Region-Based Object Extraction using Adaptive Neuro-Fuzzy Inference System Combined with Support Vector Machine

This article offers a hybrid technique for river extraction. At starting, Neuro-fuzzy utilized based on RGB values of a remote sensing image; next, a fuzzy rule is applied to segment the RGB input image that is pre-processed into two complementary fuzzy River Regions of Interest (ROI); after that, two dimensional discrete wavelet decomposition is applied on two River ROI for supplying a multi-resolution decomposition of an image; Fourth SVM is created for a learning-based classification of the parameter vectors extracted at ROIs; Finally Decision Tree is applied to examine the feasibility of neighbouring image areas for combining to make more river like regions depicted by nodes. Observational outcomes show that the proposed technique was capable to merge neighbouring small segments, which have high spectral heterogeneousness property and low river-like geometrical attributes to build more meaningful river regions, and performed excellent to the available existing methods. We conclude that our hybrid technique can give an efficient ways for extracting river over pond waters from satellite images. The experimental results show that the proposed method is effective for detection of meaningful river regions.