Linear Spectral Mixture Model in Moderate Spatial Resolution Image Data
Abstract
The concept of spectral mixture offers a wide range of applications in the
Remote Sensing area. The application of this concept, however, requires the
prior estimation of the component’s (endmembers) spectral response. This
latter requirement can be achieved by different methods, as reported in the
literature, such as techniques for the detection of pure pixels, use of spectral
libraries, and field radiometric measurements. Among those, the most often
used is the pure pixel approach. In this approach, the components’ spectral
reflectances are estimated by means of pixels covered entirely by a single
component. This approach offers the advantage of allowing the extraction of
the required spectral reflectance directly from the image data. This approach,
however, becomes increasingly unfeasible as the spatial resolution of the
image data decreases, due to the larger ground area covered by a single pixel.
In this study we propose a methodology to estimate the spectral reflectance for
each component class in moderate spatial resolution image data, by applying
the linear mixing model (MLME), and higher spatial resolution image data as
auxiliary data. It is expected that this methodology will provide a more
practical way to implement the spectral mixture approach to moderate
resolution image data, allowing in this way the expansion of the information
about the components’ proportions across larger areas, up-scaling information
in regional and global studies. Experiments were carried out using CCD (20
m ground resolution) and IRMSS (80 m ground resolution) and WFI (260 m
ground resolution) CBERS-2 image data, as medium and moderate spatial
resolution data, respectively. The spectral reflectances for the components in
the IRMSS and WFI CBERS-2 spectral bands are estimated by applying the
proposed methodology. The reliability of the proposed methodology was
assessed by both analyzing scatter plots for CBERS-2 data and by comparing
the fraction images produced by image data sets of the sensors analyzed.
Remote Sensing area. The application of this concept, however, requires the
prior estimation of the component’s (endmembers) spectral response. This
latter requirement can be achieved by different methods, as reported in the
literature, such as techniques for the detection of pure pixels, use of spectral
libraries, and field radiometric measurements. Among those, the most often
used is the pure pixel approach. In this approach, the components’ spectral
reflectances are estimated by means of pixels covered entirely by a single
component. This approach offers the advantage of allowing the extraction of
the required spectral reflectance directly from the image data. This approach,
however, becomes increasingly unfeasible as the spatial resolution of the
image data decreases, due to the larger ground area covered by a single pixel.
In this study we propose a methodology to estimate the spectral reflectance for
each component class in moderate spatial resolution image data, by applying
the linear mixing model (MLME), and higher spatial resolution image data as
auxiliary data. It is expected that this methodology will provide a more
practical way to implement the spectral mixture approach to moderate
resolution image data, allowing in this way the expansion of the information
about the components’ proportions across larger areas, up-scaling information
in regional and global studies. Experiments were carried out using CCD (20
m ground resolution) and IRMSS (80 m ground resolution) and WFI (260 m
ground resolution) CBERS-2 image data, as medium and moderate spatial
resolution data, respectively. The spectral reflectances for the components in
the IRMSS and WFI CBERS-2 spectral bands are estimated by applying the
proposed methodology. The reliability of the proposed methodology was
assessed by both analyzing scatter plots for CBERS-2 data and by comparing
the fraction images produced by image data sets of the sensors analyzed.
Keywords
Mistura espectral; resolução espacial moderada; dados CBERS; Spectral mixture; moderate spatial resolution; CBERS Data
