Investigation of correlation of empirical modes and low-frequency residues of hyperspectral image signatures

Using the hyperspectral image (HSI) Moffett Field (AVIRIS spectrometer), which is characterized by a high variety of natural and urban environments, we study the decomposition of HSI pixel signatures into empirical modes and low-frequency residues of different orders. For the study, 6 types of objects were selected: 1) vegetation; 2) concrete; 3) soil; 4) seawater (fairway); 5) shallow seawater; 6) salt. HSI noisy channels are automatically detected and excluded from consideration. It is established that the first mode is least correlated with the signature and its low-frequency residue. Correlation grows with an increase in the mode order. Modes of 1-3 orders can be attributed to weakly correlated with the pixel signature, but they carry information about the image. To determine the physical sense of high-frequency modes, the correlation of empirical modes of the same order belonging to different materials was estimated, and a high correlation level between the modes was established, decreasing with increasing the order. The average value of the mutual correlation coefficients of the modes of different objects is: for the first mode - 0.90; for the second - 0.82; the third - 0.74. Supposed that the high correlation of the first modes is associated with the influence of the atmosphere. The potential possibility of using empirical modes in algorithms for correcting the influence of the atmosphere and detecting glare pixels by the magnitude of the amplitude bursts has been established.