Satellite image enhancement using Transforms and deep learning algorithm To delineate waterbody and vegetation Region for precise area measurement

Abstract

newline Remote sensing methodologies delineates the vegetation and water bodies of a specific geographical location. Delineation of vegetation and water bodies plays a vital role for urban, rural area planning and development. Delineation of vegetation and water bodies from different terrains is a challenging task because of geometric distortion and mixed pixels over the boundary and curvature area. Geometric distortion occurs as a result of isotropic frame, angular selectivity, change in satellite velocity and speed during image acquisition. Mixed pixels occur due to different classes such as vegetation and water bodies single pixel. In this thesis, to remove the geometric distortion and mixed pixel for precise delineation and accurate measurement algorithms are proposed such as TDyWT, SWT, DnCNN for medium and high-resolution satellite images for precise delineation and accurate measurement of vegetation and water bodies with respect to ground truth verification. newlineThe existing methods uses classifiers such as Support Vector Machine (SVM), Neural Network (NN), and fuzzy algorithms for the vegetation and water area delineation in multi resolution satellite images. The existing algorithms require large dataset for training and consumes more time for interpretation. The existing algorithms such as SVM, KNN, Fuzzy genetic algorithms misclassify pixels on boundary and curvature regions such as urban, vegetation, water bodies, and hilly land surfaces due to geometric distortion and mixed pixels. Furthermore, spectral analysis and pixel classification of boundary region plays an important role in accurate area measurement. The exact and enhanced boundary and curvature region enables the accurate area measurement