Real-Time Implementation of the Pixel Purity Index Algorithm for Endmember Identification on GPUs

Abstract

Spectral unmixing amounts to automatically finding the signatures of pure spectral components (called endmembers in the hyperspectral imaging literature) and their associated abundance fractions in each pixel of the hyperspectral image. Many algorithms have been proposed to automatically find spectral endmembers in hyperspectral data sets. Perhaps one of the most popular ones is the pixel purity index (PPI), which is available in the ENVI software from Exelis Visual Information Solutions. This algorithm identifies the endmembers as the pixels with maxima projection values after projections onto a large randomly generated set of random vectors (called skewers). Although the algorithm has been widely used in the spectral unmixing community, it is highly time consuming as its precision asymptotically increases. Due to its high computational complexity, the PPI algorithm has been recently implemented in several high-performance computing architectures, including commodity clusters, heterogeneous and distributed systems, field programmable gate arrays, and graphics processing units (GPUs). In this letter, we present an improved GPU implementation of the PPI algorithm, which provides real-time performance for the first time in the literature.

The index array approach and the dual tiled similarity algorithm for UAS hyper-spatial image processing

“…Some research has been done in the GPU-based processing of remote sensing imagery. For example, references [4][5][6][7][8][9][10][11][12] explored hyperspectral image processing by the GPU with a focus on handling hundred spectral bands on the same image pixel. References [13][14][15][16][17][18] proposed the GPU methods for geocorrection and orthorectification (also called geo-referencing) for imagery acquired by UAS commercial off-the-shelf cameras, an airborne pushbroom imager, and high resolution satellite sensors.…”

GPU Implementation of Three-Stage Prediction with Adaptive Search Threshold for Hyperspectral Image Compression

“…In all cases, less than 34 percent of the total solver time is consumed by data transfers. Taking into consideration that the AVIRIS scanning rate is 12 Hz [27], more satellite hyperspectral sensors such as Hyperion feature 220 Hz cross-line scanning rates. This means that a hyperspectral sensor data like the AVIRIS Cuprite scene could be collected in about 5s.…”

Development of efficient GPU parallelization of WRF Yonsei University planetary boundary layer scheme

“…These make GPUs more effective than a massively parallel system built from commodity central processing units (CPUs). Usage of GPUs has been applied very successfully to deal with numerous computational problems in various domains, for instance, porting marine ecosystem model spin-up using transport matrices to GPUs (Siewertsen et al, 2013), GPU-accelerated long-wave radiation scheme of the rapid radiative transfer model for general circulation (RRTMG) models (Price et al, 2014), advances in multi-GPU smoothed particle hydrodynamics simulations (Rustico et al, 2014), speeding up the computation of WRF double moment 6-class microphysics scheme with GPU (Mielikainen et al, 2013), real-time implementation of the pixel purity index algorithm for end-member identification on GPUs (Wu et al, 2014), fat vs. thin threading approach on GPUs: application to stochastic simulation of chemical reactions (Klingbeil et al, 2012), ASAMgpu V1.0 -a moist fully compressible atmospheric model using GPUs (Horn, 2012), GPU acceleration of predictive partitioned vector quantization for ultra-spectral sounder data compression (Wei, 2011), clusters vs. GPUs for parallel automatic target detection in remotely sensed hyperspectral images (Paz et al, 2010), and a GPUaccelerated wavelet decompression system with set partitioning in hierarchical trees (SPIHT) and Reed-Solomon decoding for satellite images (Song et al, 2011), to name several.…”