Enhancement of high spectral resolution remote-sensing data by a noise-adjusted principal components transform

High-spectral-resolution remote-sensing data are first transformed so that the noise covariance matrix becomes the identity matrix. Then the principal components transform is applied. This transform is equivalent to the maximum noise fraction transform and is optimal in the sense that it maximizes the signal-to-noise ratio (SNR) in each successive transform component, just as the principal component transform maximizes the data variance in successive components. Application of this transform requires knowledge or an estimate of the noise covariance matrix of the data. The effectiveness of this transform for noise removal is demonstrated in both the spatial and spectral domains. Results that demonstrate the enhancement of geological mapping and detection of alteration mineralogy in data from the Pilbara region of Western Australia, including mapping of the occurrence of pyrophyllite over an extended area, are presented