URBAN AREA PRODUCT SIMULATION FOR THE ENMAP HYPERSPECTRAL SENSOR P.Gamba, A. Villa, A. Plaza,  J. Chanussot, J. A. Benedik...
OUTLINE <ul><li>Introduction </li></ul><ul><li>The EnMap sensor </li></ul><ul><li>Experimental test </li></ul><ul><li>Conc...
INTRODUCTION <ul><li>Urban remote sensing is a crucial task : </li></ul><ul><li>Monitoring cities development </li></ul><u...
INTRODUCTION <ul><li>Can hyperspectral data be helpul for urban monitoring? </li></ul><ul><li>Pro:  very detailed spectral...
AIM OF THE WORK <ul><li>Preliminary investigation of hyperspectral sensors potentialities: </li></ul><ul><li>- Can urban s...
OUTLINE <ul><li>Introduction </li></ul><ul><li>The EnMap sensor </li></ul><ul><li>Experimental test </li></ul><ul><li>Conc...
EnMap mission <ul><li>Dedicated imaging pushbroom hyperspectral sensor mainly based on modified existing or pre-developed ...
SYNTHETIC REALISTIC IMAGES Original image Low resolution image EnMap PSF (900 nm) Spatial resolution degradation <ul><li>C...
Low resolution image SVM SVM-SU 1 <ul><li>Villa  et al. , Spectral Unmixing to obtain classification maps at a finer resol...
<ul><li>SVM – Spectral Unmixing (*) : </li></ul><ul><li>1) Probabilistic SVM determines which pixels can be considered as ...
<ul><li>SVM – Spectral Unmixing: </li></ul><ul><li>1) Probabilistic SVM determines which pixels can be considered as pure ...
<ul><li>SVM – Spectral Unmixing: </li></ul><ul><li>1) Probabilistic SVM determines which pixels can be considered as pure ...
OUTLINE <ul><li>Introduction </li></ul><ul><li>The EnMap sensor </li></ul><ul><li>Experimental test </li></ul><ul><li>Conc...
DATA SET <ul><li>ROSIS Center data set: </li></ul><ul><ul><li>1096 x 712 pixels, 102 spectral bands </li></ul></ul><ul><ul...
EXPERIMENTS Reference data SVM SVM-SU ROSIS Center (3x downscale) OA (%) 98.11 79.56 81.89 Δ  (%) - 19.55 - 16.22 ROSIS Ce...
THEMATIC MAPS SVM on original HR data ( ground truth ) SVM on LR data 70.97% Finer Classification 74.32%
THEMATIC MAPS SVM on original HR data ( ground truth ) SVM on LR data 70.97% Finer Classification 74.32%
EXPERIMENTS <ul><li>Preliminary conclusions </li></ul><ul><li>-  Even slight spatial resolution degradation leads to a sig...
OUTLINE <ul><li>Introduction </li></ul><ul><li>The EnMap sensor </li></ul><ul><li>Experimental test </li></ul><ul><li>Conc...
CONCLUSIONS <ul><li>Conclusions </li></ul><ul><li>-  Low spatial resolution  is a major challenge in urban environment </l...
Upcoming SlideShare
Loading in …5
×

URBAN AREA PRODUCT SIMULATION FOR THE ENMAP HYPERSPECTRAL SENSOR.ppt

784 views

Published on

Published in: Technology
0 Comments
1 Like
Statistics
Notes
  • Be the first to comment

No Downloads
Views
Total views
784
On SlideShare
0
From Embeds
0
Number of Embeds
7
Actions
Shares
0
Downloads
14
Comments
0
Likes
1
Embeds 0
No embeds

No notes for slide
  • 1) All pixels are classified with SVM. If the probability to belong to a class is greater that a chosen treshold, the pixel is considered pure and labeled
  • 2) Spectral unmixing is applied to mixed pixels to determined each class abundance. According to the desired zoom factor, each pixel is split into a number of sub-pixels. Each sub-pixel is assigned to a class according to its abundace
  • 3) Final spatial regularization (by Simulated Annealing)
  • The overall accuracy of the Reference data is the OA obtained by classifying the high spatial resolution data set with an SVM and 100 samples per class.
  • The ellipses show that the method SVM-SU improves the classification accuracy. However, some information is definitely lost, as shown by the arrow: the red class (corresponding to the shadow) can not be found in the final map, close to the yellow class.
  • URBAN AREA PRODUCT SIMULATION FOR THE ENMAP HYPERSPECTRAL SENSOR.ppt

    1. 1. URBAN AREA PRODUCT SIMULATION FOR THE ENMAP HYPERSPECTRAL SENSOR P.Gamba, A. Villa, A. Plaza, J. Chanussot, J. A. Benediktsson
    2. 2. OUTLINE <ul><li>Introduction </li></ul><ul><li>The EnMap sensor </li></ul><ul><li>Experimental test </li></ul><ul><li>Conclusions and perspectives </li></ul>
    3. 3. INTRODUCTION <ul><li>Urban remote sensing is a crucial task : </li></ul><ul><li>Monitoring cities development </li></ul><ul><li>Infrastructure extraction </li></ul><ul><li>Land cover mapping </li></ul><ul><li>Satellites provide cheap and timely data </li></ul><ul><li>Main challenges : </li></ul><ul><li>Urban land cover classes not well distinct </li></ul><ul><li>Heterogeneous areas lead to mixed pixels </li></ul><ul><li>Can hyperspectral data be helpul for such a task? </li></ul>
    4. 4. INTRODUCTION <ul><li>Can hyperspectral data be helpul for urban monitoring? </li></ul><ul><li>Pro: very detailed spectral information of the scene </li></ul><ul><li>Con: tradeoff between spectral and spatial resolution </li></ul><ul><li>( ~tens of meters ) </li></ul><ul><li>A number of hyperspectral missions already planned: </li></ul><ul><li>PRISMA – ASI, 2014 (spatial res: 30 m) </li></ul><ul><li>EnMap – DLR, 2014 (spatial res: 30 m) </li></ul><ul><li>Hyper-J (JAXA) and HyspIRI (NASA) in the next future.. </li></ul>Need to assess the effectiveness of hyperspectral data for urban monitoring!
    5. 5. AIM OF THE WORK <ul><li>Preliminary investigation of hyperspectral sensors potentialities: </li></ul><ul><li>- Can urban structures be monitored with hyperspectral sensors? </li></ul><ul><li>- Is the rich spectral information of hyperspectral data useful for such a task? </li></ul><ul><li>Two main issues addressed: </li></ul><ul><li>1) Creation of realistic urban scenes by considering EnMap PSF </li></ul><ul><li>2) Comparison with classical and advanced methods </li></ul>
    6. 6. OUTLINE <ul><li>Introduction </li></ul><ul><li>The EnMap sensor </li></ul><ul><li>Experimental test </li></ul><ul><li>Conclusions and perspectives </li></ul>
    7. 7. EnMap mission <ul><li>Dedicated imaging pushbroom hyperspectral sensor mainly based on modified existing or pre-developed technology </li></ul><ul><li>Broad spectral range from 420 nm to 1000 nm (VNIR) and from 900 nm to 2450 nm (SWIR) with high radiometric resolution and stability in both spectral ranges </li></ul><ul><li>Swath width 30km at high spatial resolution of 30 m x 30 m and off-nadir (30°) pointing feature for fast target revisit (4 days) </li></ul><ul><li>Sufficient on-board memory to acquire 1.000 km swath length per orbit and a total of 5.000 km per day. </li></ul>S. Kaiser , B. Sang , J. Schubert , S. Hofer and T. Stuffler: &quot;Compact prism spectrometer of pushbroom type for hyperspectral imaging&quot;,  Proc. SPIE Conf. Imaging Spectrometry XIII ,  vol. 7100,  p.710001, 2008. 
    8. 8. SYNTHETIC REALISTIC IMAGES Original image Low resolution image EnMap PSF (900 nm) Spatial resolution degradation <ul><li>Create realistic hyperspectral images by downscaling the spatial </li></ul><ul><li>resolution according to the EnMap PSF </li></ul>
    9. 9. Low resolution image SVM SVM-SU 1 <ul><li>Villa et al. , Spectral Unmixing to obtain classification maps at a finer resolution, Journal of Selected Topics in Signal Processing , 2011 </li></ul>Classification map Classification map at finer resolution CLASSIFICATION COMPARISON 2) Evaluate the performances of traditional methods and sub-pixel techniques in terms of land cover classification
    10. 10. <ul><li>SVM – Spectral Unmixing (*) : </li></ul><ul><li>1) Probabilistic SVM determines which pixels can be considered as pure (if </li></ul><ul><li>prob > treshold) </li></ul><ul><li>2) Spectral unmixing is used to retrieve class abundances within mixed pixels </li></ul><ul><li>3) Final spatial regularization </li></ul>SVM - SU (*) A. Villa, J. Chanussot, J.A. Benediktsson and C. Jutten., Spectral Unmixing to obtain classification maps at a finer resolution, Journal of Selected Topics in Signal Processing, vol. 5, n. 3, May  2011
    11. 11. <ul><li>SVM – Spectral Unmixing: </li></ul><ul><li>1) Probabilistic SVM determines which pixels can be considered as pure (if </li></ul><ul><li>prob > treshold) </li></ul><ul><li>Spectral unmixing is used to retrieve class abundances within mixed pixels </li></ul><ul><li>and to fill “upsampled” sub-pixels </li></ul><ul><li>3) Final spatial regularization </li></ul>SVM - SU
    12. 12. <ul><li>SVM – Spectral Unmixing: </li></ul><ul><li>1) Probabilistic SVM determines which pixels can be considered as pure (if </li></ul><ul><li>prob > threshold) </li></ul><ul><li>2) Spectral unmixing is used to retrieve class abundances within mixed pixels </li></ul><ul><li>and to fill “upsampled” sub-pixels </li></ul><ul><li>3) Final spatial regularization is performed ( Cost function to be minimized: total perimeter of the connected areas ) </li></ul>SVM - SU The results is a thematic map at a finer resolution  useful to assess possibilities offered by HSI at low-medium spatial resolution .
    13. 13. OUTLINE <ul><li>Introduction </li></ul><ul><li>The EnMap sensor </li></ul><ul><li>Experimental test </li></ul><ul><li>Conclusions and perspectives </li></ul>
    14. 14. DATA SET <ul><li>ROSIS Center data set: </li></ul><ul><ul><li>1096 x 712 pixels, 102 spectral bands </li></ul></ul><ul><ul><li>Spatial resolution: 1.3 meters </li></ul></ul><ul><ul><li>Spatial resolution decreased of a factor 3 and 5 </li></ul></ul><ul><li>Classification with SVM and SVM-SU, 100 samples per class. </li></ul>
    15. 15. EXPERIMENTS Reference data SVM SVM-SU ROSIS Center (3x downscale) OA (%) 98.11 79.56 81.89 Δ (%) - 19.55 - 16.22 ROSIS Center (5x downscale) OA (%) 98.11 70.97 74.32 Δ (%) - 27.14 - 23.79
    16. 16. THEMATIC MAPS SVM on original HR data ( ground truth ) SVM on LR data 70.97% Finer Classification 74.32%
    17. 17. THEMATIC MAPS SVM on original HR data ( ground truth ) SVM on LR data 70.97% Finer Classification 74.32%
    18. 18. EXPERIMENTS <ul><li>Preliminary conclusions </li></ul><ul><li>- Even slight spatial resolution degradation leads to a significant classification </li></ul><ul><li>accuracy decrease </li></ul><ul><li>-- Sub-pixel information of HSI is helpful to improve classification accuracy </li></ul><ul><li>-- Potentialities of hyperspectral data depend on the desired application </li></ul>
    19. 19. OUTLINE <ul><li>Introduction </li></ul><ul><li>The EnMap sensor </li></ul><ul><li>Experimental test </li></ul><ul><li>Conclusions and perspectives </li></ul>
    20. 20. CONCLUSIONS <ul><li>Conclusions </li></ul><ul><li>- Low spatial resolution is a major challenge in urban environment </li></ul><ul><li>-> Hard to perform urban monitoring with hyperspectral sensors </li></ul><ul><li>-- Additional spectral information useful if properly exploited </li></ul><ul><li>-> Need for advanced techniques to study sub-pixel information </li></ul><ul><li>Perspectives </li></ul><ul><li>Consider more realistic models (more PSFs, geometric distortions..) </li></ul><ul><li>- Investigate lower spatial resolution (comparable to HSI) </li></ul><ul><li>- Further investigations of the possibilities offered by sub-pixel methods </li></ul>

    ×