|Ph.D Student||Almog Ophir|
|Subject||Wavelet Decomposition for Reducing Flux Density Effects|
on Hyperspectral Classification
|Department||Department of Civil and Environmental Engineering||Supervisors||Professor Maxim Shoshany|
|Dr. Victor Alchanatis|
|Full Thesis text - in Hebrew|
Of those factors affecting accurate estimation of surface reflectance, the flux density effect, as represented by the cosine of the sun's incidence angle, is well established. Accounting for this effect, however, requires very detailed information regarding the surface facets' orientation across a wide range of scales (e.g., the plants' leaves, the soil aggregates' structure). Moreover, the spatial resolution of Digital Elevation Models is frequently lower than that of satellite and airborne imagery. Thus, reflectance estimations for these images will not fully account for flux density effects and consequently, the reflectance of the same surface material would vary, resulting in increased spectral confusion. Utilization of normalization, band selection, ratioing, spectral angle (SAM), and derivative techniques for this purpose provide only partial solutions under unknown illumination conditions.
In this thesis we introduce a novel signal processing approach, based on wavelet analysis, aimed at reducing the effects of flux density variations on imagery object's identification. Wavelet analysis is a space localized periodic analysis tool, which enables analysis of a signal in both spectral and frequency domains. In this process the reflectance signature is decomposed into different scale components represented by detailed wavelet coefficients. A new technique (R2a) is then introduced, which is based on the observation that wavelet detailed coefficients vary linearly with increasing scaling levels. In this thesis it is hypothesized that rates of change of wavelet detailed coefficients would correspond to flux density effects on spectral reflectance data. Thus, the ratio (R2a) between the coefficient of variation of these detailed coefficients (a) and reflectance (R) at each wavelength position is expected to be invariant to flux density effects in particular and multiplicative effects in general.
Analysis of the advantages obtained from using the R2a technique was achieved by comparing its classification accuracies and reliabilities with those obtained with raw spectral data, PCA (Principle Component Analysis), SAM (Spectral Angle Mapper) and selected bands. This analysis was conducted both with synthetic simulations adding flux density effects artificially and with hyperspectral imagery of vegetation and lithosols. Results indicated better classifications (accuracies and reliabilities) by the implementation of the R2a technique over the use of raw spectral data, selected bands as well as using PCA and SAM techniques. The presence of noise further increased the advantage of using the R2a technique.