Research project S0/00/004 (Research action S0)
Topic and context
The project lies within the scope of the "New technologies and innovations" research field.
Production of very-high-resolution SAR images
There is growing demand for very-high-resolution images (1m). This demand is being met partially by the new satellite systems operating in the visible field (e.g. IKONOS). A similar advance is expected with the arrival of future SAR systems (e.g. TerraSAR, COSMO-Skymed, EarthWatch X-band SAR), which will have metric-resolution image capabilities courtesy of the acquisition mode known as Spotlight SAR. Despite the continued scarcity of civilian satellite data available for testing purposes, it is high time that Belgium conducted its own expert evaluation of very-high-resolution SAR image processing.
SAR physics
Various electromagnetic diffusion models have been developed in the case of soil associated with plant cover. Certain radar observation models for forests have generally adopted several superimposed layers: soil, low vegetation on the ground, trunks, canopy. The literature also carries details of complete software programs. These are applied to forest cover and based on a triple-layer model, but can often be simplified for plant cover. At UCL, a model is being developed to allow simulations of radar (and SAR) polarimetric observations of rugged soil, plant areas and forest areas. This model makes allowance for the size distribution of trunks, branches and leaves, includes the reciprocity effect, enables vertical distribution inside the canopy and provides an accurate description of vegetation and trees. It represents a fully polarimetric radiative transfer model allowing the diffused Stokes vector to be calculated in full, and thus the full polarimetric response from a large target.
SAR image-processing tools
Change detection is especially important in GMES applications (changes affecting vegetation, cities, monitoring of floods, earthquakes, etc.). For reasons of quality and sturdiness, this should ideally be carried out symbolically and form an extension to standard approaches at pixel level. As a result, change detection requires efficient tools for restoration, interim processing and image registration. Within the SAR context, conventional extraction methods are proving to be of no use and necessitate the development of specific methods, e.g. to detect edges in the case of a mono-channel SAR. In the case of multi-channel SAR, methods based on the multi-varied statistic are more appropriate and warrant greater prior exploration. Throughout this project, attempts will be made to assess the performance of the various algorithms, something not covered within present-day literature yet necessary in order to
1) assess advances in computer vision and
2) integrate techniques with a view to industrial applications.
Objectives
Production of very-high-resolution SAR images
A Spotlight SAR prototype processor will be developed in three stages:
1. Theory behind Spotlight SAR processing, including here:
2. Processor development and encoding
3. Processor validation
SAR physical aspects
The present model and coding need to be tested, improved and validated. An analysis of sensitivity vis-à-vis physical parameters must also be carried out. The ultimate objective is to produce a model and coding that allow the best possible simulation of physical reality.
1. Code testing and improvement
2. Code validation and extensions
3. Sensitivity analysis
SAR image processing tools
Our objective is to explore and extract the information contained in very-high-resolution SAR images.
1. Image restoration: develop and assess noise-reduction techniques based on Markovian models in the field of wavelets and adapted to the SAR image statistic.
2. Interim image processing: develop interim processing tools and investigate their accuracy and sturdiness. Two topics will be examined: detection of multi-varied contours (taking account of the physical properties of the SAR system); and regional analysis (segmentation, texture, classification).
3. Image registration: automation of the registration process in order to increase operational data availability.
4. Change detection: examination of SAR information relating to change detection. The object description used during registration will be applied in order to detect symbolic changes and measure them more rigorously and in greater detail.
5. Performance evaluation of the various tasks on the basis of quantitative indicators (e.g. fraction of correctly classified objects, quality of attributes).
