Multi‐sensor OBIA methods for conflict research and humanitarian relief applications

Abstract

Two case studies are presented showing the potential of remote sensing focusing on advanced semi-automated Object-Based Image Analysis (OBIA) methods to support the monitoring of conflict affected areas where traditional field assessments are hampered by security concerns or other hindering factors making field access very difficult. First, the exploitation of natural resources is examined by Synthetic Aperture Radar (SAR) polarimetry. Decomposition of polarimetric SAR data is applied to extract polarimetric parameters which have strong relation to the physical scattering mechanisms of the ground target. Then, a hereon based unsupervised land cover classification is conducted. Next, the result of the aforementioned pixel-based classification is improved by OBIA post-processing procedures. The second method described in this article concentrates on the application of state of the art machine learning techniques in order to provide a generic OBIA approach for the extraction of very small scaled features such as dwellings in Internally Displaced Persons (IDP) or refugee camps. Based on current Very High Resolution (VHR) optical data, a comprehensive set of descriptive attributes is derived comprising spectral, geometrical and relational information in order to characterize the features of interest. In the next step, the target features are identified and specified applying the non-parametric classification algorithm Support Vector Machines (SVM).