Hybrid classifier for water quality classification from satellite image trained with modified multi texton features

Abstract

A new WQC model is introduced in this work, where, the quality of water from Baitarani and Brahmani River were considered using Satellite Image. Initially, both the input data and images were preprocessed using Improved 2-step min-max normalization and proposed adaptive median filtering approach, respectively. After that, raw features as well as entropy features were derived from preprocessed data, while, VI based features, deep features and modified MT features were derived from preprocessed image. Further, the feature fusion was done to concatenate the data features and image features. The fused feature set was subjected to the hybrid water quality classification step, which involved ECA-MSNet and LeNet. The average of these classifiers determined if the quality of water was satisfactory or not. From analysis, ECA-MSNet + LeNet achieved a high sensitivity of 0.96385% at TD = 90%, while Bi-GRU, LeNet, Dense net, Link net, Shuffle net, SVM + RF+MLR and CNN got lower sensitivity values.