OGRSS Data Fusion 2014

informations here

Submission

Proposed framework

We develop a complete framework to exploit jointly RGB and LWIR data. It consists in extracting relevant features from images, combine them and perform supervised classifications based on this set of features. Then, classification maps are combined using classifier fusion methodology.

The main steps of the proposed framework are:

Feature extraction

We exploit several features, from both data using:

- Dimensionality reduction using Principal Component Analysis (PCA) - Band selection - Statistics - OBIA ?

Supervised classification

- Pixel Based - Object Based

Implementation

This methodology is based on algorithms all available in the ORFEO ToolBox library [3]. ORFEO Toolbox (OTB) is distributed as an open source library and offers particular functionalities for remote sensing image processing in general and for high spatial resolution images in particular. OTB is funded by the French Space Agency (CNES) and distributed under a free software license CeCILL (similar to GPL) to encourage contribution from users and to promote reproducible research.

Reproducibility

The product of this research will be a paper describing the methodology with the full computational environment used to produce the results.

References

[3] http://www.orfeo-toolbox.org/

Notes

Data Description

Training samples description

0 unclassified

1. road
2. trees
3. red roof
4. grey roof
5. concrete roof
6. vegetation
7. bare soil

Tests

Training Samples description

TODO

Processing chain

different strategies (top down, bottom up):

• OBIA based : segment data and the classify each object(each segmented area) using high level features computed on each object
• pixel based : segment data at pixel level and then use post processing on classified map (segmentation, majority voting ...)

• Dimensionnality Reduction

PCA -> Matlab OTB (more than 99.99% info inone band) ICA , NAPCA ? TODO

• Reprojection

Using MVD1 -> Good results

• Extraction
  • spectral distance

• Segmentation
  • Meanshift
  • TODO add spectral data
  • watershed ...

• Classification

• • SVM ?

• PostProcessing

• • majority voting (on pixel based classification, useless if a segmentation step is present)
  • Dempster Shaffer

First Processing chain

• SVM Classification on spectral data

TODO

Results

TODO

• First Processing chain -> OK

Add Background value to compute image statistics

Classification MAP regularization

Add Post Processing using MeanShift (OBIA code Mickael)

Test classification only on vegetation/tree class

• NOTES

Conf Mat SVM sur RGB

1. 909 4 4 36 12 1 8
2. 1 851 0 0 0 128 0
3. 3 0 959 1 0 2 52
4. 73 0 28 853 51 3 1
5. 6 0 0 13 951 1 1
6. 1 111 0 1 0 861 22
7. 1 1 14 0 4 20 909

Conf Mat SVM sur RGB+PCA_LWIR

1. 908 1 15 79 4 2 1
2. 0 843 0 0 0 153 0
3. 4 0 1037 3 1 0 6
4. 27 0 11 937 38 0 0
5. 2 0 1 11 903 1 0
6. 0 80 1 1 0 887 19
7. 0 1 1 0 2 18 993

Meilleurs résultats de la carte concaténée a part pour la séparation Vegetation/Arbre

TODO

Work Schedule

• Step 1 of the Classification Contest: NOW!
• Step 2 of the Classification Contest and opening of the Paper Contest: release of the full data set between February 16 and February 23, 2014. Announcement will be provided through this website and through the mailing list and the Linkedin group (see below) of the IADF Technical Committtee.
• Submission deadline for the Classification Contest: two weeks after Step 2