Singular value decomposition in applied remote sensing

Abstract

Singular value decomposition (SVD) has previously been applied to the problems of forest and agricultural land classification. It has been proven by the author's to be an extremely useful and efficient technique. SVD has previously been used to characterise agricultural species with accuracies approaching 95% and has also been used to differentiate sub-classes of winter wheat. SVD has many uses in the field of remote sensing, due to the inherent attributes associated with this technique, which include: key vector analysis, dimensional reduction, robustness, efficiency and noise reduction. Key vector analysis is a technique which is primarily used for signal characterisation and allows land-cover types to be classified. The increasing use and dissemination of hyperspectral data is causing many data analysis problems, with traditional classification techniques, due to the large number of channels. SVD can be used to significantly reduce the dimensionality of data sets. The computational efficiency is very high when using SVD, as the number of computations increases linearly, with the number of dimensions. SVD is a very robust technique, with excellent noise reduction properties. This work applies SVD to optical airborne data at varying resolution. The use of SVD as a tool in remote sensing is displayed with applications. In particular SVD is applied to characterise agricultural species covering a highly dynamic terrain in Bavaria, Germany. A novel use of SVD for extracting mixed pixel quantities is also shown.