This paper details the use of a genetic algorithm (GA) as a method to preselect spectral feature variables for
chemometric algorithms, using spectroscopic data gathered on explosive threat targets. The GA was applied to laserinduced
breakdown spectroscopy (LIBS) and ultraviolet Raman spectroscopy (UVRS) data, in which the spectra
consisted of approximately 10000 and 1000 distinct spectral values, respectively. The GA-selected variables were
examined using two chemometric techniques: multi-class linear discriminant analysis (LDA) and support vector
machines (SVM), and the performance from LDA and SVM was fed back to the GA through a fitness function
evaluation. In each case, an optimal selection of features was achieved within 20 generations of the GA, with few
improvements thereafter. The GA selected chemically significant signatures, such as oxygen and hydron peaks from
LIBS spectra and characteristic Raman shifts for AN, TNT, and PETN. Successes documented herein suggest that this
GA approach could be useful in analyzing spectroscopic data in complex environments, where the discriminating
features of desired targets are not yet fully understood.
The detection, tracking, and classification of humans in video imagery is of obvious military and civilian importance.
The problem is difficult under the best of circumstances. In infrared (IR) imagery, or any grayscale imagery, the problem
is compounded by the lack of color cues. Sometimes, human detection in IR imagery can take advantage of the thermal
difference between humans and background-but this difference is not robust. Varying environmental conditions
regularly degrade the thermal contrast between humans and background. In difficult data, humans can be effectively
camouflaged by their environment and standard feature detectors are unreliable. The research described here uses a
hybrid approach toward human detection, tracking, and classification. The first is a feature-based correlated body parts
detector. The second is a pseudo-Hough transform applied to the edge images of the video sequence. The third relies on
an optical flow-based vector field transformation of the video sequence. This vector field permits a multidimensional
application of the feature detectors initiated in the previous two methods. Then a multi-dimensional oriented Haar
transform is applied to the vector field to further characterize potential detections. This transform also shows potential
for distinguishing human behavior.
In this paper, we investigate sensor fusion along three avenues: statistical, biological, and categorical. The first two
approaches are analyzed simultaneously to provide a precise and rigorous sensor fusion methodology. The statistical
model currently enhances Bayesian methods for tracking, and suggests further application to target identification and
fusion - involving both low level feature extraction and higher level sensor output combination. The biological model is
also applied to multiple levels of the fusion problem. On the lowest level, it utilizes biologically-inspired results for
improved feature extraction. On the higher levels, it develops biologically-inspired evolutionary and agency algorithms
for sensor output combination and sensor network analysis. Ultimately, we model the entire fusion process with category
theory. Category theory allows for the application of advanced mathematical theory to fusion analysis. In addition to
using category theory as a modeling tool, in this paper we adapt categorical logic via topos theory to provide an
advanced framework for decision fusion - initially using the topos of graphs. Graphs are a simpler representation. We
suggest formulations which will be richer - toward the goal of a theoretically robust and computationally practical sensor
fusion system for assisted/automatic target recognition.
The importance of network science to the present and future military is unquestioned. Networks of some type pervade every aspect of military operations-a situation that is shared by civilian society. However, several aspects of militarily oriented network science must be considered unique or given significantly greater emphasis than their civilian counterparts. Military, especially battlespace, networks must be mobile and robust. They must utilize diverse sensors
moving in and out of the network. They must be able to survive various modes of attack and the destruction of large segments of their structure. Nodes often must pass on classifications made locally while other nodes must serve as combined sensor/classifiers or information coordinators. They must be capable of forming fluidly and in an ad hoc manner.
In this paper, it will be shown how category theory, higher category theory, and topos theory provide just the model required by military network science. Category theory is a well-developed mathematical field that views mathematical structures abstractly, often revealing previously unnoticed correspondences. It has been used in database and software modeling, and in sensor and data fusion. It provides an advantage over other modeling formalisms both in its generality and in its extensive theory. Higher category theory extends the insights of category theory into higher dimensions, enhancing robustness. Topos theory was developed, in part, through the application of category theory to logic, but it also has geometric aspects. The motivation behind including topos theory in network science is the idea that a mathematical theory fundamental to geometry and logic should be applicable to the study of systems of spatially distributed information and analysis flow. The structures presented in this paper will have profound and far-reaching applications to military networks.
This paper describes a method for developing and training a classifier for detecting military vehicles in FLIR (Forward Looking Infrared) imagery. Often image analysis is done via constructing feature vectors from the original two-dimensional image. In this effort, a genetic algorithm is used to evolve a group of linear filters for constructing these feature vectors. Training is performed on collections of target chips and non-target or clutter chips drawn from FLIR image datasets. The evolved filters produce multi-dimensional feature vectors from each sample. First the fitness function for the genetic algorithm rewards maximal separation of target from non-target vectors measured by clustering the two sets and applying a vector space norm. Next, the entire method is adapted to supply feature vectors to a support vector machine classifier (SVM) in order to optimize the SVM's performance, i.e. the genetic algorithm's fitness function rewards effective SVM class distinction. Finally, supplemental features are incorporated into the system, resulting in an improved, hybrid classifier. This classification method is intended to be applicable to a wide variety of target-sensor scenarios.
The acquisition a robust set of IR imagery is frequently impossible through the traditional image collection process. On the other hand, the use of a full-scale simulation is too time consuming and frequently produces unrealistic images. Therefore, other methods are sought that would exploit a small subset of sample real-world images for rapid database prototyping. This paper presents a fast and simple method of high-resolution target image insertion into a low-resolution image of a terrain. The method exploits a naive physics paradigm. First, a high-resolution target image is diffused using a Gaussian kernel and on-target zooming effect. A target binary mask guides the diffusion process. The diffused image is re-sampled onto a low-resolution target image. Next, a down-sampled target image is inserted into a given terrain image using two target insertion/diffusion processes and additional effects. These diffusion processes eliminate contrasts at the border area of a target and on the terrain/background. Background-to-target diffusion extends the heat of overlapped terrain pixels over a target section. Target-to-background diffusion radiates and overlaps target heat over the border area of the adjacent section of the terrain. Developed processes mirror the physics of heat propagation and diffusion, and apply weighted pixel mixing to eliminate target insertion contrasts. Given a set of high-resolution turntable data and a set of terrain images, a training database can be generated within a short time. The number of parameters controlling the insertion process has been decreased to the minimum and brought into a control panel. Each parameter has understandable physical meaning and has assigned a meaningful range of values.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.