CLASSIFICATION CHAPTER 12 The Classification Problem A Dermanis

CLASSIFICATION CHAPTER 12 The Classification Problem A. Dermanis

Absolute Classification Prior determination of the spectral reflectance characteristics of all possible classes Creation of spectral libraries. Restricting factor: Assessment of atmospheric parameters Requires: Effective reduction of atmospheric effects (effective global monitoring of the atmosphere). Large number of well-distributed bands (hyperspectral or ultraspectral sensors) Relative Classification Pixels are classified in the same class when their values in all bands are similar. No Restriction: Atmospheric Requires: influence is External data, collected by field work (at the same time epoch with satellite imagery). the same, also for pixels with ground data Supervised Classification: Ground data introduced before classification. information Unsupervised Classification: Ground data introduced after classification. A. Dermanis

Absolute Clasification: Class centers determined from spectral library. Relative Clasification: Unsupervised: Class centers determined from clustering algorithm. Relative Clasification: Supervised: Class centers determined from ground collected data (pixel samples for each class) Classification by pixel position in spectral space A. Dermanis

Classification Problems Absolute definition of the classes not possible: Variation within each land cover type - No distinct class separation. Dependence on the particular application. Correction for atmospheric influence not completely possible: Global atmospheric monitoring: determines at atmospheric absorbance (Tθ, Tφ) but not atmospheric diffusion due to scattering (ED, LP) z(λ) LS(λ) + A(λ) ρ(λ) + Β(λ) 1 A(λ) = π Τφ(λ) [Τφ(λ) Ε 0(λ) cosω + ΕD(λ) Β(λ) = LP(λ) A. Dermanis

Classification Problems Limited number of bands in multispectral sensors: Only a discrete version of spectral firm is viewed Variation of the spectral signature within single class Presence of mixed pixels A. Dermanis