Verification of statistical properties for hyperspectral images: Heteroscedasticity and non-stationarity

This paper investigates the heteroscedasticity and non-stationarity, two statistical properties, of hyperspectral remote sensing data. In the field of mathematical sciences, a collection of variables is heteroscedastic if there are sub-populations that have different variances or volatilities than others, while a non-stationary process refers to a stochastic process whose joint probability distribution are changing when shifted in time or space. To be treat as sequences, hyperspectral data are investigated via Bartlett Test and Wald-Wolfowitz Runs Test to verify the heteroscedasticity and non-stationarity, respectively. Most experimental results fail to pass Bartlett Test and Wald-Wolfowitz Runs Rest statistically significant, indicating that both heteroscedasticity and non-stationarity are intrinsic properties of spectral response sequence.