Land cover feature recognition by fusion of polsar, polinsar and optical data

M. Shimoni, D. Borghys, R. Heremans, N. Milisavljević, D. Derauw, C. Pernel, A. Orban

Research output: Contribution to journalConference articlepeer-review

Abstract

Classification of land cover is one of the primary objectives in the analysis of remotely sensed data. To aid in this process, data from multiple sensors are often utilised, since each potentially provides different information about the characteristics of the land cover. The main research goal of this study is to fuse different frequency E-SAR PolSAR data as well as PolInSAR with Daedalus optical data for land cover classification and land cover feature recognition. The studied area is located in Croatia and is covered by several land features including different types of crops, residential areas, roads, pastures, bare soil, forest and river. Two different level fusion techniques are applied and compared in this study: the Logistic Regression (LR) as 'feature based fusion' method and the Fuzzy method for higher decision level fusion. The LR technique was used to fuse in total 102 features extracted from the complete data set. Iterations of the different land cover training areas using different features sets produced different probability images of the detected land cover types. The outputs of the feature-based fusion are 19 supervised classifications that were used as database for the higher-level fusion. For both fusion methods the overall accuracy for each of the fused sets is better than the accuracy for the separate sets of features. Based on the results presented in this research we found that: (1) Fused features from different SAR frequencies are complementary and adequate for land cover classification; (2) PolInSAR features are complementary to the PolSAR information and essential for producing accurate classification of different land cover types as man-made object, water bodies, forest, crops and bare soils; (3) The optical data is complementary information for the SAR data but not necessary for the production of accurate land-cover classification.

Original languageEnglish
JournalEuropean Space Agency, (Special Publication) ESA SP
Issue numberSP-644
Publication statusPublished - Mar 2007
EventPolInSAR 2007: 3rd International Workshop on Science and Applications of SAR Polarimetry and Polarimetric Interferometry - Frascati, Italy
Duration: 22 Jan 200726 Jan 2007

Fingerprint

Dive into the research topics of 'Land cover feature recognition by fusion of polsar, polinsar and optical data'. Together they form a unique fingerprint.

Cite this