High-level synthesis acceleration for an FPGA implementation of an optimized automatic target detection and classification algorithm for hyperspectral image analysis with Intel oneAPI

Recent advances in heterogeneous high performance computing (HPC) have opened new avenues for demanding remote sensing applications. Perhaps one of the most popular algorithm in target detection and identification is the automatic target detection and classification algorithm (ATDCA) widely used in the hyperspectral image analysis community. Previous research has already investigated the mapping of ATDCA on graphics processing units (GPUs) and field programmable gate arrays (FPGAs), showing impressive speedup factors that allow its exploitation in time-critical scenarios. Based on these studies, our work explores the performance portability of a tuned OpenCL implementation across a range of processing devices including multicore processors, GPUs and other accelerators. This approach differs from previous papers, which focused on achieving the optimal performance on each platform. Here, we are more interested in the following issues: (1) evaluating if a single code written in OpenCL allows us to achieve acceptable performance across all of them, and (2) assessing the gap between our portable OpenCL code and those hand-tuned versions previously investigated. Our study includes the analysis of different tuning techniques that expose data parallelism as well as enable an efficient exploitation of the complex memory hierarchies found in these new heterogeneous devices. Experiments have been conducted using hyperspectral data sets collected by NASA's Airborne Visible Infra- red Imaging Spectrometer (AVIRIS) and the Hyperspectral Digital Imagery Collection Experiment (HYDICE) sensors. To the best of our knowledge, this kind of analysis has not been previously conducted in the hyperspectral imaging processing literature, and in our opinion it is very important in order to really calibrate the possibility of using heterogeneous platforms for efficient hyperspectral imaging processing in real remote sensing missions.

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.

In hyperspectral image analysis one of the most important tasks is target detection, requiring the execution of algorithms with high computational complexity. Recently, research efforts have focused on on-board real-time target detection to provide timely responses for swift decisions. Therefore, it is necessary to use a technology that provides the performance needed for real-time target detection, and at the same time meets the satellite payload requirements. Field-programmable gate arrays (FPGAs) have very interesting properties in terms of performance, size and power consumption, which have become the standard option for on-board processing. In this letter, we present a hardware optimized implementation for FPGAs of the automatic target detection and classification algorithm (ATDCA) using the Gram–Schmidt (GS) method for orthogonalization purposes. The ATDCA-GS algorithm is directly coded using VHDL and verified on a Virtex-7 XC7VX690T FPGA using real hyperspectral data (collected by Hyperspectral Digital Imagery Collection Experiment (HYDICE) sensor and by NASA's Airborne Visible/Infrared Imaging Spectrometer (AVIRIS)) and a synthetic image. Experimental results demonstrate that our hardware version of the ATDCA-GS algorithm outperforms previous implementations (multicore processors, GPUs and accelerators) in both computation time (obtaining real-time performance) and power consumption, demonstrating the suitability of FPGAs for this purpose.

Automatic target detection in hyperspectral images is a task that has attracted a lot of attention recently. In the last few years, several algoritms have been developed for this purpose, including the well-known RX algorithm for anomaly detection, or the automatic target detection and classification algorithm (ATDCA), which uses an orthogonal subspace projection (OSP) approach to extract a set of spectrally distinct targets automatically from the input hyperspectral data. Depending on the complexity and dimensionality of the analyzed image scene, the target/anomaly detection process may be computationally very expensive, a fact that limits the possibility of utilizing this process in time-critical applications. In this paper, we develop computationally efficient parallel versions of both the RX and ATDCA algorithms for near real-time exploitation of these algorithms. In the case of ATGP, we use several distance metrics in addition to the OSP approach. The parallel versions are quantitatively compared in terms of target detection accuracy, using hyperspectral data collected by NASA's Airborne Visible Infra-Red Imaging Spectrometer (AVIRIS) over the World Trade Center in New York, five days after the terrorist attack of September 11th, 2001, and also in terms of parallel performance, using a massively Beowulf cluster available at NASA's Goddard Space Flight Center in Maryland.

In the last decades, the problem of target detection has received considerable attention in remote sensing applications. When this problem is tackled using hyperspectral images with hundreds of bands, the use of high-performance computing (HPC) is essential. One of the most popular algorithms in the hyperspectral image analysis community for this purpose is the automatic target detection and classification algorithm (ATDCA). Previous research has already investigated the mapping of ATDCA on HPC platforms such as multicore processors, graphics processing units (GPUs), and field-programmable gate arrays (FPGAs), showing impressive speedup factors (after careful fine-tuning) that allow for its exploitation in time-critical scenarios. However, the lack of standardization resulted in most implementations being too specific to a given architecture, eliminating (or at least making extremely difficult) code reusability across different platforms. In order to address this issue, we present a portability study of an implementation of ATDCA developed using the open computing language (OpenCL). We focus on cross-platform parameters such as performance, energy consumption, and code design complexity, as compared to previously developed (hand-tuned) implementations. Our portability study analyzes different strategies to expose data parallelism as well as enable the efficient exploitation of complex memory hierarchies in heterogeneous devices. We also conduct an assessment of energy consumption and discuss metrics to analyze the quality of our code. The conducted experiments—using synthetic and real hyperspectral data sets collected by the Hyperspectral Digital Imagery Collection Experiment (HYDICE) and NASA’s Airborne Visible Infra-Red Imaging Spectrometer (AVIRIS)—demonstrate, for the first time in the literature, that portability across different HPC platforms can be achieved for real-time target detection in hyperspectral missions.

The automatic mixed pixel classification (AMPC) considered in this chapter is fully computer automated and can be implemented to automatically detect and classify targets with no human intervention. Like the automatic subpixel detection discussed in Chapters 5–6 AMPC can be also categorized into unsupervised mixed pixel classification and anomaly classification. The former classifies mixed pixels in an unsupervised manner, where the required unsupervised target knowledge is the a posteriori target information generated directly from the image data as noted in Chapter 5. By contrast, the latter extends anomaly detection to anomaly classification, in which case the detected anomalies can be classified with no need of unsupervised target knowledge. Depending upon availability of a priori target knowledge two versions of unsupervised MPC, referred to as desired target detection and classification algorithm (DTDCA) and automatic target detection and classification algorithm (ATDCA), are presented in this chapter. The DTDCA is applied to a situation that there is knowledge about specific targets to be classified, whereas ATCDA can be used to classify targets of interest present in an unknown image scene without a priori target knowledge. As a consequence, they result in different applications.

This study investigates the information content of polarimetric SAR imagery for use within automatic target detection and classification algorithms. Key questions such as the stability of polarimetric information and its relationship with image resolution are addressed. Using relatively simple polarimetric features, such as the percentage of pure odd and even bounce scattering events, we show how it is possible to identify the differences between two classes of military vehicle. The use of the radial power spectral density is proposed as a measure of the spatial distribution of these odd and even bounce scattering events, again enabling the two classes to be distinguished.

This study investigates the information content of polarimetric SAR imagery for use within automatic target detection and classification algorithms. Key questions such as the stability of polarimetric information and its relationship with image resolution are addressed. Using relatively simple polarimetric features, such as the percentage of pure odd and even bounce scattering events, we show how it is possible to identify the differences between two classes of military vehicle. The use of the radial power spectral density is proposed as a measure of the spatial distribution of these odd and even bounce scattering events, again enabling the two classes to be distinguished.

Automatic target recognition (ATR) in hyperspectral imagery is a challenging problem due to recent advances of remote sensing instruments which have significantly improved sensor's spectral resolution. As a result, small and subtle targets can be uncovered and extracted from image scenes, which may not be identified by prior knowledge. In particular, when target size is smaller than pixel resolution, target recognition must be carried out at subpixel level. Under such circumstance, traditional spatial-based image processing techniques are generally not applicable and may not perform well if they are applied. The work presented here investigates this issue and develops spectral-based algorithms for automatic spectral target recognition (ASTR) in hyperspectral imagery with no required a priori knowledge, specifically, in reconnaissance and surveillance applications. The proposed ASTR consists of two stage processes, automatic target generation process (ATGP) followed by target classification process (TCP). The ATGP generates a set of targets from image data in an unsupervised manner which will subsequently be classified by the TCP. Depending upon how an initial target is selected in ATGP, two versions of the ASTR can be implemented, referred to as desired target detection and classification algorithm (DTDCA) and automatic target detection and classification algorithm (ATDCA). The former can be used to search for a specific target in unknown scenes while the latter can be used to detect anomalies in blind environments. In order to evaluate their performance, a comparative and quantitative study using real hyperspectral images is conducted for analysis.

This study investigates the information content of polarimetric SAR imagery for use within automatic target detection and classification algorithms. Key questions such as the stability of polarimetric information and its relationship with image resolution are addressed. Using relatively simple polarimetric features, such as the percentage of pure odd and even bounce scattering events, we show how it is possible to identify the differences between two classes of military vehicle. The use of the radial power spectral density is proposed as a measure of the spatial distribution of these odd and even bounce scattering events, again enabling the two classes to be distinguished.

There are two main approaches to hyperspectral target detection: anomaly detection techniques, which detect outliers substantially different from the background, and spectral signature techniques, which require as an input a user-defined target signature. Oftentimes, however, the target signature may not be known, or there may be unexpected targets in the image, which are unknown but still of interest. As a result, algorithms that can automatically extract potential target signatures without any a priori knowledge are of great interest. In this work, a fusion-based algorithm is developed that takes advantage of both spatial and spectral information to automatically extract the spectral signatures of potential targets of interest. The performance of several target detection algorithms is compared for both the proposed spatially-spectrally estimated (SSE) target signature and the initial target signature used by the Automatic Target Detection and Classification Algorithm (ATDCA). It is shown that the SSE signature leads to improved automatic spectral target recognition (ATSR) performance than the ATDCA algorithm for the test conducted on the AVIRIS Indian Pines dataset.

Least square unmixing approach has been successfully applied to hyperspectral remotely sensed images for subpixel target detection. It can detect target with size less than a pixel by estimating its abundance fraction resident in each pixel. In order for the this approach to be effective, the number of bands must be larger than or equal to that of signatures to be classified, i.e., the number of equations should be no less than the number of unknowns. This ensures that there are sufficient dimensions to accommodate orthogonal projections resulting from the individual signatures. It is known as band number constraint (BNC). Such inherent constraint is not an issue for hyperspectral images since they generally have hundreds of bands, which is more than the number of signatures resident within images. However, this may not be true for multispectral images where the number of signatures to be classified might be greater than the number of bands. In order to relax this constraint, an extension of the least square approach is presented. With a set of least square filters that are nonlinearly combined, endmember detection for multispectral images can be realized. Furthermore, to detect targets in unknown background is a greater challenge. That is the well known Automatic Target Recognition (ATR) programs. In this paper, we also proposed a Multispectral Target Generation Process (MTGP) that will automatic search for potential targets in the image scene. The effectiveness of the proposed method is evaluated by SPOT images. The experimental results show significantly improves in classification performance than Orthogonal Subspace Projection (OSP) and Automatic Target Detection and Classification Algorithm (ATDCA).

Hyperspectral remotely sensed imagery is rapidly developed recently. It collects radiance from the ground with hundreds of channels which results in hundreds of co-registered images. How to process this huge amount of data is a great challenge, especially when no information of the image scene is available. Under this circumstance, anomaly detection becomes more difficult. Several methods are devoted to this problem, such as the well-known RX algorithm and high-moment statistics approaches. The RX algorithm can detect all anomalies in single image but it can not discriminate them. On the other hand, the high-moment statistics approaches use criterion such as skewness and kurtosis to find the projection directions to detect anomalies. In this paper we propose an effective algorithm for anomaly detection and discrimination extended from RX algorithm, called Background Whitened Target Detection Algorithm. It first modeled the background signature with Gaussian distribution and applied the whitening process. After the process, the background will distribute as i.i.d. Gaussian in all spectral bands. Those pixels did not fit in the distribution will be the anomalies. Then Automatic Target Detection and Classification Algorithm (ATDCA) is applied to search for those distinct spectrum automatically and classify them as anomalies. Since ATDCA can also estimated the abundance fraction of each target resident in one pixel by applying Sum-to-one and Nonnegativity constraints, the proposed method can also be applied in a constrained fashion. The experimental results show that the proposed method can improve RX algorithm by discriminate the anomalies and also outperform high-moment approaches in terms of computational complexity.

Detecting concealed targets in an unknown environment presents a great challenge in hyperspectral image analysis since the prior knowledge about targets, the background and environment is not available. In this paper, a computer-aided detection and classification method (CADCM) for concealed targets is proposed. It is fully computer-automated and requires no a priori information at all. It consists of three successive processes: (1) a band selection procedure; (2) a band ratioing approach; and (3) an automatic target detection and classification algorithm (ATDCA). The effectiveness of the proposed CADCM is evaluated by HYperspectral Digital Imagery Collection Experiment (HYDICE) image scenes. The results show that targets which are concealed by shade, natural background or covered by man-made objects can be effectively detected and further classified by CADCM.

Band selection has been widely used in hyperspectral image processing for dimension reduction. In this paper, a recursive SAM-based band selection (RSAM-BBS) method is proposed. Once two initial bands are given, RSAM-BBS is performed in a sequential manner, and at each step the band that can best describe the spectral separation of two hyperspectral signatures is added to the bands already selected until the spectral angle reaches its maximum. In order to demonstrate the utility of the proposed band selection method, an anomaly detection algorithm is developed, which first extracts the anomalous target spectrum from the original image using automatic target detection and classification algorithm (ATDCA), followed by maximum spectral screening (MSS) to estimate the background average spectrum, then implements RSAM-BBS to select bands that participate in the subsequent adaptive cosine estimator (ACE) target detection. As shown in the experimental result on the AVIRIS dataset, less than five bands selected by the RSAM-BBS can achieve comparable detection performance using the full bands.