) dy (, X) = min ( dyz (, )), Z NY ( ) dy (, X) T. Stop when no more candidates are found Resulting binary tree * Baatz, M. und Schäpe, A. (2000): Multiresolution Segmentation - an Optimization Approach for High Quality Multi-Scale Image Segmentation. In: Angewandte Geographische Informationsverarbeitung XII. Beiträge zum AGIT-Symposium Salzburg Wichmann, S Slide 12

13 Multiresolution Segmentation (MRS) Example for MRS on a single image Slide 13

14 Segmentation for object- based change detection 1. Segment I 1 using MRS 2. Apply this segmentation to I 2 and recalculate distances 3. Check every merge if it is consistent with the new data 4. Remove inconsistent segments by a specific segment removal strategy (universal, global, local) 5. Re-run MRS to obtain the final segmentation Universal segment removal strategy Slide 14

of I1 Final segmentation of I2 Segmentation of

15 Segmentation for object- based change detection Universal segment removal strategy Segmentation of I1 Final segmentation of I2 Segmentation of I2 with inconsistent segments removed Slide 15

two segmentations, it is possible to create one

16 Integration of segmentations over time Change detection using intersected objects By intersecting two segmentations, it is possible to create one single segmentation with one-to-one-relationships. Slide 16

17 Feature space transformation Transform random vectors U and V in order to emphasize relevant changes Canonical Correlation Analysis (CCA), Iteratively Reweighted Multivariate Alteration Detection (IR-MAD*) Different MAD-variates m i show different types of changes Z= m i 2 ~ Χ 2 gives change probability Transformed data (m i ) Change indicator (Z) Source data (Mean, Std) * Nielsen, A.A., Conradsen, K. un d Simpson, J.J. (1998): Multivariate Alteration Detection (MAD) and MAF processing in multispectral, bitemporal image data: New approaches to change detection studies. In: Remote Sensing of Environment 64, p Slide 17

18 Classification of changes Thresholding Sample Z-value of change and no-change segments Estimate probability densities Determine optimal threshold using Bayes theorem Histogram of NoChange (green) and Change (red) objects Threshold determination Slide 18

19 Classification of changes Thresholding Image I 1 Image I 2 Z-value Classification Slide 19

20 Classification of changes Clustering based on MAD- variates using FMLE* Fuzzy K-Means clustering and Expectation Maximization (EM) Number of clusters as parameter MAD-Variates m i as input Iterative approach 1. Assign an arbitrary cluster C i to all segments 2. Run FMLE-algorithm on all segments that belong to a cluster 3. Identify no-change cluster C NC 4. If found: mark all segments assigned to cluster C NC as unclassified and go to step 2 5. Else: Stop * Gath, I. und Geva, A.B. (1989): Unsupervised optimal fuzzy clustering. In: IEEE Transactions on Pattern Analysis and Machine Intelligence, ITPAMI 11. 7, p Slide 20

21 Classification of changes Image I 1 Image I 2 Z-value Threshold Classification Initial FMLE Classification Final FMLE Classification Slide 21

22 Advances in the Processing of VHR Optical Imagery in Support of Safeguards Verification Introduction Methods Examples Conclusions & Outlook Joint Programme Germany- IAEA on the Technical Development and Further Improvement of IAEA Safeguards COPERNICUS The European Earth Observation Programme Slide 22

G-SEXTANT aims to develop a portfolio of Earth Observation (EO) products and services to support the geo-spatial information needs of EU External Action users and stakeholders, such as the European

23 G-SEXTANT aims to develop a portfolio of Earth Observation (EO) products and services to support the geo-spatial information needs of EU External Action users and stakeholders, such as the European External Action Service. The G-SEXTANT project addresses the following scenarios: Humanitarian Crisis; Natural Resources; Land Conflict Situation Awareness; Monitoring of Nuclear Sites and Activities; Illicit crops; Border Surveillance. Slide 23

EEAS Purpose NUCLEAR ACTIVITIES SCENARIO To provide tools in support of User s activities related to

Monitoring of nuclear decommissioning activities Monitoring of nuclear activities in the context of the

Section DG-AIDCO DG-ECHO EUSC Further users Need for improvement e.g.

24 EEAS Purpose NUCLEAR ACTIVITIES SCENARIO To provide tools in support of User s activities related to monitor nuclear- related sites Key elements Tools related to the monitoring of nuclear-related sites: Monitoring of nuclear decommissioning activities Monitoring of nuclear activities in the context of the Nuclear Non-Proliferation Treaty (NPT) Users IAEA UNODC UN Cart. Section DG-AIDCO DG-ECHO EUSC Further users Need for improvement e.g. contributions to an integrated platform to the non-proliferation imagery analyst Envisaged Tools 2D/3D optical change detection 2D/3D SAR change detection Integration of information sources etc.

25 Detect 2D changes at nuclear sites using optical data: Rapid mapping Preprocessing Segmentation Feature Extraction Transformation Classification Postprocessing Change visualization tool Slide 25

26 Detect 2D changes at nuclear sites using optical data: Rapid mapping Slide 26

27 Advances in the Processing of VHR Optical Imagery in Support of Safeguards Verification 1 Introduction 2 3 Methods Examples 4 Summary & Outlook Slide 27

28 Advances in the Processing of VHR Optical Imagery in Support of Safeguards Verification Object-based change detection procedure, available as plugins for ecognition at Automated image processing tools can assist satellite imagery analysis by highlighting areas where changes occured. Results achieved so far indicate that further improvement is needed with regard to false alarms and classification of changes. Further developments: Time series analysis, multi-sensor processing, European research programmes (e.g. FP7, Horizon2020) offers multiple possibilities for safeguards R&D besides MSSPs. Slide 28

29 Thank you for your attention. This presentation was prepared as an account of work sponsored by the Government of the Federal Republic of Germany within the Joint Programme on the Technical Development and Further Improvement of IAEA Safeguards between the Federal Republic of Germany and the IAEA. Slide 29

CHANGE DETECTION BY THE IR-MAD AND KERNEL MAF METHODS IN LANDSAT TM DATA COVERING A SWEDISH FOREST REGION

CHANGE DETECTION BY THE IR-MAD AND KERNEL MAF METHODS IN LANDSAT TM DATA COVERING A SWEDISH FOREST REGION Allan A. NIELSEN a, Håkan OLSSON b a Technical University of Denmark, National Space Institute