Multitemporal Multispectral Image Fusion and Sub Pixel Classification Framework for Detecting Surface Water Changes in Landsat Images

Abstract

The gradual depletion of surface water in major lakes and their impact in the sustainable development of local water resources has been a great challenge. Surface water changes are usually identified by extracting the water features from individual time series multispectral images. A novel framework comprising of pixel fusion and classification is designed to detect the changes in multitemporal multispectral images. An effective spatial frequency and maximum amplitude selection based fusion rule on undecimated wavelet transform coefficients is proposed for extracting and injecting the spatial, spectral content from multispectral (MS) and panchromatic (PAN) images of the same scene captured at different time periods. Landsat Thematic Mapper (TM), Landsat Enhanced Thematic Mapper (ETM+) and Landsat Operational Land Imager (OLI) of 2010, 2000 and 2018 images acquired from US Geological Survey were fused to investigate the potential of the proposed spatial frequency based undecimated wavelet transform (UDWT-SF) in highlighting the changes. Several component substitution and multiresolution analysis based fusion algorithms were applied and found that undecimated wavelet transform with spatial frequency effectively extracts the spectral information from MS image and spatial information from PAN image. Visual interpretation, spatial and spectral quality assessment metrics were evaluated to analyse the performance of the proposed fusion strategy. The endmembers of the fused image were selected using pixel purity index endmember extraction algorithm and the abundance estimation by the application of fully constrained least square spectral unmixing algorithm. Devised an efficient sub-pixel classification process by employing the spectral signature, Normalized Difference Water Index (NDWI) and the abundance map generated from the pansharpened image in random forest classifier. Integration of spatial frequency based undecimated wavelet transform fusion and spectral unmixing based random forest classifier yields better resul

Description

Keywords

Citation

item.page.endorsement

item.page.review

item.page.supplemented

item.page.referenced