Multi-sensor data fusion for urban area classification

Abstract

Nowadays many sensors for information acquisition are widely employed in remote sensing and different properties of the objects can be revealed. Unfortunately each imaging sensor has its own limits on scene recognition in the sense of thematic, temporal, and other interpretation. Integration (fusion) of different data types is expected to increase the quality of scene interpretation and decision making. In recent time integration of synthetic aperture radar (SAR), optical, topography or geographic information system data is widely performed for many tasks such as automatic classification, mapping or interpretation. In this paper we present an approach for very high resolution multi-sensor data fusion to solve several tasks such as urban area automatic classification and change detection. Datasets with different nature are integrated using the INFOFUSE framework, consisting of feature extraction (information fission), dimensionality reduction, and supervised classification. Fusion of WorldView-2 optical data and laser Digital Surface Model (DSM) data allows for different types of urban objects to be classified into predefined classes of interest with increased accuracy. Numerical evaluation of the method comparing with other established methods illustrates advantage in the accuracy of structure classification into low-, medium-, and high-rise buildings together with other common urban classes.