Noise-Adjusted Principle Component Analysis For Hyperspectral Remotely Sensed Imagery Visualization

Abstract

Introduction In recent years, hyperspectral imaging has been developed in remote sensing, which uses hundreds of co-registered spectral channels to acquires images for the same area on the earth. Its high spectral resolution enables researchers and scientists to detect features, classify objects, and extract ground information more accurately. PCA [1] is a typical approach for high-dimensional data analysis, which assembles the major data information into the first several principal components (PCs) based on variance maximization. However, variance is not a good criterion to rank the data features because part of the variance may be from noise. The noise should be whitened before PCA, which is equivalently to rank the PCs in terms of signal-to-noise ratio. The resultant technique is called Noise-Adjusted Principal Component Analysis (NAPCA) [2]. In our research, NAPCA is employed to visualize images taken by Hyperion, the first spaceborne hyperspectral sensor onboard NASA’s EO-1 satellite.