Band selection provides performance improvement in hyperspectral applications such as target detection, spectral unmixing and classification. Signal-to-noise ratio estimation (SNRe) as a method can be adjusted for different specific applications. SNRe is usually used to remove some low SNR bands from original hyperspectral data in a preprocessing stage and then other band selection methods are applied for the remaining high SNR bands of hyperspectral data to make the operations more efficient. In this paper, we take advantage of SNRe to select the bands which contain the largest amount of information. The wavelet transform is first used to realize the signal-noise separation and get the noise standard deviation of each band, and then the SNRs of all bands are calculated orderly. Considering some time-consuming operations in SNRe algorithm which can’t satisfy some real time applications are very suitable for high performance computing(HPC) in parallel, we design a new massively parallel algorithm to accelerate the SNR estimation algorithm on graphics processing units(GPUs) using the compute device unified architecture(CUDA) language. In addition the implementation of our GPU-based SNRe algorithm has extremely explored the possible parallelism in the C code and been debugged carefully to verify its correctness and efficiency. Experiments are conducted on two sets of real hyperspectral images and considerable acceleration is obtained.
Proc. SPIE. 9263, Multispectral, Hyperspectral, and Ultraspectral Remote Sensing Technology, Techniques and Applications V
KEYWORDS: Signal to noise ratio, Hyperspectral imaging, Data modeling, Remote sensing, Computer programming, Image analysis, Reconstruction algorithms, Chemical elements, Electrical engineering, Current controlled current source
A great challenge in hyperspectral image analysis is decomposing a mixed pixel into a collection of endmembers and their corresponding abundance fractions. This paper presents an improved implementation of Barycentric Coordinate approach to unmix hyperspectral images, integrating with the Most-Negative Remove Projection method to meet the abundance sum-to-one constraint (ASC) and abundance non-negativity constraint (ANC). The original barycentric coordinate approach interprets the endmember unmixing problem as a simplex volume ratio problem, which is solved by calculate the determinants of two augmented matrix. One consists of all the members and the other consist of the to-be-unmixed pixel and all the endmembers except for the one corresponding to the specific abundance that is to be estimated. In this paper, we first modified the algorithm of Barycentric Coordinate approach by bringing in the Matrix Determinant Lemma to simplify the unmixing process, which makes the calculation only contains linear matrix and vector operations. So, the matrix determinant calculation of every pixel, as the original algorithm did, is avoided. By the end of this step, the estimated abundance meet the ASC constraint. Then, the Most-Negative Remove Projection method is used to make the abundance fractions meet the full constraints. This algorithm is demonstrated both on synthetic and real images. The resulting algorithm yields the abundance maps that are similar to those obtained by FCLS, while the runtime is outperformed as its computational simplicity.
In order to effectively extract endmembers for hyperspectral imagery where linear mixing model may not be appropriate due to multiple scattering effects, this paper extends the simplex growing algorithm (SGA) to its kernel version. A new simplex volume formula without dimension reduction is used in SGA to form a new simplex growing algorithm (NSGA). The original data are nonlinearly mapped into a high-dimensional space where the scatters can be ignored. To avoid determining complex nonlinear mapping, a kernel function is used to extend the NSGA to kernel NSGA (KNSGA). Experimental results of simulated and real data prove that the proposed KNSGA approach outperforms SGA and NSGA.