River Boundary Delineation from Remotely Sensed Imagery Based on Svm and Relaxation Labeling of Water Index and Dsm

Topographic mapping of river boundary area is important in the environment management and hydrological modeling. In recent years, delineation of river by means of remotely sensed image classification has been widely investigated. Prime advantages of remote sensing technology for river delineation are reducing labor cost and time and elevating restrictions on ground surveys. However, these techniques are as yet to produce satisfactory result due to, for instance, the lack of an effective means of removing shadow and spurious noise. This paper proposes the boundary delineation of river from satellite image using Support Vector Machine (SVM) and two well accepted water indices (NDWI2 and MNDWI). In our framework, relaxation labeling was also incorporated in order to ensure reliable classification while suppressing spurious artifacts, inherited from the satellite imaging. The subsequent assessment suggested that MNDWI index yielded the most accurate classification and hence water extraction. The resolution of the delineated river braches was further enhanced by incorporating DSM data. The technique was then benchmarked against labor intensive manual tracing with promising consistency.