levenberg / TSCBPT

Binary Partition Tree for Polarimetric SAR data

TSCBPT: Binary Partition Tree for Polarimetric SAR data.

This repository contains part of the C++ software developed during my PhD thesis:

Alonso-Gonzalez, Alberto. Multidimensional and temporal SAR data representation and processing based on binary partition trees, PhD Thesis, May 2014

The software contains the code that implements the Binary Partition Tree technique for processing Polarimetric SAR data, as described in [1]. Additionally, the command line tool is able to process several acquisitions, in order to perform the time series BPT-based processing described in [2]. Finally, the tool also implements the Distance Based Bilateral filtering technique, as described in [3], in order to filter & regularize covariance matrices and preserve the spatial resolution.

[1] Alonso-Gonzalez, A.; Lopez-Martinez, C.; Salembier, P., Filtering and Segmentation of Polarimetric SAR Data Based on Binary Partition Trees, Geoscience and Remote Sensing, IEEE Transactions on, vol.50, no.2, pp.593, 605, Feb. 2012

[2] Alonso-Gonzalez, A.; Lopez-Martinez, C.; Salembier, P., PolSAR Time Series Processing with Binary Partition Trees, Geoscience and Remote Sensing, IEEE Transactions on, vol.52, no.6, pp.3553, 3567, June 2014

[3] Alonso-González, A.; López-Martínez, C.; Salembier, P.; Deng, X., Bilateral Distance Based Filtering for Polarimetric SAR Data, Remote Sensing, 2013, 5, 5620-5641.

• If you use this code in a scientific work, publication or public presentation, please cite the appropriate references from above.

Compilation / Installation

• To get the complete software you can download the zip file from the project page or you can clone the project with git (requires the program to be installed):

$ git clone https://github.com/aalgo/TSCBPT.git

• The software requires the Armadillo C++ library for linear algebra and matrix computations. It is available in most of the Linux distributions. On Ubuntu/Debian, for instance, it may be installed with the apt package manager:

$ sudo apt-get install libarmadillo-dev

Note the dev suffix. It is required since the C++ headers libs need also to be installed!

• Additionally, a small subset of the C++ Boost libraries is employed. On Ubuntu/Debian it may be installed with:

$ sudo apt-get install libboost-dev

• To compile the code, execute the make command on the project folder (the TSCBPT folder).

• The executable files will be placed within the bin folder. For instance, to execute the TEBPT command line program:

$ bin/TEBPT

• NOTE: The software has only been tested under Linux with the GCC compiler.

TSCBPT C++ library

The repository contains a set of C++ header libraries to perform the Binary Partition Tree (BPT) processing mentioned in the previous references. The libraries are located within the include folder. They are "header only" libraries, based on generic template classes, intended to be employed for many other types of data, models, etc.

As an utility and as a usage example, a file called TEBPT.cpp is included within the src folder. It is employed to build the TEBPT command line tool to process time series PolSAR data according to [2].

Usage of the TEBPT tool

The TEBPT tool is a command line tool for processing time series datasets with the Temporal Evolution BPT, as described in [2]. There are different compilations of the tool for processing different types of data:

• TEBPT processes fully polarimetric SAR data into the C3 covariance matrix.
• TEBPT-T3 processes fully polarimetric SAR data in the Pauli basis into the T3 covariance matrix.
• TEBPT-Dual processes dual polarimetric SAR data into the C2 covariance matrix.
• TEBPT-Single processes single polarimetric SAR data into the C1 covariance matrix (intensity).

NOTE: Although all of the previous tools will generate the results in a PolSARPro friendly format, the PolarType entry in the config.txt file may not be correct for Dual and Single polarimetric case, since the tool does not know this information (the same case as in fully polarimetric data is written). This needs manual edit of the config.txt to set the appropriate PolarType value.

The tool (any of the above variants) may be executed with the syntax:

$ bin/TEBPT [options] pruning_factors rows cols file1 [file2 ... fileN]

• With respect to the input files & format, it is important to notice:

  • The input files are in binary format (complex float) with no header. Alternatively, files with a two 32bit integer header containing the rows and cols size information are allowed. In this case, specify 0 0 as rows and cols in the command line and the tool will read this data from the files.
  • NOTE: The binary endian of the input files can be switched with the --swap-endian option.
  • The files are assumed to be a sequence of (HH, HV, VH, VV) binary files. Then a number of files multiple of 4 is expected.
  • The files are converted to (C3, T3, C2 or C1) covariance matrices and the results are stored in Prune_N folders (where N stands for the prune factor, in dB) in PolSARPro format. In fact they can be opened with this software for visualization and processing.
  • For time series datasets, all the data will be saved in the same folder with the consecutive files for the additional matrix elements, and only the results for the first acquisition data will be accessed by the PolSARPro software.

• The rows and cols parameters represent the size of the input files.

• The pruning factor is expressed in dB. It may be a number (e.g. -2) or a sequence of numbers, specified as start:inc:end (e.g. -3:0.5:0). If a sequence is given, all the different prunes will be generated after BPT construction, which is much faster and efficient than with different executions of the tool.

• The different options include:

  • --out outpath Changes the output directory to given one. Otherwise the results are written in the current folder. Note: remember to use a folder that already exists!
  • --cut start_row start_col height width Process only the specified crop of the input data.
  • --bpt file Read the BPT merging sequence from the given file. Once the BPT has been constructed for a dataset, the merging sequence file BPT.msq is generated. With this, the later BPT reconstruction may be performed much faster, as there is no need for the computation of the adjacency graph and the similarity measures.
  • --nl rows cols Apply a multilook as initial filtering of the given size rows by cols.
  • --bl rows cols Apply a distance based bilateral [3] as initial filtering of the given size rows by cols.
  • --blf-sigma_s, --blf-sigma_p, --blf-sigma_t, --blf-iterations value Change the parameters of the distance based bilateral, as described in [3].
  • --no-ts Do not compute temporal stability measures and images.
  • --dist-all Generate distance images between all pairs of acquisitions.
  • --no-write Do not write pruned images data.

In the future, more examples of using the generic TSCBPT template library will be added.