The problem of detecting an anomaly based on two sets of data, the first one that is assumed to represent a quiescent condition, and the second that may contain an anomaly, is addressed. Using estimated mutual information as a discriminating indicator of change, a detector is configured and interpreted, with the changing parameter modeled as the outcome of a random variable. The relationship of the proposed detector to the standard generalized likelihood ratio test is also examined. It is found that the resulting approach merges concepts in information theory, for which a Bayesian assumption is made for the underlying change parameter, with classical decision theory, for which a frequentist assumption for the change parameter is utilized.

Generation and control of humidity in a testing environment is crucial when evaluating a chemical vapor sensor as water vapor in the air can not only interfere with the sensor itself, but also react with a chemical analyte changing its composition. Upon constructing a split-flow humidity generator for chemical vapor sensor development, numerous issues were observed due to instability of the generated relative humidity level and drift of the humidity over time. By first fixing the initial relative humidity output of the system at 50%, we studied the effects of flowrate on stabilization time along with long term stability for extended testing events. It was found that the stabilization time can be upwards of 7 h, but can be maintained for greater than 90 h allowing for extended experiments. Once the stabilization time was known for 50% relative humidity output, additional studies at differing humidity levels and flowrates were performed to better characterize the system. At a relative humidity of 20% there was no time required to stabilize, but when increased to 80% this time increased to over 4 h. With this information we were better able to understand the generation process and characterize the humidity generation system, output stabilization and possible modifications to limit future testing issues.

Solid particle aerosol generation can be a costly technique that may have limited applications for a researcher. Herein, we discuss a low-cost method of solid aerosol generation for less than $1000 USD. The aerosol generation system was validated with acetaminophen and syloid 244 by studying the aerosolization into a chamber using this lab-built low-cost solid aerosol generator. This method used an inexpensive Venturi aspirator valve to pull the material from a hopper and disperse it into an 85 L chamber, creating a non-recirculating aerosol environment. The demonstrated system is a modification of a previously reported low-cost aerosol generator by the addition of electronic control valves automating the aerosolization process resulting in increased repeatability of air volume ejected into the chamber as well as decreasing the retrograde emission of materials. In each experiment, an initial spike of material was observed on the particle counter with exponential decay of total particles as they fell out of suspension or were consumed by the particle counter. In addition, the lab-built system was directly compared to a more expensive commercially available belt-fed Venturi aerosol generator and our experiments show that both methods produced similar results in regards to the particle distribution and time to create a stable aerosol environment. The addition of inexpensive electronic valves to this simple Venturi aspirator opens the area of solid particle aerosol generation to a larger audience without the high-cost burden normally associated with other commercially available technologies.

Understanding a system's performance while operating under different scenarios is difficult because of the vast number of varying parameters that need to be accounted for. To mitigate some of the difficulty a model can be developed that provides some predictability in a system's performance thereby reducing material usage and laboratory time. It is therefore prudent to understand these parameters and capture that information in order to increase the predictability of a system, especially prior fielding. Through modeling, we connect laboratory scale data with potential scenarios in the field to accomplish this. In this paper, we show that through the modeling of a combination of spectra and instrument operating characteristics we can provide a predictive capability of a system's performance. Our anomaly detection algorithm can predict a limit of anomaly detection (LOAD) for potential scenarios and then compare them to actual data for validation of our predictive capability. We show similar LOADs in both simulation and actual data collected. We further develop our model to account for realistic field scenarios and evaluate changes in performance.

Lifelong learning capitalizes on the shared skill structure present in a stream of tasks that arrive over time to improve upon the performance of single-task learners. In contemporary lifelong learning applications, it is often the case that there are multiple sensing modalities or views associated with each task. A crucial aspect in lifelong multitask multiview learning is to capture not only the shared structure among the tasks but also across views effectively. In this work, a nonparametric kernel-based learning framework is adopted to model even nonlinear shared structures in the tasks and views in a flexible and robust way. An efficient lifelong learning formulation is derived by judicious approximation of the per-task learning objectives, based on which the shared skill libraries can be updated online in function space. Numerical tests verify the efficacy of the proposed approach.

Building on our previous development of a compact, portable, and low SWaP gas analyzer (11” x 6.7” x 5.1”, 7.8 lbs) based on photoacoustic spectroscopy and using broadband quantum cascade laser arrays, we demonstrate here compositional analysis of airborne aerosols using this instrument. With an integration time of 330-ms per laser, and ~70 seconds for a spectrum covering 950-1500 cm-1, our instrument showed a detection sensitivity at the mg/m3 level for solid and liquid-loaded solid aerosols. Additionally, Malathion-loaded aerosols can be discriminated from pure Syloid aerosols based on their absorption features. The preliminary results show a potential path for developments of a portable real-time aerosol composition analyzer.

The detection of bulk materials is well-understood and many transduction methodologies exist. In contrast the detection of distributed or dispersed materials is still under study due the unique sequence of events under which this this occurs. For dispersed materials the problem is twofold, first you need to intercept or sample a location containing an analyte of interest and second you must be able to detect and identify that analyte. In addition, intercepting or sampling from sparsely contaminated areas is a more difficult problem as there is more background clutter due to less analyte available for interrogation and identification. Potential dispersed threats may include IED residues or disseminated materials dispersed in order contaminate an area with harmful chemicals. Using technologies such as Raman spectroscopy can provide real-time unique chemical-specific information to detect dispersed materials. However, understanding adequate sampling methods based on the instrument physical operation characteristics can help reduce false negatives and improve maneuverability through contested areas by bounding operational limitations. Since disseminated materials are deposited on a surfaces in a log-Normal fashion, the deposition pattern can be modeled and the potential ability to detect can be determined by understanding the probability of intercept of an analyte by the sampling method., i.e., for Raman the potential of a focused laser to illuminate an analyte containing location. The operating characteristics in question are the area of interrogation, repetition rate of the sampling method, and the speed at which the sampling is completed. In this paper, deposition patterns are modeled, and a CW Raman instrument is used to determine probability of intercept for several area-based concentrations, at different speeds, and with different interrogation areas. The data is analyzed based on both a predicted model and actual data. Determining and understanding these operating characteristics will aid in understanding of the necessary sampling, i.e., laser intercept, in order to provide desired confidence levels for detection.

The inherent wealth of information associated with hyperspectral data provides a data stream that could be leveraged for situational awareness or providing immediate user feedback. However, the enormous amount of data that is produced by some system's data stream requires longer processing times and often post-processing techniques. Therefore, it is prudent to develop real-time hyperspectral processing techniques that are capable of operating at maneuver speeds. Anomaly detection techniques applied to higher order statistics of the hyperspectral data can provide immediate user feedback for awareness. Determining capabilities prior to applying directly to a system is also informative and provides an in silico point of reference. In this paper, we show, through the use of a real-time simulator (RTS) in the MATLAB environment, a method for simulating the processing speed of a data stream based on how data is received from the instrument. In this work, the RTS provides sub 100ms capabilities based on non-optimized code within the MATLAB environment and is largely limited by the write speed in MATLAB. Utilizing virtual memory and the flexibility of MATLAB allows for simulating real-time capabilities of already obtained hyperspectral data prior to implementing it on a device. Additionally, applying the algorithm to a simulated ground truth data provides a theoretical limit of anomaly detection (LOAD). We further compare theoretical LOADs with actual anomaly detection capabilities in a laboratory environment.

Policy gradient methods have been widely used in reinforcement learning (RL), especially thanks to their facility to handle continuous state spaces, strong convergence guarantees, and low-complexity updates. Training of the methods for individual tasks, however, can still be taxing in terms of the learning speed and the sample trajectory collection. Lifelong learning aims to exploit the intrinsic structure shared among a suite of RL tasks, akin to multitask learning, but in an efficient online fashion. In this work, we propose a lifelong RL algorithm based on the kernel method to leverage nonlinear features of the data based on a popular union-of-subspace model. Experimental results on a set of simple related tasks verify the advantage of the proposed strategy, compared to the single-task and the parametric counterparts.

"Intelligent Systems", machines that respond to the world around them and machines that interact with humans to change the dynamic of physical or social interaction. In the chemical, biological, radiological, nuclear, and explosives (CBRNE) detection world the focus is currently on the former definition with the advent of what some call "smart systems" based upon the common goal of creating CBRNE sensors that can respond and adapt to the environment in which they operate. Responses can be as simple as tipping and cuing of additional assets or resources to address changes in environment or operating conditions. Or on deeper level, the control systems and algorithms that operate/control these systems autonomously adapting to changes in both the operational characteristics and current conditions. Ideally a system could self-monitor its inherent capabilities and, for example, adjust dwell or sampling times base upon learned or defined characteristics. The concept of self-learning or machine learning within a sensor aligns with the current popularism of artificial intelligence (AI). However, within the CBRNE sensor community there is an inherent lack of the depth and breadth of data to actualize a functional AI to address these problems. In reality the information or data could be quite limited and the need to be able to operate anywhere in the world without long periods of acclimation must be stressed. Therefore CBRNE Intelligent Systems must be able to operate in a traditional sense, turn it on and function, and be adaptable to "long term" operations adjusting to both environmental and operational characteristic changes.

The detection of steady state visual evoked potentials (SSVEPs), an evoked response to visual stimuli, has been identified as an effective solution for brain computer interface systems and as a probe for neurocognitive investigations of visually related tasks. However, since they recorded as part of the scalp-based electroencephalogram (EEG) signals their detection is challenging as they are buried amongst the normal brain signals. Blind source separation methods, such as independent vector analysis (IVA), have been shown to be capable of enhancing and improving signal detection by exploiting the diversity within individual datasets while simultaneously exploiting the complimentary information across datasets. In general, IVA is highly flexible with a general solution space; however, it is not guaranteed to converge to a meaningful minimum, by incorporating a problem specific constraint we can shrink the solution space insuring a relevant solution. In this work, we present a novel multiset data framework for EEG recordings and apply our constrained power spectra IVA (CP-IVA) to a publicly available SSVEP dataset. We compare the prediction accuracy of CP-IVA with that of an optimized processing stream developed for that dataset, as well as a canonical correlation analysis (CCA) based approach, showing that CP-IVA achieves better average performance and is more robust across the population of subjects with a higher minimum detection rate. More importantly, CP-IVA achieves this performance with minimal pre-processing and without the need to train complex classifiers

Hyperspectral imaging (HSI) has become increasingly popular for sensing in defense, commercial, and academic research for its ability to acquire vast amounts of information, relatively quickly, at stand-off distances. As such, the need for rapid or near-real time data reduction is becoming more evident especially when immediate knowledge of the area under investigation is required such as in contested areas, the scene of natural disasters, and other similar scenarios. While analysis of the underlying spectral information may provide specific information about materials present, in HSI determining an anomaly can be just as informative in scenarios such as CB detection for avoidance. Therefore, a rapid, real-time HSI anomaly detection algorithm is merited. In this paper, we present work towards an algorithm for near-real time anomaly detection utilizing higher-order statistics and, in particular, implications due to changes in skewness and kurtosis, the 3rd and 4th central moments. We demonstrate using a visible-SWIR hyperspectral line scanner that anomalies (thiodiglycol and acetaminophen) can be detected in data that is updated to simulate real-time analysis. Changing spectral features result in changes in the probability density function, and can be specifically realized with comparisons of higher order statistics (i.e. skewness and kurtosis), thereby reducing a full spectral analysis at each voxel to a comparison of two values at each pixel. This paper explores utilizing this concept as a means for anomaly detection, evaluating different surfaces that an analyte may be present on, and lastly presents work towards automated background updates for anomaly detection on dynamic surfaces.

The detection of chemical hazards on surfaces continues to be a challenge for the sensing community. In order to minimize risk to users, a desirable configuration is a non-contact (standoff) system, which can operate a safe distance from the hazard. A conceptual solution to this challenge is the Wide-Area Mapping and Identification (WAMId) system. The WAMId prototype breadboard combines two distinct technologies, hyperspectral imaging and standoff Raman spectroscopy, operating in tandem to locate and identify anomalous areas of interest and then presumptively identify surface contaminates. In the developed configuration, a single short to mid wave infrared (SWIR/MWIR) hyperspectral camera images a scene of interest, data is processed to locate anomalous materials and the resulting coordinates from the scene are uploaded to a gimbal control which then slews an 830 nm Raman system to perform presumptive identification measurements. In this work, we present the results of the program, to include system development, and sample testing data for three chemicals.

Independent component analysis (ICA) has found wide application in a variety of areas, and analysis of functional magnetic resonance imaging (fMRI) data has been a particularly fruitful one. Maximum likelihood provides a natural formuiation for ICA and allows one to take into account multiple statistical properties of the data-forms of diversity. While use of multiple types of diversity allows for additional flexibility, it comes at a cost, leading to high variability in the solution space. In this paper, using simulated as well as fMRI-like data, we provide insight into the trade-offs between estimation accuracy and algorithmic consistency with or without deviations from the assumed model and assumptions such as the statistical independence. Additionally, we propose a new metric, cross inter-symbol interference, to quantify the consistency of an algorithm across different runs, and demonstrate its desirable performance for selecting consistent run compared to other metrics used for the task.

Small target detection is a problem common to a diverse number of fields such as radar, remote sensing, and infrared imaging. In this paper, we consider the application of feature extraction for detection of small hazardous materials in multiwavelength imaging. Since various materials may exist in the area of study each with varying degrees of reflectivity and absortion at different wavelengths of light, flexible, data-driven methods are needed for feature extraction of relevant sources. We propose the use of independent component analysis (ICA), a widely-used blind source separation method based on the statistical independence of the underlying sources. We compare 3 different prominent flavors of ICA on simulated data in a variety of environments. Then, we apply ICA to 2 multi-wavelength imaging datasets with results that suggest that features extracted are useful.

Steady state visual evoked potentials (SSVEPs) have been identified as an effective solution for brain computer interface (BCI) systems as well as for neurocognitive investigations. SSVEPs can be observed in the scalp-based recordings of electroencephalogram signals, and are one component buried amongst the normal brain signals and complex noise. We present a novel method for enhancing and improving detection of SSVEPs by leveraging the rich joint blind source separation framework using independent vector analysis (IVA). IVA exploits the diversity within each dataset while preserving dependence across all the datasets. This approach is shown to enhance the detection of SSVEP signals across a range of frequencies and subjects for BCI systems. Furthermore, we show that IVA enables improved topographic mapping of the SSVEP propagation providing a promising new tool for neuroscience and neurocognitive research.

UV Raman spectra were measured using a novel experimental configuration. This configuration allows many of the difficulties associated with UV excitation and high-power pulsed laser sources to be mitigated. Large sample areas are imaged into the detection system allowing high power excitation sources to be used while simultaneously avoiding sample degradation and multi-photon absorption effects. Such large detection areas allow large numbers of molecular scatters to be probed even with minimal penetration depth. Alignment issues between sample and collection optics are also simplified. Several common solvents were studied using 213 nm light and their spectra reported.

The availability of multi-set data, i.e., multiple datasets originating from one or more sensors at different conditions and/or from multiple subjects, enables the exploitation of complimentary information across datasets. However, popular methods to analyze such data suffer from poor performance due to unrealistic assumptions and constrained solution spaces, in addition they ignore potentially informative signal properties, forms of diversity, such as higher order statistics and associations across datasets. We propose to overcome these issues using data-driven techniques including multi-set singular value decomposition (MSVD) [1] and independent vector analysis (IVA) [2] to leverage associations across datasets and, in the case of IVA, leverage diversity using second as well as higher-order statistics [3]. We propose to leverage MSVD and IVA to enhance the detection of steady state visually evoked potentials signals in electroencephalography data using a hybrid approach.

The detection of steady state visual evoked potentials (SSVEPs) has been identified as an effective solution for brain computer interface (BCI) systems as well as for neurocognitive investigations of visually related tasks. SSVEPs are induced at the same frequency as the visual stimuli and can be observed in the scalp-based recordings of electroencephalogram signals, though they are one component buried amongst the normal brain signals and complex noise. Variations in individual response latencies as well as the presence of multiple biological artifacts complicate the use of direct frequency analysis, thus making blind source separation methods, such as independent component (ICA) and independent vector analysis (IVA) desirable solutions. IVA is a recent extension of ICA that decomposes multiple datasets simultaneously and has been been shown to be capable of enhancing and improving the detection of SSVEPs by exploiting the complimentary information that exists across EEG channels. In this work, we present a novel extension of IVA which incorporates a priori information to constrain the power spectral density (PSD) of the source estimates, known as constrained PSD IVA (CP-IVA) and demonstrate its improved SSVEP detection performance as well as stability over standard IVA and temporally constrained IVA (C-IVA).

Spectroscopic analysis is used throughout industry, academia, and other areas to differentiate and identify compounds. In many cases the compounds have highly similar spectral structures, i.e., spectral overlap and may only readily be identified as belonging to a class of materials. Current analytical methods perform well when there are clearly discernible peaks within the spectra but are known to lose discrimination power as the spectra of interest become more and more similar. To overcome this loss in detection power we propose a novel method for determining the maximum discrimination spectral bands, known as the maximum discrimination approach (MDA). MDA is based upon determining the statistical distance between two spectra for each band, and is derived by assuming each spectrum is the result of estimating the power spectral density of Gaussian noise. We demonstrate the ability of MDA to find maximum discrimination spectral bands using the spectral data of gasoline and kerosene, two related mixtures with similar spectral content.