The role of multispectral image transformations in change detection

Abstract

In recent decades, remote sensing techniques have been applied as a powerful tool to provide temporal variation of Earth related phenomena. To understand the impact of climate change and human activities on Earth water resources, monitoring the variation of water storage over a long period is a primary issue. On the other hand, this variation is fundamental to estimate the hydroelectric power generation variation and fresh water recreation. Among the spaceborne sensors, optical and SAR satellite imagery provide the opportunity to monitor the spatial change in coastline, which can serve as a way to determine the water extent repeatedly in an appropriate time interval.

While water absorbs nearly all the sunlight in near-infrared wavelength, the water bodies appear very dark at this band in an optical image. So applying a threshold on the image histogram is a common way to build the water mask. Despite its straightforward procedure, precise distinctions among water bodies may not be possible in some regions or seasons because of the complicated relationship between water and land and also because of the effect of vegetation.

As well as thresholding, other change detection method are widely used to monitor the extent of water bodies. Multispectral transformation analyses like PCA and CCA are able to highlight the important information about the change in all spectral bands and also to reduce the dimension of data. In this way, their potential to improve the quality of satellite images and also reduce the noise level must be assessed.

In this thesis, we have two general objectives. First improving the quality of the multispectral image applying PCA on spectral bands and then reconstructing the image using just a certain number of PCs. Number of the PCs and the selecting strategy appropriate PCs are the main challenge of this procedure. Highlighting the change between two multispectral images applying transformation like PCA and CCA is the other objective. Interpreting the transformed images are not straightforward and in most cases, comparing with the original images could be a solution. We examine different scenarios to examine the performance of the spectral transformations in change detection.