ARTIFICIAL NEURAL NETWORKS IN SPECTRAL-SPATIAL LANDUSE CLASSIFICATION

ABSTRACT

Landuse is the most important and dynamic terrain attribute that greatly influences many hydrological parameters, like, infiltration, evapotranspiration, runoff, etc. With fine resolutions, multiple bands and more importantly repetitive coverage of large areas, remotely sensed (RS) data is best suited for mapping landuse. The large volumes of RS multi-spectral (MS) data that result, needs more efficient methods for accurate landuse classification. Maximum Likelihood (MLH), a statistical method of classification uses only spectral information. Texture (spatial) features have also been extracted and used in classification. However, nonparametric methods like Artificial Neural Networks (ANNs), with combination of spectral and spatial attributes have come into wider usage for their higher accuracies. Studies have been carried out formulating different cases to assess different combination of properties like, i) spectral only, ii) spatial only and iii) spectral and spatial combined directly, to assess their utility in landuse classification by ANNs.

KEYWORDS:

• Neural networks
• Landuse
• Spectral
• Spatial
• Classification