Artificial Neuarl Network Based Coral Cover Classifiers using Indian Remote Sensing (IRS LISS-III) Sensor Data: A Case Study in Gulf of Kachcch
要旨Artificial neural network have become popular in classification of reomtly sensed satellite data where they demonstrate better accuracy than conventional methods. A back propagation neural network algorithm has been developed to classify eco-morphological zonation of coral reef as well as benthic communities. IRS P6 LISS III satellite data of March 2, 2006 has been used to map coral reefs using hybrid analysis of user based knowledge. The traditional method used for classification of coral reefs gives substantial amount of mis-classifications due to the similarity in reflectance values. The optimized neural network, made for the classification of coral reef image with high rate of noise, shows a better accuracy, as it was able to remove mis-classifications up to certain extent. The developed classifier uses the radiance values of 300 homogeneous pure pixels per class as training samples from a radiometrically and geometrically corrected image of the study area. The optimized network was applied on the complete coral reef image. Accuracy was checked by cross validating ground truth data which showed considerable improvement (84% at 90% confidence level to 91.14% at 90% confidence level) in mis-classifications.