ANN based robust LULC classification technique using spectral, texture and elevation data

Albeit the advent of fast computing facilities, digital image classification of remotely sensed data is still remain the topic of research. This might be due to the reason that the ancillary information such as texture and topography is absent in image classification. Since two decades, texture is widely applied in
image classification but there is no explicit icon in most popularly used remote sensing software. Hence the aim of this study is to classify the Landsat ETM+
captured in 2000 using spectral information, topographic information and texture information. This study helps to throw light into statistical texture analysis i.e., the effect window size i.e., 3×3 to 9×9, on image classification. The ability of Grey Run Length Matrix (GRLM), which is computationally complex compared to industrially well-known Grey Level Cooccurrence Matrix (GLCM) but encompasses greater potential to discriminate between two classes, is explored. Eight spectral bands, 11 texture parameters extracted from Landsat ETM+ data and elevation,
slope, aspect extracted from DEM data are classified individually using Artificial Neural Network (ANN) and the individually classified information is integrated
using endorsement theory. Validations of classified results are performed using Google Maps and Landmap services updated in 2009. The results are compared with Maximum Likelihood classification (MLC) and hence all the evidence (spectral, texture and topography) with 5×5 texture window provided maximum classification accuracy of 70.44 %.