SAR image and polarimetric SAR images can be used to produce thematic maps, and normally special statistical classification methods are used to derive the necessary information. In this project more robust knowledge-based or model-based techniques for classification will be investigated.
A SAR (Synthetic Aperture Radar) is an airborne or spaceborne radar, and it acquires images of the surface of the Earth with very good spatial resolution (i.e. down to a few meters). An advanced version of the SAR is the socalled polarimetric SAR, where radar waves with both vertical and horizontal polarization is transmitted and received, and hence information about the polarization characteristics of objects is collected. A large number of applications of SAR and polarimetric SAR exist within different topics: monitoring of sea ice in the Greenland Sea, monitoring of agricultural crops, determination of soil moisture on fields, determination of biomass and other vegetation parameters for crops and trees, topographic mapping, and mapping of glaciers and the inland ice in Greenland. The SAR images contain other types of information than conventional image data from, e.g. visible and infrared sensors, because the radar operates at microwave frequencies, where other characteristics of the surface (e.g. water content in plants or the soil, and the structure of the plants) control the appearance of the objects in the images. Furthermore, the radar images are independent of sun light and cloud cover, in contrast to the visible and infrared images.
An important application of remote sensing images, and especially polarimetric SAR images, is to produce thematic maps of various types from the acquired images. This is often performed using classification methods. Standard classification methods cannot be used for SAR images, due to an inherent noise problem called speckle. Also, special classification methods are needed for polarimetric SAR images due to the special data formats and information content. A number of classification methods exist, both for SAR images and polarimetric SAR images, and both using supervised and unsupervised classification. The latter methods are often based on polarimetric decomposition techniques. These classification methods lead, however, seldom to very robust approaches, but they have to be adapted to the individual images acquired from specific geographical locations and under specific environmental characteristics. Therefore, the goal of the present project is to investigate the possibility of using knowledge-based or model-based methods for the entire classification procedure or eventually only parts of the procedure. Such methods, based on common knowledge of how various objects backscatter the radar waves and/or electromagnetic models of the scattering from objects, have the potential of being more robust than the statistical methods.
The project will start with a literature review to seek information about existing classification techniques. Hereafter, appropriate techniques, existing as well as new, will be identified, analysed theoretically, implemented and the performance evaluated. Eventually, new methods will be developed based on a knowledge-based or model-based approach.
30350 Remote Sensing. Experience with programming