الفهرس 
introductionOnly 14 pages are availabe for public view
 |  | from | 16 |  |  |
| | | from | 16 | | |
Abstract
This dissertation is devoted to applying the spiked random neural network (RNN) model in some signal and image processing problems. Although various studies of applying neural network models have been done by many researchers, there remains significant room for introducing other models that simulate the behavior of the biological neurons more closely. Recently, spiked neural networks have attracted many researchers’ attention due to their intrinsic properties that represent in a great extent the manner in which signals are transmitted in biological neural networks where they often are voltage (action potential) spikes rather than analog levels. However, very few techniques based on spiked neural networks have been applied to real world problems so far. The random neural network, developed by Erol Gelenbe, is a biologically inspired spiked model which differs substantially from existing models such as the multi layer perceptron. The network has a compact closed form solution for network state even in the recurrent case. In this model, contrary to other models, the internal state is a nonnegative integer; it rises or falls depending on the excitatory or inhibitory nature of incoming spikes, and it drops each time the neuron fires, recent work on a variety of significant applications of this model has shown that the random neural network is able to efficiently carry out desirable functions such as optimization, learning, associative memory, pattern recognition, and function approximation. This dissertation covers our research work in several areas related to signal and image processing. Although each of them has its own specific history, methodology, and applica¬tions, they all use the random neural network model as a non-parametric computational tool. In particular, we present: • A random neural network model for mine detection and false alarm fil¬tering. We used some pre-published statistical approaches to pre-process the input data before using the random neural network. It is shown through extensive simula¬tions that the proposed model is very effective in detecting mines and rejecting false alarms. The random neural network filter proved to be a robust non-parametric technique that is based on training on a limited calibration data on a certain mine¬field.