Sebastián López

GPU Implementation of an Automatic Target Detection and Classification Algorithm for Hyperspectral Image Analysis

The detection of (moving or static) targets in remotely sensed hyperspectral images often requires real-time responses for swift decisions that depend upon high computing performance of algorithm analysis. The automatic target detection and classification algorithm (ATDCA) has been widely used for this purpose. In this letter, we develop several optimizations for accelerating the computational performance of ATDCA. The first one focuses on the use of the Gram-Schmidt orthogonalization method instead of the orthogonal projection process adopted by the classic algorithm. The second one is focused on the development of a new implementation of the algorithm on commodity graphics processing units (GPUs). The proposed GPU implementation properly exploits the GPU architecture at low level, including shared memory, and provides coalesced accesses to memory that lead to very significant speedup factors, thus taking full advantage of the computational power of GPUs. The GPU implementation is specifically tailored to hyperspectral imagery and the special characteristics of this kind of data, achieving real-time performance of ATDCA for the first time in the literature. The proposed optimizations are evaluated not only in terms of target detection accuracy but also in terms of computational performance using two different GPU architectures by NVIDIA: Tesla C1060 and GeForce GTX 580, taking advantage of the performance of operations in single-precision floating point. Experiments are conducted using hyperspectral data sets collected by three different hyperspectral imaging instruments. These results reveal considerable acceleration factors while retaining the same target detection accuracy for the algorithm.

Hyperspectral Unmixing on GPUs and Multi-Core Processors: A Comparison

One of the main problems in the analysis of remotely sensed hyperspectral data cubes is the presence of mixed pixels, which arise when the spatial resolution of the sensor is not able to separate spectrally distinct materials. Due to this reason, spectral unmixing has become one of the most important tasks for hyperspectral data exploitation. However, unmixing algorithms can be computationally very expensive, a fact that compromises their use in applications under real-time constraints. For this purpose, in this paper we develop two efficient implementations of a full hyperspectral unmixing chain on two different kinds of high performance computing architectures: graphics processing units (GPUs) and multi-core processors. The proposed full unmixing chain is composed for three stages: (i) estimation of the number of pure spectral signatures or endmembers, (ii) automatic identification of the estimated endmembers, and (iii) estimation of the fractional abundance of each endmember in each pixel of the scene. The two computing platforms used in this work are inter-compared in the context of hyperspectral unmixing applications. The GPU implementation of the proposed methodology has been implemented using the compute devide unified architecture (CUDA) and the cuBLAS library, and tested on two different GPU architectures: NVidia™ GeForce GTX 580 and NVidia™ Tesla C1060. It provides real-time unmixing performance in two different analysis scenarios using hyperspectral data collected by NASAs Airborne Visible Infra-Red Imaging Spectrometer (AVIRIS) over the Cuprite mining district in Nevada and the World Trade Center complex in New York City. The multi-core implementation, developed using the applications program interface (API) OpenMP and the Intel Math Kernel Library (MKL) used for matrix multiplications, achieved near real-time performance in the same scenarios. A comparison of both architectures in terms of performance, cost and mission payload considerations is given based on the results obtained in the two considered data analysis scenarios.

The Promise of Reconfigurable Computing for Hyperspectral Imaging Onboard Systems: A Review and Trends

Hyperspectral imaging is an important technique in remote sensing which is characterized by high spectral resolutions. With the advent of new hyperspectral remote sensing missions and their increased temporal resolutions, the availability and dimensionality of hyperspectral data is continuously increasing. This demands fast processing solutions that can be used to compress and/or interpret hyperspectral data onboard spacecraft imaging platforms in order to reduce downlink connection requirements and perform a more efficient exploitation of hyperspectral data sets in various applications. Over the last few years, reconfigurable hardware solutions such as field-programmable gate arrays (FPGAs) have been consolidated as the standard choice for onboard remote sensing processing due to their smaller size, weight, and power consumption when compared with other high-performance computing systems, as well as to the availability of more FPGAs with increased tolerance to ionizing radiation in space. Although there have been many literature sources on the use of FPGAs in remote sensing in general and in hyperspectral remote sensing in particular, there is no specific reference discussing the state-of-the-art and future trends of applying this flexible and dynamic technology to such missions. In this work, a necessary first step in this direction is taken by providing an extensive review and discussion of the (current and future) capabilities of reconfigurable hardware and FPGAs in the context of hyperspectral remote sensing missions. The review covers both technological aspects of FPGA hardware and implementation issues, providing two specific case studies in which FPGAs are successfully used to improve the compression and interpretation (through spectral unmixing concepts) of remotely sensed hyperspectral data. Based on the two considered case studies, we also highlight the major challenges to be addressed in the near future in this emerging and fast growing research area.

Analysis of fast block matching motion estimation algorithms for video super-resolution systems

In general, all the video super-resolution (SR) algorithms present the important drawback of a very high computational load, mainly due to the huge amount of operations executed by the motion estimation (ME) stage. Commonly, there is a trade-off between the accuracy of the estimated motion, given as a motion vector (MV), and the computational cost associated. In this sense, the ME algorithms that explore more exhaustively the search area among images use to deliver better MVs, at the cost of a higher computational load and resources use. Due to this reason, the proper choice of a ME algorithm is a key factor not only to reach real-time applications, but also to obtain high quality video sequences independently of their characteristics. Under the hardware point of view, the preferred ME algorithms are based on matching fixed-size blocks in different frames. In this paper, a comparison of nine of the most representative Fast Block Matching Algorithms (FBMAs) is made in order to select the one which presents the best tradeoff between video quality and computational cost, thus allowing reliable real-time hardware implementations of video super-resolution systems.

Performance Evaluation of the H.264/AVC Video Coding Standard for Lossy Hyperspectral Image Compression

In this paper, a performance evaluation of the state-of-the-art H.264/AVC video coding standard is carried out with the aim of determining its feasibility when applied to hyperspectral image compression. Results are obtained based on configuring diverse parameters in the encoder in order to achieve an optimal trade-off between compression ratio, accuracy of unmixing and computation time. In this sense, simulations are developed in order to measure the spectral angles and signal-to-noise ratio (SNR), achieving compression ratios up to 0.13 bits per pixel per band (bpppb) for real hyperspectral images. Moreover, in this work it is detected which blocks in the encoder contribute the most to performance improvements of the compression task for the particular case of this type of images, and which ones are not relevant at all and hence could be removed. This conclusion yields to reduce the future design complexities of potential low-power/real-time hyperspectral encoders based on H.264/AVC for remote sensing applications.

A Novel Architecture for Hyperspectral Endmember Extraction by Means of the Modified Vertex Component Analysis (MVCA) Algorithm

There is presently a high interest in the spatial industry to develop high-performance on-board processing platforms with a high degree of flexibility, so they can adapt to varying mission needs and/or to future space standards. For this purpose, Field Programmable Gate Array (FPGA) devices have demonstrated to offer an excellent compromise between flexibility and performance. This work presents a novel FPGA-based architecture to be used as part of the hyperspectral linear unmixing processing chain. In particular, this paper introduces a new architecture for hyperspectral endmember extraction accordingly to the Modified Vertex Component Analysis (MVCA) algorithm, which provides a better figure of merit in terms of endmember extraction accuracy versus computational complexity than the Vertex Component Analysis (VCA) algorithm. Two versions of the MVCA algorithm which differ on the use of floating point or integer arithmetic for iteratively projecting the hyperspectral cube onto a direction orthogonal to the subspace spanned by the endmembers already computed have been mapped onto a Xilinx Virtex-5 FPGA. The results demonstrate that both versions are capable of processing hyperspectral images captured by the NASAs AVIRIS sensor in real-time, showing the latter a better performance in terms of hardware resources and processing speed. Furthermore, our proposal constitutes the first published architecture for extracting the endmembers from a hyperspectral image based on the VCA principle and thus, it provides a basis for future FPGA implementations of state-of-the-art hyperspectral algorithms with similar characteristics, such as the Automatic Target Generation Process (ATGP) or the Orthogonal Subspace Projection (OSP) algorithms.

A Low-Computational-Complexity Algorithm for Hyperspectral Endmember Extraction: Modified Vertex Component Analysis

Endmember extraction represents one of the most challenging aspects of hyperspectral image processing. In this letter, a new algorithm for endmember extraction, named modified vertex component analysis (MVCA), is presented. This new technique outperforms the popular vertex component analysis (VCA) by applying a low-complexity orthogonalization method and by utilizing integer instead of floating-point arithmetic when dealing with hyperspectral data. The feasibility of this technique is demonstrated by comparing its performance with VCA on synthetic mixtures as well as on the well-known Cuprite hyperspectral image. MVCA shows promising results in terms of much lower computational complexity, still reproducing similar endmember accuracy than its original counterpart. Moreover, the features of this algorithm combined with state-of-the-art hardware implementations qualify MVCA as a good potential candidate for all those applications in which real time is a must.

Mapping control-intensive video kernels onto a coarse-grain reconfigurable architecture: the H.264/AVC deblocking filter

Deblocking filtering represents one of the most compute intensive tasks in an H.264/AVC standard video decoder due to its demanding memory accesses and irregular data flow. For these reasons, an efficient implementation poses big challenges, especially for programmable platforms. In this sense, the mapping of this decoders functionality onto a C-programmable coarse-grained reconfigurable architecture named ADRES (architecture for dynamically reconfigurable embedded systems) is presented in this paper, including results from the evaluation of different topologies. The results obtained show a considerable reduction in the number of cycles and memory accesses needed to perform the filtering as well as an increase in the degree of instruction parallelism (ILP) when compared with an implementation on a very long instruction word (VLIW) dedicated processor. This demonstrates that high ILP is achievable on the ADRES even for irregular, data-dependent kernels

A New Preprocessing Technique for Fast Hyperspectral Endmember Extraction

Hyperspectral image processing represents a valuable tool for remote sensing of the Earth. This fact has led to the inclusion of hyperspectral sensors in different airborne and satellite missions for Earth observation. However, one of the main drawbacks encountered when dealing with hyperspectral images is the huge amount of data to be processed, in particular, when advanced analysis techniques such as spectral unmixing are used. The main contribution of this letter is the introduction of a novel preprocessing (PP) module, called SE2PP, which is based on the integration of spatial and spectral information. The proposed approach can be combined with existing algorithms for endmember extraction, reducing the computational complexity of those algorithms while providing similar figures of accuracy. The key idea behind SE2PP is to identify and select a reduced set of pixels in the hyperspectral image, so that there is no need to process a large amount of them to get accurate spectral unmixing results. Compared to previous approaches based on similar spatial and spatial-spectral PP strategies, SE2PP clearly outperforms their results in terms of accuracy and computation speed, as it is demonstrated with artificial and real hyperspectral images.

Parallel Implementation of the Modified Vertex Component Analysis Algorithm for Hyperspectral Unmixing Using OpenCL

Hyperspectral imaging represents the state-of-theart technique in those applications related to environmental monitoring, military surveillance, or rare mineral detection. However, one of the requirements of paramount importance when dealing with such scenarios is the ability to achieve real-time constraints taking into account the huge amount of data involved in processing this type of images. In this paper, the authors present for the first time a combination of the newly introduced modified vertex component analysis (MVCA) algorithm for the process of endmembers extraction together with the ability of GPUs to exploit its parallelism, giving, as a result, important speedup factors with respect to its sequential counterpart, while maintaining the same levels of endmember extraction accuracy than the vertex component analysis (VCA) algorithm. Furthermore, OpenCL ensures the use of generic computing platforms without being restricted to a particular vendor. The proposed approach has been assessed on a set of synthetic images as well as on the well-known Cuprite real image, showing that the most time-consuming operations are located on the matrix projection and the maximum search processes. Comparison of the proposed technique with a single-threaded C-based implementation of the MVCA algorithm shows a speedup factor of 8.87 for a 500 × 500 pixel artificial image with 20 endmembers and 7.14 for the wellknown Cuprite hyperspectral data set, including in both cases I/O transfers. Moreover, when the proposed implementation is compared with respect to a C-based sequential implementation of the VCA algorithm, a speedup of 115 has been achieved. In all the cases, the results obtained by the MVCA are the same as the ones obtained with the VCA; thus, the accuracy of the proposed algorithm is not compromised.