Stephen Wayne Soliday (Selected Publications)

Hybrid Fuzzy-Neural Classifier for Feature Level Data Fusion in LADAR Autonomous Target Recognition

(Down Load PDF/Postscript)

This paper will discuss the design of a hybrid fuzzy-neural classifier for fusion of range and intensity channels coming from a LADAR sensor. Fusion was performed on a feature rather than pixel level. Results will be compared between ATR performance with and with out fusion. Also, discussed in this paper is the use of genetic algorithms for the training and optimization of the ATR system with a limited set of ground truth.

Soliday, Stephen W. and Perona, Melissa T. and McCauley, Daniel G., Proceedings for Automatic Target Recognition XI, Jul 2001, ISBN 0-8194-4074-4

A Genetic Algorithm Model for Mission Planning and Dynamic Resource Allocation of Airborne Sensors

(Down Load PDF/Postscript)

Genetic Algorithms (GA) have been very successful in combinatorially--explosive problems such as; Job Shop Scheduling, Non-linear Transportation Model, and the Traveling Salesman Problem. The problem of mission planning for large numbers of airborne platforms contains elements of all three of these classic GA problems. This paper will outline the application of genetic algorithms to mission planning and dynamic allocation of airborne sensors. An experiment will consist of a mission simulation containing (n) airborne sensors in different locations and states of readiness, and (m) requests for imagery with varying mission priorities. The GA will match sensors with requests in order to minimize the cost and time--line and maximize the execution of high priority requests. Results of simulation will be discussed.

Soliday, Stephen W., National Symposium on Sensor and Data Fusion, Applied Physics Lab., Laural, MD, May 1999

Genetic Algorithm Approach to the Search for Golomb Rulers

 (Down Load Postscript)

The success of genetic algorithm in finding relatively good solutions to NP-complete problems such as the traveling salesman problem and job-shop scheduling problem provided a good starting point for a machine intelligent method of finding Golomb Rulers. These rulers have been applied to radio astronomy, X-ray crystallography, circuit layout and geographical mapping. Currently the shortest lengths of the first sixteen rulers are known. The nature of NP-complete makes the search for higher order rulers difficult and very time consuming. While the shortest lengths for each order are important as a mathematical exercise, finding relatively short high order valid rulers has a more important impact on real world applications. Genetic algorithm has shown good results in finding usable Golomb Rulers in minutes or hours instead of weeks or months.

Soliday, Stephen W. and Homaifar, Abdollah and Lebby, Gary L., Genetic Alogorithms: Proceedings of the Sixth International Conference on Genetic Algorithms, University of Pittsburgh PA, vol: 1, pp 528-535, Morgan Kaufman, July 1995, ISBN: 1-55860-370-0

Fuzzy Controller for Two Wheel Independent Drive Pivot Steering Vehicle

 (Down Load Postscript)

A description for a vehicle consisting of two independently driven drive wheels and one free wheeling castor is given. The equations of motion are derived using Lagrange's equations and a model is developed using Runge-Kutta numerical integration. A fuzzy controller is then developed to control both the speed and direction of the vehicle. The parameters of the discourse sets for the fuzzy controller were optimized using a primitive form of Genetic Algorithm. Results of the controller's performance are discussed.

Soliday, Stephen W., First Industry/Academy Symposium on Research for Future Supersonic and Hypersonic Vehicles, vol: 1, pp 106-111, TSI Press Series, Dec 1994, ISBN: 0-9627451-8-9

Programmable Transfer Functions for Neural Networks

 (Down Load Postscript)

Implementing neural networks in the computational environment of computing requires a careful analysis of loop optimization. The transfer function of the network's processing elements, and their associated derivatives, should be designed in such a way that it is both fully programmable and contain as few in-loop floating point operations as possible. The purpose of this discussion is to analyze some of the well behaved transfer functions traditionally used in fully interconnected networks, and to derive a faster transfer function that meets the above requirements.

Soliday, Stephen W., First Industry/Academy Symposium on Research for Future Supersonic and Hypersonic Vehicles, vol: 1, pp 142-147, TSI Press Series, Dec 1994, ISBN: 0-9627451-8-9