Hyperspectral Data Processing: Algorithm Design and Analysis

33.6 Hyperspectral Compression

Hyperspectral compression becomes a necessity when hyperspectral data volume goes beyond the capability of used computers. This is particularly crucial for space-borne hyperspectral imaging sensors with limited computing power due to the payload constraint. In addition, due to limitation on bandwidth and data transmission rate, data preprocessing is generally required prior to data being down-linked to the ground stations for transmission and communication. Data compression provides an effective means of reducing data volumes and can be interpreted in various ways. From an information theory point of view, data compression performs data compaction with either no loss of entropy or entropy reduction. From a signal/image processing point of view, data compression performs lossless and lossy coding in the sense of some optimal criteria such as MSE, SNR, and peak SNR (PSNR). Eventually, the effectiveness of a data compression technique must be evaluated by a certain performance measure. One generally used for this purpose is compression ratio (CR), which is defined by the ratio of the original data size to compressed data size. Unfortunately, this criterion may not be applicable to hyperspectral data exploitation where subsample and mixed-sample signal sources may be inadvertently suppressed if no extra care is taken due to the fact that such signal sources may occupy only a single data sample vector or very few data sample vectors. To address this issue, ...