Search

The challenges for providing war fighters with the best possible actionable information from diverse sensing modalities using advances in big-data and machine learning are addressed in this paper. We start by presenting intelligence, surveillance, and reconnaissance (ISR) related big-data challenges associated with the Third Offset Strategy. Current approaches to big-data are shown to be limited with respect to reasoning/understanding. We present a discussion of what meaning making and understanding require. We posit that for human-machine collaborative solutions to address the requirements for the strategy a new approach, Qualia Exploitation of Sensor Technology (QuEST), will be required. The requirements for developing a QuEST theory of knowledge are discussed and finally, an engineering approach for achieving situation understanding is presented.

Current cyber-related security and safety risks are unprecedented, due in no small part to information overload and skilled cyber-analyst shortages. Advances in decision support and Situation Awareness (SA) tools are required to support analysts in risk mitigation. Inspired by human intelligence, research in Artificial Intelligence (AI) and Computational Intelligence (CI) have provided successful engineering solutions in complex domains including cyber. Current AI approaches aggregate large volumes of data to infer the general from the particular, i.e. inductive reasoning (pattern-matching) and generally cannot infer answers not previously programmed. Whereas humans, rarely able to reason over large volumes of data, have successfully reached the top of the food chain by inferring situations from partial or even partially incorrect information, i.e. abductive reasoning (pattern-completion); generating a hypothetical explanation of observations. In order to achieve an engineering advantage in computational decision support and SA we leverage recent research in human consciousness, the role consciousness plays in decision making, modeling the units of subjective experience which generate consciousness, qualia. This paper introduces a novel computational implementation of a Cognitive Modeling Architecture (CMA) which incorporates concepts of consciousness. We apply our model to the malware type-classification task. The underlying methodology and theories are generalizable to many domains.

For over 30 years, the U.S. Army Aviation and Missile Research, Development, and Engineering Center (AMRDEC) has specialized in characterizing the performance of infrared (IR) imaging systems in the laboratory and field. In the late 90’s, AMRDEC developed the Automated IR Sensor Test Facility (AISTF) which allowed efficient deployment testing of aviation and missile IR sensor systems. More recently, AMRDEC has tested many uncooled infrared (UCIR) sensor systems that have size, weight, power, and cost (SWAPC) benefits for certain fielded U.S. Army imaging systems. To compensate for relatively poor detector sensitivities, most UCIR systems operate with very fast focal ratio or F-number (f/#) optics. AMRDEC has recently found that measuring the Noise Equivalent Temperature Difference (NETD) with traditional techniques used with cooled infrared systems produce biased results when applied to systems with faster f/# values or obscurations. Additionally, in order to compare these camera cores or sensor systems to one another, it is imperative to scale the NETD values for f/#, focus distance, and waveband differences accurately. This paper will outline proper measurement techniques to report UCIR camera core and system-level NETD, as well as demonstrate methods to scale the metric for these differences.

Computational imaging techniques can be used to extend the depth of field of imaging sensors such that the sensors become less expensive to build and athermalize with no loss to performance. Optical phase can be manipulated to create an image that is optimized for a detection and tracking algorithm as well as reconstructed digitally to form an image suitable for viewing. A typical low-cost sensor which is used for target detection and tracking may run an algorithm which requires different features and resolution from its imagery than would a system optimized for a human. This offers a unique opportunity to optimize both optics and image processing for a system which can maximize mission performance as well as minimize production cost. Simple computational techniques have not yet been successful in passive, low-signal environments due to noise issues. This study examines the use of a simple computational technique in an algorithmic application in which optimal reconstruction may occur with lower noise. This paper will describe the model, simulation, and prototype which resulted from a detailed and novel system design and modeling process. The goal of this effort is to accurately model the anticipated performance and to prove actual cost savings of a tracking sensor which employs computational imaging techniques.

SPICE is a high resolution imaging spectrometer operating at extreme ultraviolet wavelengths, 70.4 – 79.0 nm and 97.3 - 104.9 nm. It is a facility instrument on the Solar Orbiter mission. SPICE will address the key science goals of Solar Orbiter by providing the quantitative knowledge of the physical state and composition of the plasmas in the solar atmosphere, in particular investigating the source regions of outflows and ejection processes which link the solar surface and corona to the heliosphere. By observing the intensities of selected spectral lines and line profiles, SPICE will derive temperature, density, flow and composition information for the plasmas in the temperature range from 10,000 K to 10MK. The instrument optics consists of a single-mirror telescope (off-axis paraboloid operating at near-normal incidence), feeding an imaging spectrometer. The spectrometer is also using just one optical element, a Toroidal Variable Line Space grating, which images the entrance slit from the telescope focal plane onto a pair of detector arrays, with a magnification of approximately x5. Each detector consists of a photocathode coated microchannel plate image intensifier, coupled to active-pixel-sensor (APS). Particular features of the instrument needed due to proximity to the Sun include: use of dichroic coating on the mirror to transmit and reject the majority of the solar spectrum, particle-deflector to protect the optics from the solar wind, and use of data compression due to telemetry limitations.

A qualia exploitation of sensor technology (QUEST) motivated architecture using algorithm fusion and adaptive feedback loops for face recognition for hyperspectral imagery (HSI) is presented. QUEST seeks to develop a general purpose computational intelligence system that captures the beneficial engineering aspects of qualia-based solutions. Qualia-based approaches are constructed from subjective representations and have the ability to detect, distinguish, and characterize entities in the environment Adaptive feedback loops are implemented that enhance performance by reducing candidate subjects in the gallery and by injecting additional probe images during the matching process. The architecture presented provides a framework for exploring more advanced integration strategies beyond those presented. Algorithmic results and performance improvements are presented as spatial, spectral, and temporal effects are utilized; additionally, a Matlab-based graphical user interface (GUI) is developed to aid processing, track performance, and to display results.

Detecting network intruders and malicious software is a significant problem for network administrators and security experts. New threats are emerging at an increasing rate, and current signature and statistics-based techniques are failing to keep pace. Intelligent systems that can adapt to new threats are needed to mitigate these new strains of malware as they are released. This research develops a system that uses contextual relationships and information across different layers of abstraction to detect malware based on its qualia, or essence. By looking for the underlying concepts that make a piece of software malicious, this system avoids the pitfalls of static solutions that focus on predefined signatures or anomaly thresholds. If successful, this type of qualia-based system would provide a framework for developing intelligent classification and decision-making systems for any number of application areas.

For over half a century, scientists and engineers have worked diligently to advance computational intelligence. One application of interest is how computational intelligence can bring value to our war fighters. Automatic Target Recognition (ATR) and sensor fusion efforts have fallen far short of the desired capabilities. In this article we review the capabilities requested by war fighters. When compared to our current capabilities, it is easy to conclude current Combat Identification (CID) as a Family of Systems (FoS) does a lousy job. The war fighter needed capable, operationalized ATR and sensor fusion systems ten years ago but it did not happen. The article reviews the war fighter needs and the current state of the art. The article then concludes by looking forward to where we are headed to provide the capabilities required.

The interest in new infrared materials has grown rapidly during the last decade, one reason being the increasing cost of traditional Germanium, with in the meantime, a decrease of the cost of infrared detectors. In response to this Umicore has developed the GASIR® range of optical materials. A key strength of the new material is that it can be molded, leading to particularly cost effective solutions for high volume requirements. This paper reviews the GASIR®1 material relative to some of the existing materials and presents a case study of an optical design using GASIR®. In particular the effect of thermal cycling on survivability and performance are examined. The case study includes Umicore's recently developed iDLC(tm) coating. This coating complies with the specifications for "Diamond Like Carbon" coatings and can thus suit a broad range of applications such as thermography, fire fighting, etc.

E.C Adam defined Situational Awareness (SA) as "the mental representation and understanding of objects, events, people, system states, interactions, environmental conditions, and other situation-specific factors affecting human performance in complex and dynamic tasks. Stated in lay terms, SA is simply knowing what is going on so you can figure out what to do." We propose a novel idea to assist the human in gaining SA. Our hypothesis is that nature uses qualia as a compression scheme to represent the many concepts encountered in everyday life. Qualia enable humans to quickly come up with SA based on many complex measurements from their sensors, (eyes, ears, taste, touch, memory, etc.), expectations, and experiences. Our ultimate objective is to develop a computer that uses qualia concepts to transform sensor data to assist the human in gaining and maintaining improved SA. However, before any computer can use qualia, we must first define a representation for qualia that can be implemented computationally. This paper will present our representation for qualia. The representation is not simply a hierarchical aggregation of input data. Instead, it is a prediction of what will happen next, derived from computations resulting from sensory inputs and the computational engine of a qualia generator and qualia processor.

An x-ray anti-scatter grid alignment aid has been developed and its performance evaluated for portable chest radiographic exams. The grid alignment aid consists of one line laser and one cross-hair laser mounted on a supporting frame. During an x-ray exam, the two lasers generate a predetermined light pattern on the x-ray collimator to indicate the optimal x-ray source position in the three-dimensional (3D) space. Two grids commonly used in portable chest exams (40 lp/cm resolution, 6:1 and 8:1 ratio, and 130 cm focal distance) were characterized in terms of the signal-to-noise ratio improvement factor (SIF), as a function of the source image distance (SID), and x-ray tube lateral displacement from which the optimal grid operation range was determined. Data indicated the grid alignment aid was able to provide sufficient accuracy for positioning the x-ray tube to achieve optimal grid usage.

Extension of coded apertures to the MWIR introduces the effects of diffraction and other distortions not observed in shorter wavelength systems. A new approach is being developed under the DARPA/SPO funded LACOSTE (Large Area Coverage Optical search-while Track and Engage) program, that addresses the effects of diffraction while gaining the benefits of coded apertures, thus providing flexibility to vary resolution, possess sufficient light gathering power, and achieve a wide field of view (WFOV). The photonic MEMS "eyelid" array technology is currently being instantiated in this DARPA Surveillance program study as the "heart", mediating the flow of the incoming signal. However, speed, lifetime, packaging and scalability are critical factors for the MEMS "eyelid" technology which will determine system efficacy as well as military and commercial usefulness. The electronic eyelid array is the fundamental addressable unit for adaptive code generation and will allow the system to multiplex in time for increased resolution. The binary code which determines whether a 500μm eyelid is open or closed is referred to as the "eyelid code." Groups of eyelids can work together as a "super aperture" by virtue of a "macro-code." A macro code becomes relevant to describe how dispersed eyelids across the 0.19m x 0.19m aperture will function together. Dynamic aperture arrays were fabricated on both quartz and sapphire substrates for operation in the visible to MWIR. Both 8x8 and 40x40 element arrays were designed, fabricated, and tested with macro-codes consisting of 4, 8, and 16 unique combinations. The die were packaged and tested in ambient for robust eyelid operations. The point spread function was also measured in an optical setup with the eyelid arrays located in the aperture plane.

Umicore, known for its activities in the infrared materials and molded optics, this year launches a new infrared glass called GASIR® 3. This material can be molded using Umicore's proprietary molding technology and allows serving a wide range of new markets. Examples are a new automotive commercial application and sensing applications with their need for very small optics. Parallel to the materials development, a new coating has been developed by Umicore that allows the use of GASIR® molded optics in extremely harsh environments. The extreme performance of this new type of coating which complies with the toughest military specs will also be described.

A novel IR transmissive hard coating that offers protection to harsh environmental conditions on GASIR® and other IR materials for thermal imaging and sensing applications. iDLC has been developed to maximise both spectral and environmental performance for GASIR®. This coating can be applied to the outside surface of molded optics and windows and offers high spectral efficiencies from 1.4μm to 15μm. The ability to deposit a multi-layer structure allows broad band high efficiency anti-reflection coatings to be produced. Compared to conventional DLC, this coating offers significantly less absorption, lower reflection and thus allows higher transmission over a wide spectral band. Tests have shown that the coating offers exceptional resistance to abrasion, salt spray and humidity. The process used to manufacture iDLC has been configured for production volumes and offers a process for a wide range of applications on IR electro-optic materials.

Full-leg and full-spine imaging with standard computed radiography (CR) systems requires several cassettes/storage phosphor screens to be placed in a staggered arrangement and exposed simultaneously to achieve an increased imaging area. A method has been developed that can automatically and accurately stitch the acquired sub-images without relying on any external reference markers. It can detect and correct the order, orientation, and overlap arrangement of the subimages for stitching. The automatic determination of the order, orientation, and overlap arrangement of the sub-images consists of (1) constructing a hypothesis list that includes all cassette/screen arrangements, (2) refining hypotheses based on a set of rules derived from imaging physics, (3) correlating each consecutive sub-image pair in each hypothesis and establishing an overall figure-of-merit, (4) selecting the hypothesis of maximum figure-of-merit. The stitching process requires the CR reader to over scan each CR screen so that the screen edges are completely visible in the acquired sub-images. The rotational displacement and vertical displacement between two consecutive sub-images are calculated by matching the orientation and location of the screen edge in the front image and its corresponding shadow in the back image. The horizontal displacement is estimated by maximizing the correlation function between the two image sections in the overlap region. Accordingly, the two images are stitched together. This process is repeated for the newly stitched composite image and the next consecutive sub-image until a full-image composite is created. The method has been evaluated in both phantom experiments and clinical studies. The standard deviation of image misregistration is below one image pixel.

Modern capture and storage of digital information have given rise to tremendous repositories of data, but the ability to efficiently and effectively use this data has lagged. A prominent example is the generation of video imagery by Unmanned Aerial Vehicles. This imagery currently must be analyzed by a human observer either in real-time or playback mode (dynamic or frame-by-frame). These manpower-intensive activities become increasingly unpalatable as personnel resources shrink, so an automated system is desired that will assist in transforming the large data sets into vastly smaller sets that still retain the pertinent information. We propose a computational vision model based on physiological and behavioral characteristics of the human visual system to perform the registration of multiple video frames, followed by a fusion process that produces a video mosaic preserving the `important' features (those the human observer would deem important) from each individual frame.

We introduce an object recognition system using Gabor filters to model the biological visual field and a saccadic behavioral model to emulate biological active vision. A high resolution image containing an object of interest is first processed by an ensemble of multi-resolution, multi- orientation Gabor filters. The object can then be described by an alternating sequence of fixation coordinates and the Gabor responses at pixel locations surrounding that fixation point. Once this sequence is memorized, a complex image can be searched for the same location/feature sequence, indicating the presence of the memorized object. The model is suitable for memorization of arbitrary objects, and a simple example is presented using a human face as the object of interest.

In this paper, it is shown that Evolution Programs can be used to search the weight space for Bayesian training of a Neural Network. Bayesian Analysis is an integration problem (as opposed to an optimization problem) over weight space. The first application of the Bayesian method primarily focused on using a Gaussian approximation of the posterior distribution in an area of high probability in the weight space instead of using formal integration. More recently, training a neural network in a Bayesian fashion has been accomplished by searching weight space for areas of high probability density which obviates the need for the Gaussian assumption. In particular, a hybrid Monte-Carlo method was used to search weight space in a logical manner to obtain an arbitrarily close approximation of the integration involved in a Bayesian analysis. Genetic Algorithms have been used in the past to determine the weights in an ANN, and (with some slight modifications) are ideally suited for searching the weight space to approximate the Bayesian integration. In this respect, the Bayesian framework provides a simple and elegant way to apply Evolution Programs to the ANN training problem. While this paper concentrates on using ANNs as classifiers, the generalization to regression problems is straightforward.

A Radial Basis Function (RBF) Iterative Construction Algorithm (RICA) is presented that autonomously determines the size of the network architecture needed to perform classification on a given data set. The algorithm uses a combination of a Gaussian goodness-of-fit measure and Mahalanobis distance clustering to calculate the number of hidden nodes needed and to estimate the parameters of the hidden node basis functions. An iterative minimum squared error reduction method is used to optimize the output layer weights. RICA is compared to several neural network algorithms, including a fixed architecture multilayer perceptron (MLP), a fixed architecture RBF, and an adaptive architecture MLP, using optical character recognition and infrared image data.

This paper explores using linear regression and artificial neural networks (ANN) to model the performance of an ATR algorithm based on a given set of data. Here, a probability of detection response surfaces as a function of relevant parameters is simulated. It is then shown that this surface can be approximated using either linear regression or an ANN with good results. These regression surfaces can provide valuable information to the ATR developer/customer in terms of trying to predict ATR performance in untested areas. The application of this ATR performance modeling methodology becomes clear when we consider applying it to a common problem, such as air-to-ground target detection, where the changing parameters of the target can give a good set of data points from which to build the response curve.© (1997) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.

TheMLP Iterative Construction Algorithm (MICA) designs a Multi-Layer Perceptron (MLP) neural network as it trains. MICA adds Hidden Layer Nodes (HLNs) one at a time, separating classes on a pair-wise basis, untilthe data is projected into a linear separable space by class. Then MICA trains the Output Layer Nodes (OLNs),which results in an MLP that achieves 100% accuracy on the training data. MICA, like Backprop,'2 produces an MLP that is a minimum mean squared error approximation of the Bayes optimal discriminant function. Moreover,MICA's training technique yields novel feature selection technique and hidden node pruning technique. 1 INTRODUCTION The MLP Iterative Construction Algorithm (MICA) constructs a Multi-Layer Perceptron (MLP) neural net-work for solving classification problems. MICA trains the Hidden Layer Nodes (HLNs), then trains the Output Layer Nodes (OLNs). The resulting MLP network correctly classifies the training data to 100% accuracy. MICA generalization results compare favorably to Backprop's. On difficult training sets such as spiral data, MICA is

This paper demonstrates the usefulness of neural networks in classifying environmental samples from compound mixture data. This problem was solved by careful determination of neural network learning parameters and forward sequential selection of input features. Finally, the fundamental limit of any classifier on this data was determined using Bayes error bounding.

A model-based vision (MBV) approach to automatic target cuing/recognition (ATC/R) using real infrared (IR) image data will be presented. The MBV-ATC/R is comprised of three parts: Focus of attention (FOA), indexing, and match/search. The FOA module analyzes the IR image and extracts (segments) regions of interest that may contain targets. The focus of this article will be on the FOA portion of the MBV-ATC/R approach. In particular, three methods of FOA will be optimized, compared, and fused. The first FOA module is a Least Asymmetrical Daubechies wavelet decomposition method. The second FOA module is a physiologically based Difference of Gaussians. The third FOA module is a Morphological hit- and-miss transform. The three FOA algorithms are individually optimized using a genetic algorithm. Then an adaptive pulse coupled neural network is used to fuse the results.

This paper demonstrates the application of new pattern recognition techniques that can be used to characterize space objects. The feature space trajectory neural network (FST NN) was first presented by Leonard Neiberg and David P. Casasent in 1994 as a target identification tool. Kenneth H. Fielding and Dennis W. Ruck recently applied the hidden Markov model (HMM) classifier to a 3D moving light display identification problem and a target recognition problem, using time history information to improve classification results. This paper shows how the FST NN and HMM can be used to automate optical detection of space object anomalies. Time sequenced images produced by a simulation program are used for testing these anomaly detection algorithms. Two data sets are tested. The second data set is tested with various levels of shot noise. The first data set is more difficult to classify, with the best FST NN test achieving a 100% anomaly detection rate with 5% false alarm rate. FST NN and HMM tests on the second data set achieved 100% anomaly detection with no false alarms. With various levels of shot noise added, the FST NN achieves a 100% anomaly detection rate with 4% false alarm rate. A variety of Fourier features are tested with energy normalized low frequency coefficients producing the best consistent results across data sets and noise levels. A new FST NN test is presented that measures how well the order of a test sequence matches other sequences in the database. The original FST NN is based strictly on feature space distance, but when the order of the sequence is important, the new test is useful.© (1996) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.

An automated method for detecting microcalcification clusters is presented. The algorithm begins with a digitized mammogram and outputs the center coordinates of regions of interest (ROIs). The method presented uses a non-linear function and a 12-tap least asymmetric Daubechies (LAD12) wavelet in a tree structured filter bank to increase the signal to noise level by 10.26 dB. The signal to noise level gain achieved by the filtering allows subsequent thresholding to eliminate on average 90% of the image from further consideration without eliminating actual microcalcification clusters 95% of the time. Morphological filtering and texture analysis are then used to identify individual microcalcifications. Altogether, the method successfully detected 44 of 53 microcalcification clusters (83%) with an average of 2.3 false positive clusters per image. A cluster is considered detected if it contains 3 or more microcalcifications within a 6.4 mm by 6.4 mm area. The method successfully detected 13 of the 14 malignant cases (93%).© (1996) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.

This paper uses a high level vision model to describe the information passing and linking within the primate visual system. Information linking schemes, such as state dependent modulation and temporal synchronization, are presented as methods the vision system uses to combine information using expectation to fill in missing information and remove unneeded information. The possibility of using linking methods derived from physiologically based theoretical models to combine current image processing techniques for pattern recognition purposes is investigated. These image processing techniques are transforms such as (but not limited to) wavelet filters, hit or miss filters, morphological filters, and difference of gausian filters. These particular filters are chosen because they simulate functions that are performed in the primate visual system. To implement the physiologically motivated linking methods, the Pulse Coupled Neural Network (PCNN) is chosen as a basic building block for the vision model which performs linking at the neuronal pulse level. Last, an image fusion network which incorporates information linking based on the PCNN is described, and initial results are presented.© (1996) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.

Derivative-based feature saliency techniques were used to define the best of 25 Laws texture features for the classification of 101 malignant mass and benign mass regions. Statistical and derivative-based saliency techniques were used to select the best size, shape, contrast, and Laws texture features for the mass model. Nine features were chosen to define the model, of which four have been used by other researchers. Using this model, the regions were classified using a multilayer perceptron neural network architecture trained with an imbalanced training set weight update algorithm to obtain an overall classification accuracy of 100 percent for the segmented malignant masses with a false-positive rates of 1.8/image. The system has shown a sensitivity of 92 percent for locating malignant ROIs. The database contained 284 images (12 bit, 100 micrometers ).

This research applies statistical and artificial neural network analysis to data obtained from measurement of organic compounds in the breath of Fisher-344 rats. The Research Triangle Institute (RTI) developed a breath collection system for use with rates in order to collect and measure volatile organic compounds (VOCs) exhaled. The RTI study tested the hypothesis that VOCs, including endogenous compounds, in breath can serve as markers to exposure to various chemicals such as drugs, pesticides, or carcinogens. A comparative analysis of chromatograms showed that the administration of carbon tetrachloride dramatically altered the VOCs measured in breath; both the compounds detected and their concentrations were greatly impacted. This research demonstrated that neural network analysis and classification discriminates between exposure to carbon tetrachloride versus no exposure. It also identified the chemical compounds in rat breath that best discriminate between carbon tetrachloride exposure and either a vehicle control or no dose. For the data set analyzed, 100 percent classification accuracy was achieved in classifying cases of exposure versus no exposure. The top three marker compounds were identified. The results obtained show that neural networks can be effectively used to analyze complex chromatographic data.

Optical techniques for space object identification have been a topic of research for several decades. Imaging techniques have been widely examined for this problem. However, it has been shown that high spatial resolution cannot always be obtained due to atmospheric turbulence. Spectral measurements are of interest to determine the material content when images cannot be obtained. In this paper spectra obtained from a Fourier transform spectrometer are used to identify materials using pattern recognition-based techniques. The signal-to-noise ratio of the Sagnac-type interferometer is derived, and the errors in material identification and abundance estimation that arise from using the measured spectra to estimate these quantities are studied. Fisher discriminants and multi-layer perceptrons were used to identify materials, and a constrained least squares technique was used to estimate abundances. Results for material identification and abundance estimation are presented as a function of signal-to-noise ratio.

More than 50 million women over the age of 40 are currently at risk for breast cancer in the United States. Computer-aided diagnosis, as a second opinion to radiologists, will aid in decreasing the number of false readings of mammograms. Neural network benefits are exploited at both the classification and feature selection stages in the development of a computer-aided breast cancer diagnostic system. The multilayer perceptron is used to classify and contrast three features (angular second moment, eigenmasses, and wavelets) developed to distinguish benign from malignant lesion in a database of 94 difficult-to-diagnose digitized microcalcification cases. System performance of 74 percent correct classifications is achieved. Feature selection techniques are presented which further improve performance. Neural and decision boundary-based methods are implemented, compared, and validated to isolate and remove useless features. The contribution from this analysis is an increase to 88 percent correct classification in system performance. These feature selection techniques can also process risk factor data. © (1995) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.

Time series prediction has widespread application, ranging from predicting the stock market to trying to predict future locations of scud missiles. The embedology theorem sets forth the procedures for state space manipulation and reconstruction for time series prediction. This includes embedding the time series into a higher dimensional space in order to form an attractor, a structure defined by the embedded vectors. In this paper, embedology is combined with neural technologies in an effort to create a more accurate prediction algorithm. The algorithms tested are embedology, neural networks, and Euclidean space nearest neighbors. Local linear training methods are compared to the use of the nearest neighbors as the training set for a neural network. The results of these experiments determine that the neural algorithms have the best prediction accuracies. The performance of the nearest neighbor trained neural network validates the applicability of the local linear training set.

More than 50 million women over the age of 40 are currently at risk for breast cancer in the United States. Computer-aided diagnosis, used as a `second opinion' to radiologists, will aid in decreasing the number of false readings of mammograms. A novel feature extraction method is presented that provides increased classification power. Wavelets, previously only exploited for their segmentation benefits, are explored as features for classification. Daubechies4, Daubechies20, and biorthogonal wavelets are each investigated. Applied to 94 difficult-to- diagnose digitized microcalcification cases, performance is 74 percent correct classifications. Feature selection techniques are presented which further improve wavelet classification performance to 88 percent correct classification.© (1995) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.

Images of a microlithographic sample obtained using a new near field scanning optical microscope (NSOM) that uses force regulation of the sample-tip separation are presented. The NSOM is a research instrument fitted with a metal covered glass tip probe that defines a small aperture at the sharp end. The aperture is estimated to be on the order of 100 nanometers in diameter resulting in a resolution exceeding that of diffraction limited systems. This form of microscopy can be done both in the transmission and the reflection modes. The force regulation mechanism produces a simultaneously obtained scanned force microscope image of the topography thus permitting correlative imaging of the sample. The samples are imaged in transmission and reflection near field optical format, with white light and with coherent light. The results are compared with other forms of IC imaging and characterization, namely scanned force microscopy and scanning electron microscopy.

This paper describes a wavelet-based system for computing localized velocity fields associated with time-sequential imagery. The approach combines the mathematical rigor of the multiresolution wavelet analysis with well known spatiotemporal frequency flow computation principles. The foundation of the approach consists of a unique, nonhomogeneous multiresolution wavelet filter bank designed to extract moving objects in a 3D image sequence based on their location, size and speed. The filterbank is generated by an unconventional 3D subband coding scheme that generates twenty orientation tuned filters at each spatial and temporal resolution. The frequency responses of the wavelet filter bank are combined using a least-squares method to assign a velocity vector to each spatial location in an image sequence. Several examples are provided to demonstrate the flow computation abilities of the wavelet vector motion sensor.

The most difficult stage of automated target recognition is segmentation. Current segmentation problems include faces and tactical targets; previous efforts to segment these objects have used intensity and motion cues. This paper develops a color preprocessing scheme to be used with the other segmentation techniques. A neural network is trained to identify the color of a desired object, eliminating all but that color from the scene. Gabor correlations and 2D wavelet transformations will be performed on stationary images; and 3D wavelet transforms on multispectral data will incorporate color and motion detection into the machine visual system. The paper will demonstrate that color and motion cues can enhance a computer segmentation system. Results from segmenting faces both from the AFIT data base and from video taped television are presented; results from tactical targets such as tanks and airplanes are also given. Color preprocessing is shown to greatly improve the segmentation in most cases.© (1994) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.

Two verification systems, face and speaker, are fused to form a single identity verification system. The Karhunen-Loeve Transform (KLT) is used for dimensional reduction, and a back- propagation neural net is used for classification. Verification involved training a net for each individual in the database for two classes of outputs, `Joe' or `not Joe.' The base speaker identification system used Cepstral analysis for feature extraction and a distortion measure for classification. Verification in this case involved performing the KLT on the Cepstral coefficients and then classifying using a two-class neural net for each individual. KLT feature reduction is compared to alternative linear methods, and the KLT is found to provide superior performance. The fusion of the two base verification systems is shown to provide superior performance over either system alone.

A machine that can read unconstrained handwritten words remains an unsolved problem. For example, automatic entry of handwritten documents into a computer is yet to be accomplished. Most systems attempt to segment letters of a word and read words one character at a time. Segmenting a handwritten word is very difficult and often, the confidence of the results is low. Another method which avoids segmentation altogether is to treat each word as a whole. This research investigates the use of Fourier Transform coefficients, computed from the whole word, for the recognition of handwritten words. To test this concept, the particular pattern recognition problem studied consisted of classifying four handwritten words, `Buffalo', `Vegas', `Washington', and `City' from the SUNY post office database. Several feature subsets of the Fourier coefficients are examined. The best recognition performance of 76.2% was achieved when the Karhunen-Loeve transform was computed on the Fourier coefficients and those features were fed into a multilayer perceptron.© (1994) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.

A neural network approach to obtaining upper and lower bounds on the Bayes error rate for pattern recognition problems is presented. The approach is developed using the key concept of resubstitution and leave-one-out testing from conventional nonparametric error estimation techniques. The neural network approach is evaluated by applying it to several 8D, two-class `toy' problems, where the Bayes error rate is known. The neural network error estimate for a high-dimensional problem with an unknown Bayes error rate is also compared to error estimates obtained using conventional nonparametric estimation techniques. Using the neural network procedure, the upper bound of the Bayes error rate is reliably found for problems with complex decision boundary surfaces. Alternative testing approaches are suggested for reducing the difference between the bounds and the true Bayes rate.

The Federal Bureau of Investigation (FBI) has recently sanctioned a wavelet fingerprint image compression algorithm developed for reducing storage requirements of digitized fingerprints. This research implements an optical wavelet transform of a fingerprint image, as the first step in an optical fingerprint identification process. Wavelet filters are created from computer- generated holograms of biorthogonal wavelets, the same wavelets implemented in the FBI algorithm. Using a detour phase holographic technique, a complex binary filter mask is created with both symmetry and linear phase. The wavelet transform is implemented with continuous shift using an optical correlation between binarized fingerprints written on a Magneto-Optic Spatial Light Modulator and the biorthogonal wavelet filters. A telescopic lens combination scales the transformed fingerprint onto the filters, providing a means of adjusting the biorthogonal wavelet filter dilation continuously. The wavelet transformed fingerprint is then applied to an optical fingerprint identification process. Comparison between normal fingerprints and wavelet transformed fingerprints shows improvement in the optical identification process, in terms of rotational invariance.© (1994) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.

A spatio-temporal method for identifying objects contained in an image sequence is presented. The Hidden Markov Model (HMM) technique is used as the classification algorithm, making classification decisions based on a spatio-temporal sequence of observed object features. A five class problem is considered. Classification accuracies of 100% and 99.7% are obtained for sequences of images generated over two separate regions of viewing positions. HMMs trained on image sequences of the objects moving in opposite directions showed a 98.1% successful classification rate by class and direction of movement. The HMM technique proved robust to image corruption with additive correlated noise and had a higher accuracy than a single look nearest neighbor method. A real image sequence of one of the objects used was successfully recognized with the HMMs trained on synthetic data. This study shows the temporal changes that observed feature vectors undergo due to object motion hold information that can yield superior classification accuracy when compared to single frame techniques.

Current U.S. Air Force interests include a desire to track an object based on its shape once it has been designated as a target. This paper discusses the use of a correlation based system to track an object through a series of images based on templates derived from previous image frames. The ability to track is extended to sequences which include multiple objects of interest within the field of view. This is accomplished by comparing the height and shape of the template autocorrelation to the peaks in the correlation of the template with the next scene. The result is to identify the region in the next scene which best matches the designated target. In addition to correlation plane postprocessing, an adaptive window is used to determine the template size in order to reduce the effects of correlator walk-off. The image sequences used were taken from a Forward Looking Infrared (FLIR) sensor mounted onboard a DC-3 aircraft. The images contain a T-55 tank and both an M-113 and a TAB-71 armored personnel carrier moving in a columnized formation along a dirt road. The goals of this research were to (1) track targets in the presence of other, and sometimes brighter, targets of similar shape; (2) to maintain small tracking errors; and (3) to reduce the effects of correlator walk-off.

The remotely measured surface vibration signatures of tactical military ground vehicles are investigated for use in target classification and identification friend or foe (IFF) systems. The use of remote surface vibration sensing by a laser radar reduces the effects of partial occlusion, concealment, and camouflage experienced by automatic target recognition systems using traditional imagery in a tactical battlefield environment. Linear Predictive Coding (LPC) efficiently represents the vibration signatures and nearest neighbor classifiers exploit the LPC feature set using a variety of distortion metrics. Nearest neighbor classifiers achieve an 88 percent classification rate in an eight class problem, representing a classification performance increase of thirty percent from previous efforts. A novel confidence figure of merit is implemented to attain a 100 percent classification rate with less than 60 percent rejection. The high classification rates are achieved on a target set which would pose significant problems to traditional image-based recognition systems. The targets are presented to the sensor in a variety of aspects and engine speeds at a range of 1 kilometer. The classification rates achieved demonstrate the benefits of using remote vibration measurement in a ground IFF system. The signature modeling and classification system can also be used to identify rotary and fixed-wing targets.

Investigation of two neural architectures is performed in two dimensions using both synthetic and real imagery. Our model follows the work performed by H. Ogmen and S. Gagne in 1990 on the fly's visual system. We extended their model to a two-dimensional architecture and also developed a new model by adding long-term memory at the input--adaptive model. Our investigation compares the response of the adaptive model against the original Ogmen and Gagne's cell-activity model. The output of both models were further processed using casual and noncausal moving average filters to help remove tonic image elements and identify direction of motion. Our simulations show that the adaptive model can be used to segment motion from sequences of imagery.© (1993) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.

A biologically inspired tactile sensor, the AFIT Neural Tactile Sensor (NTS), is proposed and experimentally evaluated. Incorporating ANN pattern recognition techniques, the NTS offers previously unattainable benefits in sensor size, simplicity, flexibility, sensitivity, and utility.

A real-time recurrent learning algorithm was applied to a five class radar target identification problem. Wideband radar signatures were generated for five aircraft classes. Since an aircraft in flight is constantly in motion, a radar can measure sequences of radar signatures as the aspect angle changes. A radar can also generate aspect angle estimates by using kinematic information from aircraft position and velocity measurements. A recurrent neural network computer program (implementing a real time recurrent learning algorithm) was trained to recognize these sequences of radar signatures. Each radar signature was described by 6 external input features: the estimated target azimuth, the estimated target width, and 4 noisy amplitude values from 2 peak range bins. Nine consecutive radar signatures were sufficient to achieve a test set accuracy of 96%.

This paper presents results from experiments on determining the sigmoidal transfer function in Intel's Electronically Trainable Analog Neural Network (ETANN), the 80170NX chip. Accurate simulator training off-chip is needed in order to reduce training time and to minimize chip-in-loop training (on-chip), which if done in excess, can decrease the chip's useful life. For this reason accurate characterization of the ETANN chip is of significant importance to application designers for off-chip simulation. A series of tests were performed to collect data from eight ETANN chips for analysis. After computing an average response value from eight chips and performing a minimum mean-square-error search for a gain coefficient (also called a hardness parameter in the sigmoidal function), a transfer function and gain coefficient were found. Using this transfer function and gain coefficient in a custom neural network simulator, called Neural Graphics, a performance evaluation was accomplished. A test shows a direct correlation between the Neural Graphics simulator output and the ETANN chip's output using the same synaptic weights and the test data. Moreover, for this test we have found that Neural Graphics, while using the characterized transfer function from this research, performed in a superior manner to that of iNNTS simulators. For researchers who desire to interface their own custom simulator with the ETANN hardware, a similar procedure as developed in this paper should be followed.© (1993) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.

User recognition is examined using neural and conventional techniques for processing speech and face images. This article for the first time attempts to overcome this significant problem of distortions inherently captured over multiple sessions (days). Speaker recognition uses both Linear Predictive Coding (LPC) cepstral and auditory neural model representations with speaker dependent codebook designs. For facial imagery, recognition is developed on a neural network that consists of a single hidden layer multilayer perceptron backpropagation network using either the raw data as inputs or principal components of the raw data computed using the Karhunen-Loeve Transform as inputs. The data consists of 10 subjects; each subject recorded utterances and had images collected for 10 days. The utterances collected were 400 rich phonetic sentences (4 sec), 200 subject name recordings (3 sec), and 100 imposter name recordings (3 sec). Face data consists of over 2000, 32 X 32 pixel, 8 bit gray scale images of the 10 subjects. Each subsystem attains over 90% verification accuracy individually using test data gathered on days following the training data.

This paper presents a 3D multiresolution wavelet analysis that provides a tool for analyzing spatial details (e.g., horizontal and vertical edges) of moving objects contained in a sequence of images. Current multiresolution wavelet analysis theory is modified to create an orthonormal wavelet basis for L2(R3) by forming the tensor product of three non- identical multiresolution wavelet analyses on L2(R). An unconventional multiresolution decomposition and reconstruction algorithm is presented which provides a new tool for analyzing moving objects in a scene. Preliminary results demonstrate the new analysis technique''s potential for segmenting key characteristics of an object moving against stationary or moving backgrounds.© (1992) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.

This paper describes the NeuralGraphics software environment used to run interactive neural network training experiments. The NeuralGraphics environment is a collection of software tools, graphical displays, and demonstrations that allow users to easily adapt many of the current neural network paradigms to their particular classification problem. The paper discusses the paradigms implemented in the NeuralGraphics environment as well as the data files required to train and test the learning capability of selected neural networks.

The main thrust of this paper is to encourage the use of neural networks to process raw data for subsequent classification. This article addresses neural network techniques for processing raw pixel information. For this paper the definition of neural networks includes the conventional artificial neural networks such as the multilayer perceptrons and also biologically inspired processing techniques. Previously, we have successfully used the biologically inspired Gabor transform to process raw pixel information and segment images. In this paper we extend those ideas to both segment and track objects in multiframe sequences. It is also desirable for the neural network processing data to learn features for subsequent recognition. A common first step for processing raw data is to transform the data and use the transform coefficients as features for recognition. For example, handwritten English characters become linearly separable in the feature space of the low frequency Fourier coefficients. Much of human visual perception can be modelled by assuming low frequency Fourier as the feature space used by the human visual system. The optimum linear transform, with respect to reconstruction, is the Karhunen-Loeve transform (KLT). It has been shown that some neural network architectures can compute approximations to the KLT. The KLT coefficients can be used for recognition as well as for compression. We tested the use of the KLT on the problem of interfacing a nonverbal patient to a computer. The KLT uses an optimal basis set for object reconstruction. For object recognition, the KLT may not be optimal.