13699

12450

كارشناسي ارشد

مخابرات

اصفهان: دانشگاه صنعتي اصفهان، دانشكده برق و كامپيوتر

۱۳۹۷

دوازده، ۸۶ص.: مصور، جدول، نمودار

محمدرضا احمدزاده

فرزانه شايق

1397/05/08

كتابنامه

برق و كامپيوتر

مهندسي برق و كامپيوتر

ID12450

Design and Implementation of a spiking neural network with IF neuron model for pattern recognition Parvaneh Rashvand p rashvand@ec iut ac ir June 13 2018 Department of Electrical and Computer Engineering Isfahan University of Technology Isfahan 84156 83111 Iran Degree M Sc Language Farsi Supervisor Prof Mohammadreza Ahmadzadeh Ahmadzadeh@cc iut ac ir Abstract Neurons in the nervous system transmit information through the action potential called spikes Neural coding discussabout how information is transmitted just through spike sequences Thus it is one of the important components in cognitivesystems The way of processing sensory inputs and distinguishing different patterns to provide higher level brain functionssuch as memory storage and retrieval is the main subject of neural coding Accordingly This study describes the operationof biological neural systems and their models of information processing and image recognition based on which spikingneural networks SNNs are developed that are the third generation of artificial neural networks ANNs In contrast to theprevious ANNs SNNs work based on temporal coding approaches In our proposed SNN the number of neurons neuronmodels encoding method and designing the learning algorithm are described in a good way For choosing the number ofinput neurons we tried to decrease the number of them so that the speed of our program is increased The proposed learningalgorithm is one of the important problems that can increase the speed and the accuracy of our network The spiking neuralnetwork presented in this study is based on the Integrate and Fire IF neuron model and uses the time to first spike codingto train the network by a new proposed method Iris and MNIST databases have been used to evaluate the performanceof the proposed network For recognition of Iris database the accuracy of our network was 95 33 with 48 input neuronsand our network is trained in only 57 itterations that shows our network has an excellent convergence rate For MNISTdatabase at first we consider the illuminance of each pixel as entrance neurons in this situation we had 480 input neuronswith 85 5 accuracy Then we used 14 structural features as the inputs in this situation we had only 168 input nuerons with95 accuracy It means we implemented a SNN with a good accuracy and a few number of input neurons Key Words Spiking Neural Networks Training algorithm Neural coding Image Recognition IFneuron model

محمدرضا احمدزاده

فرزانه شايق