Novel Neural Network Based On Memristor And Its Application In Image Processing

With the rapidly increasing requirements of information processing,a more intelligent and miniaturized information processing system is needed.So neuromorphic systems have been attracted greatly.The nervous system research is seriously limited by the size of semiconductor transistors.Memristor with advantages such as brain-like memory function,nanometer size and nonvolatile storage is expected to completely change the existing information processing methods.In this work,we apply the memristor to neural network,and propose a new memristor-neural network which can reduce the circuit complexity and integrate easily.Compared with the traditional neuron circuit,it has a simpler circuit structure and a strong potential to reduce energy consumption.The memristor,which is considered as natural electronic synapse,can be well applied in bionic system and make the memristive neural network more intelligent.In this dissertation,the characteristics of the memristor are deeply studied,based which the novel memristor-based neural network are proposed.At the same time,this dissertation also discusses the memristor-based logic circuit,and designs the memristive digital logic device.A novel memristor-discrete Hopfield neural network based on the memristor and neu MOS transistor is proposed,which is applied to color digital image restoration.Furthermore,we put forward an unordinary pulse coupled neural network with adaptive parameters and an adaptive image enhancement algorithm.Specifically,this dissertation is divided into four parts,as follows:Firstly of all,this dissertation focuses on the classical HP memristor model and threshold adaptive memristor model.Furthermore the relationship among memristance,charge and magnetic flux has been investigated.The threshold characteristic and synaptic characteristic of the models are studied by SPICE,which provides a good theoretical reference and experimental basis for the subsequent application of memristor.Then,this dissertation designes the pure memristor logic circuit based on the logical computing ability and information storage characteristics of HP memristor.Different from the traditional memristor logic circuit,the designer of this dissertation uses the voltage as the logic value directly which is more intuitive and convenient.Compared with traditional transistor logic circuit,it has obvious improvement in the complexity of the circuit.Based on this,this dissertation constructs a memristor encoder and a memristor decoder,and they are verified by simulation.The scheme is expected to promote the application of memristor in digital circuit,and provides a new way to optimize the logic device.Next,utilizing the weighted sum property and threshold controllable function of the neuron transistor,a neoteric Hopfield neural network is proposed based on the synaptic characteristics of memristor.The network is validated to apply in the associative memory and the color digital image restoration.The network is only composed of neuMOS,memristor and ordinary resistor,and eliminate the need for complex differential operational circuit,current and voltage signal conversion circuit,which are needed in traditional circuit.It is obvious that the circuit is simple and has good circuit compatibility,which is conducive to large scale integration.At the same time,the network also has the advantages of low energy consumption,dynamic threshold control,and programmable weights.The scheme can not only greatly simplify the network structure,but also enhance the network performance,which is helpful to promote the hardware realization of the neuromorphic system.Finally,the M-PCNN neural network is proposed by combining the threshold adaptive memristor and the traditional PCNN model.The model utilizes the output of memristor circuit to analog the connection strength between neurons,which can realize the connection strength adaptive dynamic changes processing with external stimuli.This new neuron model expands the dynamic characteristics of neural network,and provides a new idea for the development of parameter adaptive neural network.Furthermore,an adaptive image enhancement algorithm based on M-PCNN neuron model is proposed,which could stress details and enhance image contrast.From the visual subjective characteristics of human eyes and objective performance evaluation index,the superiority of the proposed algorithm has been demonstrated.This method can highlight the partial details of images,and enhances the constrast of light and shade.Consequently,this work lays a foundation of the application and development of the neural network in image processing.