Identifying Rivers with Varying Width Through NDWI from Landsat 8 Images

Though different water indexes can detect rivers with significant width, detection of a narrow river trail is a complex process as various other land classes have near similar water index values. In the past, various techniques on detection of river networks consider rivers with significant width and as a continuous trail primarily using digital terrain (DTM) and digital elevation (DEM) models. Further, road networks are also found to have near similar spectral characteristics with narrow rivers. Moreover, a narrow river shows high sinuosity than other river networks. The paper presents a novel technique to detect rivers with varying width without labelled dataset. Firstly, water index values are enhanced by gabor filters and adaptive histogram equalization. Next, rivers with significant width and narrow rivers are detected separately using connected component and bounding box analysis.