A chaotic neural network model based on multi-synaptic memristor

Abstract

The present invention relates to the field of signal processing, and in particular to a chaotic neural network model based on a multi-synaptic memristor. The present invention is based on a multi-neuron continuous HNN, and uses a memristor model for optimization design: on the one hand, the fixed synaptic weight is replaced with a switchable synaptic weight, and a synaptic switching memristor is used to realize chemical synaptic parameters, and nonlinearity is introduced in the synaptic weight; on the other hand, an electromagnetic radiation memristor is used to simulate external stimulation, and the external stimulation is made predictable and controllable while realizing the introduction of external stimulation, and the final chaotic neural network model realizes an excellent hyper-chaotic effect; and the model uses a memristor with good consistency to build a hardware circuit, and with the help of high-speed processing of hardware, a chaotic sequence with large cardinality and good randomness is obtained. Therefore, it can be applied to various signal processing scenarios, such as key generation in chaotic encryption of images, to cope with the ever-improving encryption and decryption technology.

Description

Technical Field

[0001] The present invention relates to the field of signal processing, and in particular to a chaotic neural network model based on a multi-synaptic memristor. Background Art

[0002] Chaotic encryption technology is a technology that uses iterative sequences generated by chaotic systems for encryption conversion. Chaos theory has the characteristics of initial value sensitivity, unpredictability and random noise, which are highly consistent with the needs of confidential communication. The sequences generated by chaotic systems with good randomness and huge key space can meet the requirements of encryption technology, so they have great promotion potential in applications. However, the initial value sensitivity of chaotic systems may also be exploited by attackers. At present, researchers have been able to identify the parameters of some low-dimensional chaotic shift keying encryption systems. In order to improve the performance of chaotic encryption technology, the current mainstream approach is to realize phase trajectory diagrams to present multi-scroll chaotic attractors and hyperchaotic systems with more complex evolution laws. However, most of the chaotic systems designed by pure mathematics are currently in the theoretical research stage, and there are few practical applications.

[0003] Biological neural networks are large and complex nonlinear systems, and the rich interactions between neurons produce diverse dynamic behaviors. Studies have found that there is chaos in the human brain, so many studies have combined artificial neural networks with encryption technology to achieve encryption functions using the chaotic state sequence generated by artificial neural networks. Since artificial neural networks simulate the complexity of real neural networks, most studies use software simulation to model neural networks. Traditional artificial neural networks are usually built using transistors, and their circuits are complex and require large-area circuits and high-power operation to achieve similar functions, which is far from the actual working performance of the human brain.

[0004] Memristors are ideal for building artificial neural synapses due to their memory properties, compatibility with complementary oxide semiconductor (CMOS) processes, simple structure, and small area. At present, the design of chaotic neural networks based on memristors is mainly based on the Hopfield Neural Network (HNN) model, in which the memristor model is integrated for construction. However, the chaotic effects presented in existing chaotic neural network research are not significantly superior to the results obtained using pure mathematical complex chaotic systems, and the chaotic effects produced by chaotic neural networks built using memristors are average. Summary of the invention

[0005] In view of the above-mentioned problems or shortcomings, in order to solve the problem of poor chaotic effect of existing chaotic encryption technology, the present invention provides a chaotic neural network model based on multi-synaptic memristor. The present invention is based on a continuous HNN of multiple neurons, uses a memristor model for optimization design, builds a chaotic neural network model based on multi-synaptic memristor, and achieves a hyper-chaotic effect; and uses devices with good consistency to build hardware circuits, and uses high-speed processing of hardware to obtain a chaotic sequence with a large cardinality and good randomness, so that it can be applied to various signal processing scenarios, such as key generation in chaotic encryption of images, to cope with the ever-improving encryption and decryption technology.

[0006] The technical solution adopted by the present invention is as follows:

[0007] A chaotic neural network model based on multi-synaptic memristor includes a continuous HNN model of multiple neurons, synaptic switching and external stimulation.

[0008] The multi-neuron continuous HNN model is the network basis and includes m neurons, where m≥3.

[0009] The present invention is optimized based on the HNN model. The HNN model naturally has chaotic phenomena and is optimized to enhance its chaotic effect. The HNN model usually uses multiple neurons. In the selection of the number of neurons in the model, a single neuron has no synaptic influence between neurons, and two neurons only have synaptic influence between two neurons, and the chaotic effect is poor. Three neurons are sufficient to represent the mutual influence between neurons, and compared with the HNN model with more neurons, the model is simpler and easier to analyze and understand. The iterative equation of the neuron state is shown in Formula 1:

[0010]

[0011] Among them, x j represents the internal state voltage of the jth neuron, t represents time, and w ij represents the synaptic weight of the ith neuron to the jth neuron, tanh(x j ) represents the independent variable x j The hyperbolic tangent function, I j Represents the external interference to the j-th neuron, and the default value is zero.

[0012] The synaptic switching is to change at least one original fixed synaptic weight in the HNN model into a switchable synaptic weight; the switchable synaptic weight is used to control the state of the chaotic system to generate a complex multi-vortex attractor, thereby enhancing the chaotic effect.

[0013] Switchable synaptic weights refer to the synaptic weights between two neurons in the HNN model that switch between three values: zero, fixed weights, and variable weights; zero value means no synapse; fixed weights correspond to electrical synapses, which means that a fixed influence is exerted when the input stimulus exceeds the threshold, and large-dose stimulus switches to electrical synapses; variable weights correspond to chemical synapses, which are floating weights, and small-dose stimulus switches to chemical synapses, which are affected by input stimulus; variable weights are implemented using memristors, and the output voltage of a neuron acts on the synaptic switching memristor to generate a variable memristor conductance as a variable weight, which acts on another neuron.

[0014] When designing and constructing chaotic systems, the control of parameter switching can be used to switch system functions, making chaotic mapping more diverse, rich, random, and disordered. From a neurological perspective, the idea of ​​switching is also feasible. The interaction between neurons can be achieved through electrical synapses, chemical synapses, and other methods, and these two types of synapses usually exist at the same time. The switching of synapses can be achieved by the strength of the stimulus. A small dose of stimulus selects the chemical synapse, and a large dose of stimulus switches to the electrical synapse.

[0015] The state of neurons in the HNN model with weight switching is currently only affected by the initial internal state of neurons and the interaction between neurons, while adding external interference to neurons can have a more complex impact on the dynamic behavior of the neural network.

[0016] The external stimulus is to generate a current I by applying a neuron state voltage to the electromagnetic radiation memristor. j , thereby reacting to the neurons to simulate controllable external disturbances, that is, applying external disturbances to neurons affects the dynamic behavior of neurons.

[0017] Furthermore, the synaptic switching is affected by the state of the synapse itself and the external stimulation received by the two neurons connected by the synapse, specifically:

[0018] When the synapse is closed, the weight value is zero; when the synapse is open, if the stimulus applied to the neuron is small, it switches to the chemical synapse weight; if the stimulus applied to the neuron is large, it switches to the electrical synapse weight; the synapse switching expression is:

[0019] w ij =wγ (2)

[0020] w ij It is obtained by multiplying the weight value w and the weight coefficient γ, where the weight value w switches between the weight value ω0 corresponding to the electrical synapse and the weight value ω1 corresponding to the chemical synapse, see formula 3;

[0021]

[0022] The switching condition is determined by the external stimulus I of the i-th neuroni and the external stimulus I of the jth neuron j The threshold of the external stimulus difference is C1. If it is higher than the threshold, it switches to ω0, and if it is lower than the threshold, it switches to ω1. ω1 is numerically equal to the memristor value of the synaptic switching memristor.

[0023] The weight coefficient γ switches between 0 and 1, indicating whether the synapse is enabled, see formula 4;

[0024]

[0025] The switching condition is determined by the external stimulus I of the i-th neuron i and the external stimulus I of the jth neuron j The difference between them and the synaptic state state between neurons i and j. The synaptic state state has a low active state l and a high active state h. When the synaptic state is low active and the stimulus difference is less than the threshold, the synapse is closed and the synaptic weight is zero; under other conditions, the synapse is open and w ij =w, the synaptic weight switches between ω0 and ω1.

[0026] Furthermore, the external stimulus generates a current I by applying a neuron state voltage to the electromagnetic radiation memristor. j , expressed as:

[0027]

[0028] Among them, k represents the influence coefficient, which is a constant; represents the electromagnetic radiation memristor flux acting on the jth neuron; represents the memristor value of the electromagnetic radiation memristor acting on the jth neuron; x j Represents the internal state voltage of the jth neuron.

[0029] In the selection of external interference, the present invention uses neuron voltage to flow through a memristor to generate current, and the reasons are summarized as follows. First, using current input to simulate external stimulation can directly use electrical physical quantities for modeling and introduce stimulation; second, using a memristor can introduce nonlinearity, simulating a more realistic environment while also introducing a stronger chaotic effect in a strict sense; third, the stimulation introduced by the memristor is controllable. Although iterating for different network initial values ​​can produce completely different stimulation effects, for a certain network initial value, the stimulation generated during the iteration process is unique and will not change due to the influence of usage time.

[0030] The present invention introduces a memristor, which not only realizes more complex dynamic behaviors by adding nonlinearity, but also provides a device foundation for subsequent hardware implementation. The application of memristors mainly has two aspects: on the one hand, it is used to replace chemical synaptic parameters in synaptic switching and introduce nonlinearity in synaptic weights; on the other hand, it simulates external stimuli, making the external stimuli predictable and controllable while realizing the introduction of external stimuli. This model meets the requirements of the coexistence of randomness and determinism of chaotic systems. The chaotic sequence generated by the network is uniquely determined by the network initial value and the value accuracy. The specific value of the obtained chaotic sequence is related to the value method.

[0031] Furthermore, the above chaotic neural network model based on multi-synaptic memristor is used as the key generation part in the symmetric encryption method. After iterating with a given initial value of the neural network, the obtained chaotic sequence is used as the key and combined with the encryption algorithm to realize the chaotic encryption function.

[0032] Furthermore, the chaotic encryption completes the image encryption function based on the encryption ideas of confusion and diffusion;

[0033] Key generation part: For the plaintext image, partition calculation is performed on it to generate the corresponding network initial value, which is the key used by the encryption party for confidential transmission; then the network initial value is used for iteration to generate a chaotic sequence, and processed into a key matrix, which is the key used by the encryption party and the decryption party in the encryption algorithm part.

[0034] Encryption algorithm part: The single-channel image data is regarded as a two-dimensional matrix, and pixel confusion is achieved through data row shift and scrambling transformation operations. Then, the data bit XOR is performed with the key matrix generated by the chaotic sequence to obtain the encrypted image.

[0035] The decryption operation is realized through the corresponding symmetric operation. After the above operation, the image encryption function with high efficiency and security can be realized.

[0036] In summary, the present invention is based on a multi-neuron continuous HNN and uses a memristor model for optimization design: on the one hand, the fixed synaptic weights are replaced with switchable synaptic weights, and the chemical synaptic parameters are realized by using synaptic switching memristors, and nonlinearity is introduced in the synaptic weights; on the other hand, electromagnetic radiation memristors are used to simulate external stimuli, and the external stimuli are made predictable and controllable while introducing external stimuli, and the final chaotic neural network model achieves excellent hyper-chaotic effects; and the model uses memristors with good consistency to build hardware circuits, and with the help of high-speed processing of hardware, a chaotic sequence with large cardinality and good randomness is obtained. Therefore, it can be applied to various signal processing scenarios, such as key generation in chaotic encryption of images, to cope with the ever-improving encryption and decryption technologies. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1Schematic diagram of a chaotic neural network model based on a multi-synaptic memristor according to an embodiment.

[0038] Figure 2 The phase trajectory diagram of the neural network iteration of the embodiment and the phase trajectory diagram of the neural network iteration using only external stimulation.

[0039] Figure 3 The figure is an architectural block diagram of an embodiment applied to an image encryption hardware system.

[0040] Figure 4 The following is a schematic diagram of the working process of the encryption hardware system of the embodiment.

[0041] Figure 5 This is a schematic diagram of the working data flow of the image encryption and decryption module. DETAILED DESCRIPTION

[0042] The present invention is described in detail below with reference to the accompanying drawings and formulas.

[0043] The present invention is a chaotic neural network model based on a multi-synaptic memristor. The specific implementation method and hardware implementation are introduced below. In this embodiment, a three-neuron continuous HNN model is preferably used as the network basis.

[0044] Figure 1 Schematic diagram of a three-neuron memristor neural network model of an embodiment. The innovation is the use of four memristors, of which three memristors used to simulate external stimulation act on three neurons respectively; and another memristor is used to replace one of the synaptic weights.

[0045] Where x1, x2, x3 represent the internal voltages of the three neurons, ω ij represents the influence weight of the ith neuron on the jth neuron. The state change of a neuron is affected by its own current state, the current state of surrounding neurons, and the external input current. The external input current is input using an electromagnetic radiation memristor, ω 21 The switching is achieved by switching between the fixed value ω0, the memristor's memristor corresponding weight ω1 and the fixed value 0.

[0046] The expression of the memristor chaotic neural network model established based on the HNN in the continuous time domain of three neurons is shown in Formula 6.

[0047] dx / dt=-x+Wtanh(x)+I M (6)

[0048] Where t is the system iteration time, and x is a vector containing three neuron state voltages x1, x2, and x3, see Formula 7.

[0049]

[0050] The interneuron synaptic weight matrix W in Formula 6 is a 3×3 matrix, see Formula 8. γw in Formula 8 represents the result of multiplying the weight value w and the weight coefficient γ, that is, the synaptic weight between the first and second neurons. tanh(x) is the hyperbolic tangent function.

[0051]

[0052] The interference current I generated by the memristor in Formula 6 M The expression is shown in Formula 9. The current expression involves the use of electromagnetic radiation memristor. represents the memristor value of the electromagnetic radiation memristor acting on the i-th neuron, represents the flux of the electromagnetic radiation memristor acting on the i-th neuron, the electromagnetic radiation memristor flux Flux of the memristor used for external stimulation of three neurons See formula 10. The iteration of is related to the state voltages x1, x2, x3 of the corresponding neurons, see formula 11.

[0053]

[0054] The synaptic weight γw between the first and second neurons in formula 8 switches between three values, one of which is a variable weight, which is calculated by the flux of the synaptic weight memristor. The iterative function is shown in Formula 12. The flux of the memristor is Its memetic derivative is numerically equal, so it is directly used in formula 13 As a memetic derivative.

[0055]

[0056] The synaptic weight γw between the first and second neurons is obtained by multiplying the weight value w and the weight coefficient γ. Figure 1 In the expression ω 21 . The weight value w is affected by the interference current. If it is above the threshold, it is a fixed value, representing an electrical synapse; if it is below the threshold, it is a weight memristor memristor value, representing a chemical synapse. The weight value w is shown in Formula 13. The weight coefficient γ is controlled by the random influence introduced by the interference current and the function value xlogistics(n) of the logic evolution algorithm (Formula 15). The weight coefficient γ is shown in Formula 14. Formula 15 represents the iterative formula of the logic mapping, f logistics (n) represents the function value when the number of iterations is n, r is a control parameter, and its value range is [3.57,4]. Only when the value is within this range can the logical mapping be kept in a chaotic state.

[0057]

[0058] f logistics (n+1)=r×f logistics (n)(1-f logistics (n)) (15)

[0059] Figure 2 This is a comparison of the phase trajectory diagram of the neural network iteration combined with synaptic switching in this embodiment and the phase trajectory diagram of the neural network iteration using only external stimulation. The results in the figure can intuitively reflect the importance of synaptic switching. The following values ​​of the parameters can present a better chaotic effect. When the parameters ω0 = 0.7, constant1 = 3.5, d = 0.5, x logistics When (0) = 0.1 and r = 4, the chaotic effect of the network is the best, and the phase trajectory diagram shows a multi-vortex attractor, which can produce a good chaotic sequence. In addition, the Lyapunov index spectrum is analyzed, and the results show that there are two indexes greater than 0, which are 0.10869 and 0.02154, respectively, indicating that the network is a hyperchaotic system. The high speed and stability of hardware can greatly improve the efficiency of encryption. The following will show the implementation plan of using this network for encryption on hardware.

[0060] like Figure 3 The figure shows the system architecture block diagram. The host computer communicates with the system controller module through the SPI bus to transmit and receive image information; the system controller module is the AMBA bus host and communicates with various slave modules through the AMBA bus; the memristor neural network iteration module communicates with the memristor control circuit module through the IO port.

[0061] The hardware implementation example is as follows: The entire system includes a system controller module, a memristor neural network iteration module, a memristor control circuit module, an image information storage module, an image encryption and decryption module, a bus communication module and a display module.

[0062] The system controller module mainly receives host computer control information through the bus, and sends control information and receives data information to the memristor neural network iteration module, image information storage module, image encryption and decryption module and display module, so as to realize state control such as host computer communication, image information storage, chaotic sequence generation, image encryption and decryption, etc.

[0063] The memristor neural network iteration module mainly receives bus information from the system controller, sends control information to the memristor control circuit module and receives data information, and iteratively generates the chaotic random sequence required by the system through the built-in neural network digital circuit.

[0064] The memristor control circuit module is used to realize the rewriting of the memristor memristor value and the reading of the memristor memristor value.

[0065] The image information storage module is used for temporary storage of image information and can be implemented using a memory such as RAM, FLASH, etc.

[0066] The image encryption and decryption module is used to implement a specific chaotic encryption algorithm using the random sequence key generated by the memristor neural network iteration module, thereby completing the encryption and decryption of image information.

[0067] The bus communication module is used for transmitting address information, control information and data information between the host computer and the system controller, and between the system controller and other modules.

[0068] The display module is used for displaying specific images and chaotic sequences, and can be a display, an oscilloscope, a logic analyzer, and other components.

[0069] Furthermore, the memristor neural network iteration module includes a bus information analysis module, a single-precision floating-point data operation module and a memristor control information generation module.

[0070] The bus information analysis module is used to analyze the control information from the AMBA bus and send the random sequence generated by the module back to the system controller module through the AMBA bus.

[0071] The single-precision floating-point data operation module mainly performs iterative operations on the digital part of the chaotic neural network, and together with the memristor control circuit module completes the generation of the chaotic sequence.

[0072] The memristor control information generation module mainly generates write voltage and read voltage digital signals for the memristor, and communicates with the memristor control circuit module to rewrite the memristor's memristor value and read out the required memristor's memristor value.

[0073] The memristor control circuit module includes a memristor, a lookup table LUT, an ADC and a DAC.

[0074] The ADC module is used to convert the write voltage or read voltage digital signal from the memristor neural network iteration module into an analog signal and apply it to the memristor. The DAC module is used to convert the current signal generated by the memristor after the voltage is applied into a digital signal for subsequent use. The lookup table LUT is used to obtain the memristor's memristor value based on the corresponding voltage and current signals and return it to the memristor neural network iteration module.

[0075] like Figure 4The flowchart of the system working in the embodiment is shown. First, the host computer transmits image information to the system controller through the SPI bus; the system controller receives the image information and temporarily stores it in the image information memory module; the system controller configures the register group of the memristor neural network iteration module through the AMBA bus; the memristor neural network iteration module starts the neural network iteration according to the register information; the memristor neural network iteration module communicates with the memristor control circuit module to obtain the corresponding memristor value; after the iteration is completed, the memristor neural network iteration module returns the random sequence to the system controller through the AMBA bus; the system controller generates a random key, and takes out the image information and transmits it to the image encryption and decryption module; the image encryption and decryption module encrypts or decrypts the image, and the processed information is stored in the image information memory module; the system controller takes out the image information from the image information memory module and transmits it back to the host computer through the SPI bus; the system controller transmits information to the display module through the AMBA bus as needed to realize the display function.

[0076] like Figure 5 The figure shows the working data flow of the image encryption and decryption module, where the red arrow indicates the encryption process and the blue arrow indicates the decryption process. There are several steps for the encryption part: Step 1, the system controller inputs the image data and the key into the image encryption and decryption module through the AMBA bus and stores them in RAM0; Steps 2 and 3, the shiftrow module takes the image data out of RAM0 and performs row shift processing, and saves it in RAM1 according to the new storage position; Steps 4 and 5, the arnold_scramble module takes the data out of RAM1 and performs Arnold transformation processing, and stores it back to RAM0; Steps 6 and 7, the data_xor module reads the image data and the key from RAM0, performs XOR, and then stores it back to RAM0; Step 8, takes the image data in RAM0 and transmits it back to the system controller through the AMBA bus.

[0077] It can be seen from the above embodiments that the present invention is based on a multi-neuron continuous HNN and uses a memristor model for optimization design: on the one hand, the fixed synaptic weights are replaced with switchable synaptic weights, and the chemical synaptic parameters are realized by using synaptic switching memristors, and nonlinearity is introduced in the synaptic weights; on the other hand, electromagnetic radiation memristors are used to simulate external stimuli, and the external stimuli are made predictable and controllable while introducing external stimuli, and the final chaotic neural network model achieves an excellent hyper-chaotic effect; and the model uses memristors with good consistency to build hardware circuits, and with the help of high-speed processing of hardware, a chaotic sequence with large cardinality and good randomness is obtained. Thus, the key generation applied to chaotic encryption of images is realized to cope with the ever-improving encryption and decryption technology.

Claims

1. A chaotic neural network model based on multi-synaptic memristor, characterized by: Includes continuous HNN models of multiple neurons, synaptic switching, and external stimulation; The multi-neuron continuous HNN model is the network basis, comprising m neurons, m≥3; The iterative equation of the neuron state is: Among them, x j represents the internal state voltage of the jth neuron, t represents time, and w ij represents the synaptic weight of the ith neuron to the jth neuron, tanh(x j ) represents the independent variable x j The hyperbolic tangent function, I j Represents the external interference to the jth neuron, the default value is zero; The synaptic switching is to change at least one original fixed synaptic weight in the HNN model into a switchable synaptic weight; the switchable synaptic weight is used to control the state of the chaotic system to generate a complex multi-vortex attractor, thereby enhancing the chaotic effect; Switchable synaptic weights refer to the synaptic weights between two neurons in the HNN model switching between three values: zero, fixed weights, and variable weights. Zero means no synapses. Fixed weights correspond to electrical synapses, which means that a fixed influence is applied when the input stimulus exceeds the threshold, and large-dose stimulus switches to electrical synapses. Variable weights correspond to chemical synapses, which are floating weights, and small-dose stimulus switches to chemical synapses, which are affected by input stimulus. Variable weights are implemented using memristors, and the output voltage of a neuron acts on the synaptic switching memristor to generate a variable memristor conductance as a variable weight, which acts on another neuron. The external stimulus is to generate a current I by applying a neuron state voltage to the electromagnetic radiation memristor. j , thereby reacting to the neurons to simulate controllable external disturbances.

2. The chaotic neural network model based on multi-synaptic memristor as claimed in claim 1, characterized in that: The synaptic switching is affected by the state of the synapse itself and the external stimulation to the two neurons connected by the synapse, specifically: When the synapse is closed, the weight value is zero; when the synapse is open, it is compared with the threshold C1 of the difference in external stimulation applied to the two neurons: if the difference in stimulation applied to the neuron is less than the threshold C1, it is switched to the chemical synapse weight; if the difference in stimulation applied to the neuron is greater than or equal to the threshold C1, it is switched to the electrical synapse weight; the synapse switching expression is: w ij =wγ (2) w ij It is obtained by multiplying the weight value w and the weight coefficient γ, where the weight value w switches between the weight value ω0 corresponding to the electrical synapse and the weight value ω1 corresponding to the chemical synapse, see formula 3; The switching condition is determined by the external stimulus I of the i-th neuron i and the external stimulus I of the jth neuron j The threshold of the external stimulus difference is C1. If it is higher than the threshold, it switches to ω0, and if it is lower than the threshold, it switches to ω1. ω1 is numerically equal to the memristor value of the synaptic switching memristor. The weight coefficient γ switches between 0 and 1, indicating whether the synapse is enabled, see formula 4; The switching condition is determined by the external stimulus I of the i-th neuron i and the external stimulus I of the jth neuron j The difference between them and the synaptic state state between neurons i and j. The synaptic state state has a low active state l and a high active state h. When the synaptic state is low active and the stimulus difference is less than the threshold, the synapse is closed and the synaptic weight is zero; under other conditions, the synapse is open and w ij =w, the synaptic weight switches between ω0 and ω1.

3. The chaotic neural network model based on multi-synaptic memristor as claimed in claim 1, characterized in that: The external stimulus generates a current I by applying a neuron state voltage to the electromagnetic radiation memristor. j , expressed as: Among them, k represents the influence coefficient, which is a constant; represents the electromagnetic radiation memristor flux acting on the jth neuron; represents the memristor value of the electromagnetic radiation memristor acting on the jth neuron; x j Represents the internal state voltage of the jth neuron.

4. The chaotic neural network model based on multi-synaptic memristor as claimed in claim 1, characterized in that: As the key generation part in the symmetric encryption method, after the initial value of the neural network is given and iterated, the chaotic sequence obtained is used as the key and combined with the encryption algorithm to realize the chaotic encryption function.

5. The chaotic neural network model based on multi-synaptic memristor as claimed in claim 4, characterized in that: The chaotic encryption completes the image encryption function based on the encryption ideas of confusion and diffusion; Key generation part: For the plaintext image, partition it and generate the corresponding network initial value, which is the key for the encryption party to transmit confidentially; Then use the network initial value to iterate, generate a chaotic sequence, and process it into a key matrix. The key matrix is ​​the key used by the encryption and decryption parties in the encryption algorithm. Encryption algorithm part: treat the single-channel image data as a two-dimensional matrix, realize pixel confusion through data row shift and scrambling transformation operations, and then perform data bit XOR with the key matrix generated by the chaotic sequence to obtain the encrypted image; The decryption operation is implemented through the corresponding symmetric operation.

Images