Parallelised fault-tolerant Integer KLT implementation for lossless hyperspectral image compression on board satellites

The Karhunen-Loève Transform (KLT) has been used as a spectral decorrelator in multi-component image compression. The Integer KLT is the integer approximation of KLT, which enables lossless hyperspectral image compression. In this paper, the effect of single-bit errors on the performance of the Integer KLT is investigated. An error detection and correction (EDAC) method that is based on the Freivald´s simple checker is proposed for the matrix factorization part of the algorithm. A technique to reduce the computational complexity is proposed too, which is based on a fixed sampling of the covariance matrix calculation. The low-complexity fault-tolerant Integer KLT is implemented on an 8-core Texas Instruments DSP (TMS320C6678) using the OpenMP^® environment. Experimental results on the compression performance, latency and power consumption are reported. The parallelized Integer KLT implementation processes a Hyperion spaceborne hyperspectral image in 3 seconds with a throughput of 66.9 Mbps.