Change detection by object-based change indications

Within a current research project remote sensing imagery in the range of 5m spatial resolution like SPOT and IKONOS are used to derive a Land Use – Land Cover (LULC) database, which is regularily updated. The revision process includes a concept for an object-based change detection attempt to efficiently update existing geo-spatial data which is described in this paper. The change information needed is derived from recent satellite images using automatic image processing and analysis. The conceptual idea includes both, manual (applying visual interpretation) and automatic image analysis steps resulting in a change layer, which highlights those areas or objects which are suspect for change and gives an indication of the direction of change. Different preprocessing steps have been implemented in order to avoid seasonal effects or changes due to different imaging conditions, such as atmospheric conditions, different sun angles, etc.. However not always ideal imaging conditions can be found which result in change indications, like shadows which becomes more dominant with increasing image resolution. Further pre-processing includes an automatic haze reduction and a shade correction using an appropriate DTM. Image coregistration and automatic cloud and shade-of-clouds detection is performed in the standard processing and thus will not be discussed here. However the attept presented uses additionally a priori knowledge of potential change for the specific object classes as input to control the subsequent image processing. The concept of the change detection starts by setting up a focusing step to selectively initiate the following steps only for those objects which are considered as changed. Thereafter all changed objects are classified either visually (manually) or by an automatic procedure depending on the type of change detected. The decision which classification procedure is used depends on a transition-probability-matrix which indicates for each class the degree of likelihood of possible and impossible class-transitions respectively in combination with a table of available classification operators which can be applied to validate the predicted change. The transition-probabilitymatrix is generated manually and contains assumed possible changes from one class to another. If an automatic classification is indicated, the procedure then consists of two parts: First it is evaluated if the object’s geometry is changed or if the object is changed as a whole. If a change in geometry is detected, the object of concern has to be re-segmented and re-classified. If not, the object has to be re-classified only. If a manual classification is indicated, changes will be mapped respectively. At the end the results of the visual/manual classification and the automatic classification are joined into one change layer, which holds for each changed object besides its change indication information, the objects’ historical classification and its new classification. This layer can then be directly used as input for updating existing GIS databases. This paper concentrates on the first part of the process chain, namely the focusing module. The focusing module has two tasks: First, objects have to be found in the GIS data which are affected by change. Second, the focusing module has to decide, whether the changed objects subsequently can be processed automatically or must be processed manually. Different pixel based change indicators are implemented based on a comparison of the input satellite data of two different dates. The decision if a change is apparent in many cases is dependant on the threshold or threshold function used. Approaches avoiding crisp thresholding and using fuzzy membership indicators at the desired object level will be discussed in this paper. The obtained results of the proposed object-based change detection process chain are compared to change detection results obtained by completely visual interpretation. Finally all results are assembled to a resultant change indication map. 1 DECOVER BACKGROUND AND CONTEXT In the context of GMES (Global Monitoring for Environment and Security), a joint initiative of European Commission and European Space Agency, several services are developed to provide spatial information in support of the monitoring and reporting obligations of European directives (Overview at www.gmes.info, Example Water Framework Directive Dworak et al 2005). These implementations take place with strong participation of German authorities, researchers and service providers. Current developments at the European level support a new European-wide land cover data set (Core Service Land Monitoring). This data set must be seen as a European consensus and will solely contain thematic land cover data information supporting European reporting obligations. Its geometric and thematic resolution will only partly support national and regional needs. DeCOVER (Büscher, et al., 2007) will complement and extend these developments at the national and regional level for German users. A set of geo-information services has been designed to support national and regional users in their monitoring and reporting obligations. The DeCOVER service concept is divided into core and additional services. The DeCOVER core service has two main focal points. First, the provision of national harmonized land cover data supports the German spatial data infrastructure (GDI-DE) in providing selected and validated geo-information and second, the development and application of change detection and interoperability methods to sustain existing data bases (namely ATKIS, CLC and BNTK). The project is co-funded by the Federal Ministry of Economics and Technology (BMWI) via the German Aerospace Center (DLR) and implemented by a consortium of 11 partners (see Table 1) each using its own expertise and specialized skills.