Analysis of Wavelet Type and Noise in Relative Radiometric Normalization Based on Multiresolution Analysis

Abstract

Change detection with satellite imagery has numerous usage areas and the number of Earth Observing Satellites which can be utilized in the analysis are expanding. Temporal differencing based change detection methods require both geometric and radiometric corrections to be applied to satellite imagery taken at different times. Radiometric corrections can be categories into two classes; absolute and relative corrections. In the absolute case, various state variables such as atmospheric conditions, time, satellite position and orientation shall be available. On the other hand, in the case of relative radiometric correction, two or more images of the same area can be utilized to estimate the correction parameters. The main focus of this study is the assessment of Wavelet Based Multivariate Alteration Detection (MAD) method for relative radiometric calibration and change detection by using bi-temporal images. Effects of different wavelet types are investigated by developing a software library in Python programming language. Various Sentinel-2 images of different areas of the world are used as test images and starting from classical Iteratively Re-weighted MAD (IR-MAD), several types of wavelet based IR-MAD are compared and results are reported. In addition the developed software library is wrapped in a QGIS Plugin to be used by the remote sensing community.