Analysis of Compressive Sensing for Hyperspectral Remote Sensing Applications.

摘要

Compressive Sensing (CS) systems capture data with fewer measurements than traditional sensors assuming that imagery is redundant and compressible in the spectral and spatial dimensions. This thesis utilizes a model of the Coded Aperture Snapshot Spectral Imager-Dual Disperser (CASSI-DD) to simulate CS measurements from traditionally sensed HyMap images. A novel reconstruction algorithm that combines spectral smoothing and spatial total variation (TV) is used to create high resolution hyperspectral imagery from the simulated CS measurements. This research examines the effect of the number of measurements, which corresponds to the percentage of physical data sampled, on the quality of simulated CS data as estimated through performance of spectral image processing algorithms. The effect of CS on the data cloud is explored through principal component analysis (PCA) and endmember extraction. The ultimate purpose of this thesis is to investigate the utility of the CS sensor model and reconstruction for various hyperspectral applications in order to identify the strengths and limitations of CS. While CS is shown to create useful imagery for visual analysis, the data cloud is altered and per-pixel spectral fidelity declines for CS reconstructions from only a small number of measurements. In some hyperspectral applications, many measurements are needed in order to obtain comparable results to traditionally sensed HSI, including atmospheric compensation and subpixel target detection. On the other hand, in hyperspectral applications where pixels must be dramatically altered in order to be misclassified, such as land classification or NDVI mapping, CS shows promise.