Geometric distortion and mixed pixel elimination via TDYWT image enhancement for precise spatial measurement to avoid land survey error modeling

Abstract

In remote sensing, land cover classification of vegetation and water area from satellite image play a vital role for rural and urban planning and development. Existing algorithms of land cover classification require more sample image datasets for training. For existing algorithms, land cover classification of vegetation and water area is a challenging task because of mixed pixel and geometric distortion over boundary and curvature region. Mixed pixel affects the precise classification and measurement of land cover. Geometric distortion arises due to frame of isotropic and angular selectivity during image acquisition and affects the contour of land cover. In this paper, the proposed transverse dyadic wavelet transform (TDyWT) enhances and classifies vegetation and water area in land cover from LANDSAT image without training datasets. The proposed TDyWT uses Haar wavelet for decomposition and Burt 5 × 7 wavelet for reconstruction. The TDyWT enhances the contour, curvature, and boundary of vegetation and water area in LANDSAT image due to reversible and lifting properties of wavelet. TDyWT removes geometric distortion and spatial scale error of mixed pixel. In traditional land surveying spatial scale error reduction eliminates through total station and error modeling techniques. From the results, the proposed TDyWT algorithm classifies the area of subclass of vegetation and water with the 95% of accuracy with respect to ground truth survey methods.