Towards the Development of next Generation Remote Sensing Technology – Erdas Imagine Incorporates a Higher Order Feature Extraction Technoque Based on Ica

Analysis of multi/hyperspectral imagery necessitates a selection of an optimal subset of bands in order to avoid the Hughes phenomena and parameter estimation problems due to interband correlation. This can be achieved by employing feature extraction techniques for significant reduction of data dimensionality. From the perspective of statistical pattern recognition, feature extraction refers to a process, whereby a data space is transformed into a feature space, in which the original data is represented by a reduced number of effective features, retaining most of the intrinsic information content. Feature extraction from remote sensing imagery must also aim to enhance the discriminability of surface materials based on their spectral characteristics. Conventional remote sensing feature extraction techniques such as Principal Component Analysis (PCA) and Minimum Noise Fraction (MNF) model the multi/hyperspectral image data with a multivariate Gaussian distribution. Inadequacy of such algorithms stems from the Gaussian distribution assumption, which is only an assumption rather than a demonstrable property of most remote sensing data. In order to overcome this fundamental limitation, we at Leica Geosystems have developed a feature extraction technique based on Independent Component Analysis (ICA) that exploits the higher order statistical characteristics of multi/hyperspectral imagery. In this paper, we will provide a theoretical formulation of ICA and prove its superiority over PCA and MNF in a mathematical framework. Our illustrations would also demonstrate the potential benefits of employing ICA in improving the performance of several multi/hyperspectral analysis techniques such as spectral unmixing, classification, and anomaly detection.