Noise robustness ICA feature extraction algorithm for hyperspectral image

Feature extraction is important to hyperspectral image processing in that it can distinguish special featured object from background clutter and remove redundant information. An ICA (independent component analysis) based on the feature extraction algorithm for hyperspectral remote sensing data is proposed. In order to handle the over-sensitivity of ICA to noise and data imperfection, the MNF (maximum noise fraction) is adopted as the replacement of conventional principal component analysis. The UICA (undercomplete ICA) led by the MNF not only raises the time efficiency, but also maintains the extracting ability of ICA. The performance of the algorithm is verified by the results of HYIDCE and PHI experiments.