Paul Kisatsky

Robust automatic target recognition using a localized boundary representation

Abstract A localized classification approach which includes a localized target boundary representation, a set of local features to characterize parts of a target boundary, and a feature matching method is developed to classify highly degraded targets. Each pixel on the target boundary is represented by the perpendicular Euclidean distance between the boundary pixel and the chord connecting the end-points of a window centered on the boundary pixel. The resulting localized representation is quite robust with respect to noise and missing segments in the targets. A local feature is the segment of the localized representation between two points selected on the target boundary. The two points on the boundary may be specified or randomly selected. The reference features of a hypothesized target class are compared with segments of the localized representation of the test target to determine a dissimilarity measure between the hypothesized target class and the test target. Segments of the test target may also be matched with the reference localized contour sequences of the targets. The test target is assigned to a target class using a minimum mismatch rule. Results from a series of experiments conducted on three targets show that it is possible to classify targets experiencing high levels of noise and high percentages of missing segments in partial targets.

Optical and electrical effects during polymerization and depolymerization in liquid sulfur: indications for the nonuniformity model for covalent liquids

The polymerization and depolymerization of liquid sulfur refers, respectively, to the processes of chain growth and chain scission promoted by changing thermodynamic conditions. These structural transitions cause major optical and electrical consequences, including: (1) color change from yellow to orange to red, (2) peaked high angle and low angle scattering and minimum in transmission of He-Ne laser light at the structural changes, (3) a decrease in the current induced by a pulsed ruby laser above the polymerization temperature T(p) compared with below T(p), (4) two changes in the sign of the temperature coefficient of resistance in the vicinity of T(p) and the depolymerization temperature T(d), and (5) superlinear current-voltage characteristics above T(d). The data are interpreted as being suggestive of the structure of a liquid being better described by a nonuniformity rather than random network model.

A DEMONSTRATION OF LASER PUMPING USING A COMPRESSED GAS LIGHT SOURCE

Xenon gas has been compressively heated to temperatures of from 8000 to 15,000°K in a ballistic compressor system. Brightness measurements of the light pulses produced indicate blackbody radiation at these temperatures. The radiation has been coupled into a neodymium‐doped glass laser rod with resulting laser action. The system affords a direct transfer of mechanical energy into pumping radiation without the necessity of intermediate energy storage devices.

Three-layer perceptron based classifiers for the partial shape classification problem

Abstract The question of classification robustness in the multi-network neural network based system for the partial shape classification problem is addressed. In order to increase the robustness in classification, an extension of the multi-network system and a new single network system are proposed. The extension increases the robustness by augmenting the training of the three-layer perceptrons in the system. The three-layer perceptron in the single network system is designed to detect the features in all of the pattern classes. In the test mode, the test pattern is hypothesized to belong to the pattern classes and the network response to the test pattern is used to determine the similarity scores for the hypothesized classes. Two partial shape classification experiments are designed to compare the performance of the original multinetwork system, the augmented training approach, and the single network system on exactly the same test set. The results indicate that there is a significant increase in the classification robustness in the proposed augmented training approach and the single network system.

Prototype selection rules for neural network training

Abstract Rules to select a set of training prototypes from a collection of training prototypes are developed so that a neural network classifier converges to a solution when pattern classes overlap in feature space. The rules are especially useful for selecting training prototypes in order to improve the network robustness and operational flexibility by retraining the network with noisy prototypes. The formulation of the selection rules are based on a distortion measure and the network response to the training prototype collection. The application and effectiveness of the selection rules are demonstrated on a synthetic pattern classification in a Gaussian noise problem and a practical automatic target recognition problem.

Performance of a ballistic compressor as a high-intensity pulsed light source.

Xenon gas has been compressively heated from 8000 K to 15 000 K at peak pressures of 800-2400 atm. Brightness measurements were made of the emitted radiation at selected wavelengths in the visible. In a few cases, a time-resolved spectrum was obtained. The results obtained from a theoretical analysis are shown to be in reasonable agreement with experimentally measured values. A high efficiency for conversion of mechanical energy to radiant energy is indicated and suggests the use of such a device as an efficient source for laser pumping.

Simulation of parafoil reconnaissance imagery

Reconnaissance from unmanned platforms is currently of interest to DoD and civil sectors concerned with drug trafficking and illegal immigration. Platforms employed vary from motorized aircraft to tethered balloons. One appraoch currently under evaluation deploys a TV camera suspended from a parafoil delivered to the area of interest by a cannon launched projectile. Imagery is then transmitted to a remote monitor for processing and interpretation. This paper presents results of imagery obtained from simulated parafoil flights in which software techniques were developed to process-in image degradation caused by atmospheric obscurants and perturbations in the normal parafoil flight trajectory induced by wind gusts. The approach to capturing continuous motion imagery from captive flight test recordings, the introduction of simulated effects, and the transfer of the processed imagery back to video tape is described.

Prototype selection rule for neural network training

Rules to select a set of training prototypes from a collection of training prototypes are developed so that a neural network classifier converges to a solution when pattern classes overlap in feature space. The formulation of the selection rules are based on distortion measure and the network response to the training prototype collection. The rules are also especially useful for selecting training prototypes in order to improve the network robustness and operational flexibility by retraining the network with noisy prototypes. The application and effectiveness of the selection rules are demonstrated on a synthetic pattern classification in Gaussian noise problem and a practical automatic target recognition problem.