Trym Vegard Haavardsholm

Real-time anomaly detection in hyperspectral images using multivariate normal mixture models and GPU processing

Hyperspectral imaging, which records a detailed spectrum of light arriving in each pixel, has many potential uses in remote sensing as well as other application areas. Practical applications will typically require real-time processing of large data volumes recorded by a hyperspectral imager. This paper investigates the use of graphics processing units (GPU) for such real-time processing. In particular, the paper studies a hyperspectral anomaly detection algorithm based on normal mixture modelling of the background spectral distribution, a computationally demanding task relevant to military target detection and numerous other applications. The algorithm parts are analysed with respect to complexity and potential for parallellization. The computationally dominating parts are implemented on an Nvidia GeForce 8800 GPU using the Compute Unified Device Architecture programming interface. GPU computing performance is compared to a multi-core central processing unit implementation. Overall, the GPU implementation runs significantly faster, particularly for highly data-parallelizable and arithmetically intensive algorithm parts. For the parts related to covariance computation, the speed gain is less pronounced, probably due to a smaller ratio of arithmetic to memory access. Detection results on an actual data set demonstrate that the total speedup provided by the GPU is sufficient to enable real-time anomaly detection with normal mixture models even for an airborne hyperspectral imager with high spatial and spectral resolution.

An airborne real-time hyperspectral target detection system

An airborne system for hyperspectral target detection is described. The main sensor is a HySpex pushbroom hyperspectral imager for the visible and near-infrared spectral range with 1600 pixels across track, supplemented by a panchromatic line imager. An optional third sensor can be added, either a SWIR hyperspectral camera or a thermal camera. In real time, the system performs radiometric calibration and georeferencing of the images, followed by image processing for target detection and visualization. The current version of the system implements only spectral anomaly detection, based on normal mixture models. Image processing runs on a PC with a multicore Intel processor and an Nvidia graphics processing unit (GPU). The processing runs in a software framework optimized for large sustained data rates. The platform is a Cessna 172 aircraft based close to FFI, modified with a camera port in the floor.

Required Spectral Resolution for Bioaerosol Detection Algorithms Using Standoff Laser-Induced Fluorescence Measurements

We have developed and tested a standoff biological aerosol detection demonstrator employing ultraviolet laser-induced fluorescence. It is based on commercially available components including a pulsed 355-nm laser and an intensified charge-coupled device camera. Biological warfare simulants and interferents were released and measured in open air field and closed-chamber laboratory tests. We analyzed the experimental data at different spectral resolutions, using statistics-based anomaly detection, and spectral angle mapping algorithms. The results show that less than 20 spectral channels in the 350-700-nm spectral region are sufficient in order to discriminate between the agents released using these methods. This corresponds to sacrificing high spectral resolution for the benefit of more photons in each channel and reduced computation time.

Compact camera for multispectral and conventional imaging based on patterned filters.

A multispectral camera concept is presented. The concept is based on using a patterned filter in the focal plane, combined with scanning of the field of view. The filter layout has stripes of different bandpass filters extending orthogonally to the scan direction. The pattern of filter stripes is such that all bands are sampled multiple times, while minimizing the total duration of the sampling of a given scene point. As a consequence, the filter needs only a small part of the area of an image sensor. The remaining area can be used for conventional 2D imaging. A demonstrator camera has been built with six bands in the visible and near infrared, as well as a panchromatic 2D imaging capability. Image recording and reconstruction is demonstrated, but the quality of image reconstruction is expected to be a main challenge for systems based on this concept. An important advantage is that the camera can potentially be made very compact, and also low cost. It is shown that under assumptions that are not unreasonable, the proposed camera concept can be much smaller than a conventional imaging spectrometer. In principle, it can be smaller in volume by a factor on the order of several hundred while collecting the same amount of light per multispectral band. This makes the proposed camera concept very interesting for small airborne platforms and other applications requiring compact spectral imagers.

Evaluation of biological aerosol stand-off detection at a field trial

We have performed a field trial to evaluate technologies for stand-off detection of biological aerosols, both in daytime and at night. Several lidar (light detection and ranging) systems were tested in parallel. We present the results from three different lidar systems; one system for detection and localization of aerosol clouds using elastic backscattering at 1.57 μm, and two systems for detection and classification of aerosol using spectral detection of ultraviolet laser-induced fluorescence (UV LIF) excited at 355 nm. The UV lidar systems were utilizing different technologies for the spectral detection, a photomultiplier tube (PMT) array and an intensified charge-coupled device (ICCD), respectively. During the first week of the field trial, the lidar systems were measuring towards a semi-closed chamber at a distance of 230 m. The chamber was built from two docked standard 20-feet containers with air curtains in the short sides to contain the aerosol inside the chamber. Aerosol was generated inside the semi-closed chamber and monitored by reference equipments, e.g. slit sampler and particle counters. Signatures from several biological warfare agent simulants and interferents were measured at different aerosol concentrations. During the second week the aerosol was released in the air and the reference equipments were located in the centre of the test site. The lidar systems were measuring towards the test site centre at distances of either 230 m or approximately 1 km. In this paper we are presenting results and some preliminary signal processing for discrimination between different types of simulants and interference aerosols.

Anisotropic Scattered Data Interpolation for Pushbroom Image Rectification

This paper deals with fast and accurate visualization of pushbroom image data from airborne and spaceborne platforms. A pushbroom sensor acquires images in a line-scanning fashion, and this results in scattered input data that need to be resampled onto a uniform grid for geometrically correct visualization. To this end, we model the anisotropic spatial dependence structure caused by the acquisition process. Several methods for scattered data interpolation are then adapted to handle the induced anisotropic metric and compared for the pushbroom image rectification problem. A trick that exploits the semiordered line structure of pushbroom data to improve the computational complexity several orders of magnitude is also presented.

A physics-based statistical signature model for hyperspectral target detection

This paper presents a statistical signature model that accounts for variability in the measured radiance spectrum from a target, based on an extensive physical model. Spectral variability is simulated in Modtran using a targets reflectance spectrum and the resulting set of possible radiance spectra are represented by a statistical distribution function. The model incorporates the likely signature variability, taking into account estimates of variability and uncertainty in the physical imaging conditions. Estimates of the adjacency effect and secondary illumination are included. The model is tested on hyperspectral data by performing signature-specific detection. A simple method for combining signature and background information for detection purposes is also presented and tested. Good detection results are obtained, even for targets in difficult illumination conditions.

Real-time georeferencing for an airborne hyperspectral imaging system

The paper describes the georeferencing part of an airborne hyperspectral imaging system based on pushbroom scanning. Using ray-tracing methods from computer graphics and a highly efficient representation of the digital elevation model (DEM), georeferencing of high resolution pushbroom images runs in real time by a large margin. By adapting the georeferencing to match the DEM resolution, the camera field of view and the flight altitude, the method has potential to provide real time georeferencing, even for HD video on a high resolution DEM when a graphics processing unit (GPU) is used for processing.

Band selection for hyperspectral target detection based on a multinormal mixture anomaly detection algorithm

The paper outlines a new method for band selection derived from a multivariate normal mixture anomaly detection method. The method consists in evaluating detection performance in terms of false alarm rates for all band configurations obtainable from an input image by selecting and combining bands according to selection criteria reflecting sensor physics. We apply the method to a set of hyperspectral images in the visible and near-infrared spectral domain spanning a range of targets, backgrounds and measurement conditions. We find optimum bands, and investigate the feasibility of defining a common band set for a range of scenarios. The results suggest that near optimal performance can be obtained using general configurations with less than 10 bands. This may have implications for the choice of sensor technology in target detection applications. The study is based on images with high spectral and spatial resolution from the HySpex hyperspectral sensor.

Physical subspace models for invariant material identification: subspace composition and detection performance

We study material identification in a forest scene under strongly varying illumination conditions, ranging from open sunlit conditions to shaded conditions between dense tree-lines. The algorithm used is a physical subspace model, where the pixel spectrum is modelled by a subspace of physically predicted radiance spectra. We show that a pure sunlight and skylight model is not sufficient to detect shaded targets. However, by expanding the model to also represent reflected light from the surrounding vegetation, the performance of the algorithm is improved significantly. We also show that a model based on a standardized set of simulated conditions gives results equivalent to those obtained from a model based on measured ground truth spectra. Detection performance is characterized as a function of subspace dimensionality, and we find an optimum at around four dimensions. This result is consistent with what is expected from the signal-to-noise ratio in the data set. The imagery used was recorded using a new hyperspectral sensor, the Airborne Spectral Imager (ASI). The present data were obtained using the visible and near-infrared module of ASI, covering the 0.4-1.0 μm region with 160 bands. The spatial resolution is about 0.2 mrad so that the studied targets are resolved into pure pixels.