CONTAM 03 Remote and Embedded Sensing in Environmental Systems

Approach We developed an object-based (OB) change detection approach enhanced by its integration with Multivariate Alteration Detection (MAD) transformation and decision tree-based post-classification methods (Figure 1). Recent studies have found that MAD methods are promising in change detection analysis. However, MAD has not been wellstudied in terms of reducing errors due to mis-registration and shadows. By incorporating object-based analysis with MAD, features such as object shapes and contrast could be used to improve change detection results. For example, misregistration along rivers and roads often results in linear features that can be detected using object-based analysis. Our hypothesis is that the resulting OB-MAD method will better differentiate real from spurious changes in very high resolution images compared to conventional approaches. In the proposed approach, MAD transformation is first implemented to find the change and no-change pixels using spectral and spatial information. Then, MAD components are combined and used for image segmentation and object extraction. Finally, we apply a decision tree classification method to detect spurious change areas caused by misregistration and shadow.