Case-Based Reasoning (CBR) for Land Use Classification Using Radar Images

This paper presents a new method for classifying satellite SAR images based on case based reasoning (CBR) techniques. Because classification is a common task in remote sensing applications, numerous methods have been deve loped for obtaining better classification results. Knowledge based systems (KBS) are considered as a good alternative to traditional classification methods with better performance. There is a need to develop such systems to facilitate the interpretation of remote sensing data in a more efficient way. KBS are useful when concrete knowledge about the application domain is available. It is expected that KBS can automatically classify remote sensing images without operator's intervention. However, these systems have a bottleneck problem in the solicitation of rules. A solution is to apply CBR method to the classification process. Traditional classification often assumes that spectral properties of a class remain stable in the whole study area. However, the spectral signature of a class is usually subject to fluctuations because of the complexity in nature. The CBR method can easily capture such fluctuations by allocating cases over different terrain features according to stratified random sampling. Moreover, the same case library developed in the previous classification can be reused for time independent classification with satisfactory results. Experiments show that the proposed method can generate the classification results with better performance in term of higher accuracy and fast computation time. The method has been successfully applied to the classification of radar SAR images in the Pearl River Delta, south China.