This project intend to develop advanced signal processing and information retrieval approaches for temporal analysis of SAR and POLSAR data, which can result in robust change detection and trend analysis algorithms. Within this research area, we will focus on two important aspects; namely multi-class surface cover classication, and the detection and measurement of temporal change. The application areas will be related to surveillance of dynamic trends of northern land and ice (also snow covered) sur faces. The research is divided into three work packages; WP1: Data acquisition and pre-processing. WP2: Statistical modeling. WP3: Feature extraction and machine learning. In WP1 all relevant ground truth and existing SAR data will be collected, prepro cessed and organized into time series. Plans for further satellite and ground truth data acquisition will be made and carried out. In WP2 physical and statistical modeling of SAR/POLSAR data will be further developed. The idea is to extract texture feat ures, which together with features derrived from target decompsition methods, will enhance the classi?cation capabilities, and enable improved change detection and trend analysis. In WP3 analysis strategies and algorithms rooted in information theoretic and statistical learning theory will be developed and adapted to multidimensional SAR data. The team comprises national and international experts on processing and analysis of multi-temporal, multidimensional SAR remote sensing data, experts in machin e learning and pattern recognotion. in addition to experts from the application areas. The latter scientists have signifcant experience from field work. The overall project adminstration will be conducted by the University of Tromsø.
Project leader: Torbjørn Eltoft
Institution: Institutt for fysikk og teknologi