Skip to content
jups23 edited this page Sep 4, 2012 · 15 revisions

General Rules

We write our code and documentation in english. We format text in UTF-8.

1. Meeting 23.07.2012

Welthungerhilfe (WHH) is interested in two different projects. We focus on the latter.

  1. Landbank, a central database which lists landowners
  2. A satellite image analyser which finds out differences in land usage

Goal

This is an exploratory prototype which will be used by WHH employees in Sierra Leone for gathering data for campaigns and informing the government. The software should automate the tedious work of comparing satellite of images from different years and creating charts out of it.

Next Steps

Until 2012-07-27 the users in Sierra Leone will be collecting user stories to fill our product backlog. In addition we will be provided with a license of ArcGIS to evaluate whether we can leverage the already done work from this commercial product.

We as the developer team decided to start hacking this very week to be able to show something in july.

Technical Decisions

We decided to separate the analysis software from a GUI. We also do a technology spike with Python for PyQT and PyOpenCV. We suspect that we can develop faster with Python instead of C++.

Second Meeting: 03.09.2012

We are done. We found IDRISI Selva which does the job. It is surely more complicated than a focused tool like we have been building but it is scriptable.

Next steps

The WHH is going to assign the actual analysis tasks to grad students. We offered our help by scripting Selva to ease common, repetitive tasks.


Knowledegebase

Band Composition

Landsat images can belong to different generation of satellite and sensors. There are three kinds of images available. ETM+ (Enhanced Thematic Mapper), TM (Thematic Mapper) and MSS (Multispectral Scanner). Landsat 7 (since 1999) with ETM+ images is the latest and greatest. A overview of band names and light spectrum for the different Landsat versions can be found here: http://landsat.usgs.gov/band_designations_landsat_satellites.php

Which Band includes which Information

The following tuples are in RGB order.

Description of each band: http://landsat.usgs.gov/best_spectral_bands_to_use.php

Check out http://landsat.usgs.gov/tools_viewer.php (Java Applet) for typical spectral characteristics to find out how to detect a certain material.

Precision

For ETM+ images (http://eros.usgs.gov/#Find_Data/Products_and_Data_Available/ETM):

  • Band 1-7 have a resolution of 30 meter per pixel
  • Band 8 has 15 meter per pixel

Related Publication

A great description on how to process remote sensing images can be found here: http://fig.net/pub/cairo/papers/ts_27/ts27_01_rwetabula_desmedt.pdf

In this paper the authors used Idrisi32 software to map the land cover and land use in Tanzania. The steps they did were:

  1. Statistics magic: principal components analysis (PCA) to "identify band images which carry the most genuine information about the full band set for digital analysis" (Page 5)

In PCA an orthogonal transformation of the coordinate axes of the multivariate system to new orientations is performed in such a way that small number of components (smaller than the original dimension) explains as much as possible the variance in the original data. The eigenvalues express the amount of variance explained by each component and the eigenvectors are the transformation equations.

RWETABULA, DE SMEDT, 2005. This also may reduce the amount of data to process

  1. Gather training data via ground truthing. This means spectral analysis of the materials which they wanted to detect
  2. Joining/Merging the most interesting bands (mosaicking) to one image
  3. Classifying the image's pixels according to the training data

In the end they found six categories: mixed short grasses with/or bare land, dense tall grassland, bush land, cultivated land, medium size grassland, and surface water.

##Sierra Leone

  • Dry Season: December-May
  • Rainy Season: May-November

In the study about land cover change (handed to us by WHH via Dropbox) the author used images from Jan-Feb because there are least clouds and the grass is dry so it is easier to discriminate.