Abstract:
Shoreline mapping and shoreline change detection are critical for safe navigation, coastal resource
management, coastal environmental protection and sustainable coastal development and planning.
The main difficulty of traditional shoreline mapping from remote sensing classification is the lack of
adequate tools to characterize and combine texture and spectral information effectively. This paper
introduces a method for unsupervised lake shoreline delineation through combination of scene texture
and spectral characteristics. The framework is based on multiresolution image segmentation via
multispectral anisotropic diffusion neural network, in combination with texture derived from 2D-
wavelet transform algorithm. We illustrate the application of this algorithm by extracting water body
pixels from Landsat ETM+, TM and MSS for case study of Lake Nakuru in Kenya. The results are
very superior compared to the conventional methods (NDWI) and indicate that the lake has reduced
by 18.8% between 1976–2001.
