CN113516138A - Image processing method based on dual-mode memristor bridge synaptic circuit - Google Patents
Image processing method based on dual-mode memristor bridge synaptic circuit Download PDFInfo
- Publication number
- CN113516138A CN113516138A CN202110825319.8A CN202110825319A CN113516138A CN 113516138 A CN113516138 A CN 113516138A CN 202110825319 A CN202110825319 A CN 202110825319A CN 113516138 A CN113516138 A CN 113516138A
- Authority
- CN
- China
- Prior art keywords
- memristor
- bridge
- synaptic
- dual
- mode
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 230000000946 synaptic effect Effects 0.000 title claims abstract description 158
- 238000003672 processing method Methods 0.000 title claims abstract description 9
- 238000013528 artificial neural network Methods 0.000 claims abstract description 78
- 238000000034 method Methods 0.000 claims abstract description 58
- 238000004088 simulation Methods 0.000 claims abstract description 23
- 230000001413 cellular effect Effects 0.000 claims abstract description 21
- 238000000605 extraction Methods 0.000 claims abstract description 21
- 238000012545 processing Methods 0.000 claims abstract description 12
- GWEVSGVZZGPLCZ-UHFFFAOYSA-N Titan oxide Chemical compound O=[Ti]=O GWEVSGVZZGPLCZ-UHFFFAOYSA-N 0.000 claims description 35
- 210000000225 synapse Anatomy 0.000 claims description 35
- 238000013178 mathematical model Methods 0.000 claims description 25
- 238000002922 simulated annealing Methods 0.000 claims description 17
- 239000010408 film Substances 0.000 claims description 15
- BASFCYQUMIYNBI-UHFFFAOYSA-N platinum Chemical compound [Pt] BASFCYQUMIYNBI-UHFFFAOYSA-N 0.000 claims description 14
- 239000004408 titanium dioxide Substances 0.000 claims description 12
- 238000011156 evaluation Methods 0.000 claims description 9
- 229910052751 metal Inorganic materials 0.000 claims description 7
- 239000002184 metal Substances 0.000 claims description 7
- 229910052697 platinum Inorganic materials 0.000 claims description 7
- 210000004027 cell Anatomy 0.000 claims description 5
- 239000010409 thin film Substances 0.000 claims description 5
- 238000006243 chemical reaction Methods 0.000 claims description 4
- 230000009977 dual effect Effects 0.000 claims description 4
- 239000004065 semiconductor Substances 0.000 claims description 4
- 238000005259 measurement Methods 0.000 claims description 3
- 239000004576 sand Substances 0.000 claims description 3
- 230000009286 beneficial effect Effects 0.000 abstract description 10
- 230000004044 response Effects 0.000 abstract description 5
- 230000008859 change Effects 0.000 description 13
- 230000006870 function Effects 0.000 description 8
- 238000010586 diagram Methods 0.000 description 5
- 230000000694 effects Effects 0.000 description 4
- 238000010276 construction Methods 0.000 description 3
- 238000007796 conventional method Methods 0.000 description 3
- 238000012986 modification Methods 0.000 description 3
- 230000004048 modification Effects 0.000 description 3
- 230000001846 repelling effect Effects 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 230000003247 decreasing effect Effects 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 230000004907 flux Effects 0.000 description 2
- 150000002500 ions Chemical class 0.000 description 2
- 210000004556 brain Anatomy 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000012938 design process Methods 0.000 description 1
- 230000005684 electric field Effects 0.000 description 1
- 230000002401 inhibitory effect Effects 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 238000010845 search algorithm Methods 0.000 description 1
- 238000004260 weight control Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/06—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons
- G06N3/063—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons using electronic means
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Evolutionary Computation (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Neurology (AREA)
- Image Analysis (AREA)
Abstract
The invention provides an image processing method based on a dual-mode memristor bridge synaptic circuit, which comprises the following steps: constructing a dual-mode memristive bridge synaptic circuit through a linear memristor bridge synaptic circuit and a nonlinear memristor bridge synaptic circuit and an output memristive bridge synaptic circuit based on positive and negative pulses; performing simulation weight according to the dual-mode memristor bridge synaptic circuit, and constructing a dual-mode memristor bridge neural network through the simulation weight; and combining the dual-mode memristor bridge synaptic neural network with the cellular neural network, and performing edge extraction on the input image through the combined neural network to identify target elements in the input image. The invention has the beneficial effects that: the response time and the weight of the dual-mode memristive bridge synaptic neural network are faster and more accurate than those of the traditional method. Therefore, the dual-mode memristive bridge synaptic neural network is expected to be more real-time and solve more complex image processing.
Description
Technical Field
The invention relates to the technical field of image processing, in particular to an image processing method based on a dual-mode memristor bridge synaptic circuit.
Background
Artificial neural networks have been the focus of research. Since there are hundreds of millions of neurons and synapses in the human brain, implementing synaptic circuits is important for building brain-like machines. In artificial neural networks, a number of research results suggest that memristors may be used to simulate artificial synapses. In addition, it has great potential due to the nature of the memory synapse circuit. Among them, the most widely studied simulated synapse structure is the memory bridge structure, which has the advantages of simple structure, precise control, high integration level, etc.
However, in the prior art:
a synaptic bridge circuit composed of five identical memristors was proposed in conventional article of IEEE Circuit and System journal I, 59: 148-. Four memristors are used for controlling the positive and negative weights, and the last memristor is used for realizing the simulation of the synaptic weight. But this circuit cannot reach a zero weight. Meanwhile, the nonlinear phenomenon of memristance in the simulated synaptic circuit is also mentioned, which causes errors of weight simulation.
Von guang (2017), memristive bridge synapse-based neural network circuit research and application, major paper, Chinese, southwest university bipolar pulse method is used for inhibiting nonlinearity of memristance. However, it uses a synaptic bridge circuit composed of five identical memristors, and cannot be represented by a value with a synaptic weight of 0.
"Chenjing (2018), neural network based on memristor electronic synapse and application thereof, [ Master thesis ] [ China ]: university of southwest ] proposes a bridge synapse circuit composed of three identical memristors and two MOS transistors, which makes up the defect that the weight of a synapse circuit composed of five identical memristors cannot be zero, but cannot inhibit the nonlinearity of the memristors.
For the above-mentioned drawbacks, there is no solution for the time being.
Disclosure of Invention
The invention provides an image processing method based on a dual-mode memristor bridge synaptic circuit, which is used for solving the defects.
An image processing method based on a dual-mode memristive bridge synaptic circuit comprises the following steps:
constructing a dual-mode memristive bridge synaptic circuit through a linear memristor bridge synaptic circuit and a nonlinear memristor bridge synaptic circuit and an output memristive bridge synaptic circuit based on positive and negative pulses;
performing simulation weight according to the dual-mode memristor bridge synaptic circuit, and constructing a dual-mode memristor bridge neural network through the simulation weight;
and combining the dual-mode memristor bridge synaptic neural network with the cellular neural network, and performing edge extraction on the input image through the combined neural network to identify target elements in the input image.
As an embodiment of the present invention: the method further comprises the following steps:
determining a physical structure of a memristor;
constructing a memristor bridge structure based on simulated synapse through a physical structure of the memristor;
according to the memristor bridge structure of the simulated synapse, a mathematical model of the memristor is built through charge control; wherein,
the memristor model includes: a mathematical model of a linear memristor and a mathematical model of a nonlinear memristor.
As an embodiment of the present invention: the physical structure of the memristor includes: a metal platinum electrode and two layers of titanium dioxide films; wherein,
the two layers of titanium dioxide films are clamped between the metal platinum motors;
the two titanium dioxide films comprise a non-doped layer and a low-resistance doped layer;
the non-doped layer has high memristance;
the low-resistance doped layer has semiconductor characteristics.
As an embodiment of the present invention: through charge control, build the mathematical model of recalling the resistance ware, include:
determining a mathematical equation of the memristor based on the physical structure and charge control of the memristor:
wherein Q (t) represents a charge quantity function; i (t) represents the current of the memristor at time t; determining a mathematical model of a linear memristor according to a mathematical equation of the memristor:
M1(t)＝M(0)+k1Q(t)
wherein M (0) represents the initial memristance;
MONan ultimate resistance value of the memristor representing a thickness of the current doped layer; mOFFThe limiting resistance value of the memristor representing the total thickness of the two TiO2 thin films; d represents the total thickness of the two TiO2 films;
determining a mathematical model of a nonlinear memristor according to a mathematical equation of the memristor:
wherein Δ M ═ MOFF-MON；
As an embodiment of the present invention: the method further comprises the following steps:
setting parameters through a simulated annealing algorithm according to the linear memristor bridge synaptic circuit and the nonlinear memristor bridge synaptic circuit;
the setting steps are as follows:
step 2: t ═ jT, j is the descent rate, j ∈ (0,1), and n is the number of iterations. An evaluation function is obtained for each iteration,
for each memristor at the current input voltage, the evaluation function is the linear memristor and the nonlinear memristor at the time tTThe memristance difference between;
and step 3: randomly selecting t within a set voltage rangeTNew input voltage V at a momentin. Then calculating the memristance and evaluation function of the nonlinear memristor
And 4, step 4: if it is not
Then
Otherwise, returning to the step 3 if n is n + 1;
and 5: if it is
Accepting the changed input voltage, otherwise judging whether to accept the voltage according to probability
Let n be n + 1;
step 6: if it is not
The algorithm is terminated;
and 7: if n is less than L, returning to the step 3;
and 8: if T is<TzThe algorithm is terminated, otherwise step 2 is returned.
As an embodiment of the present invention: the method further comprises the following steps:
acquiring a memristive bridge synaptic circuit;
connecting the memristor bridge synaptic circuit with a switch circuit switched in a dual mode to form an output memristor bridge synaptic circuit with positive and negative pulses; wherein,
the switch circuit for dual-mode switching at least comprises a positive pulse switch interface, a negative pulse switch interface and a pulse conversion control signal interface.
As an embodiment of the present invention: the dual-mode memristive bridge synapse circuit is also connected with a differential pair circuit and an effective load circuit; wherein,
the dual-mode memristive bridge synaptic circuit is used for generating symmetrical positive and negative pulses and simulation weights, and determining an expected template operator through the positive and negative pulses and the simulation weights;
the differential pair circuit comprises a differential amplifier composed of transistors, and is used for being electrically connected with the output side of the dual-mode memristive bridge synapse circuit and converting the output voltage of the dual-mode memristive bridge synapse circuit into current;
the effective load circuit is electrically connected with the output side of the differential pair circuit and converts the current of the differential pair circuit into a voltage signal through a resistance load.
As an embodiment of the present invention: the neural network combining the dual-mode memristive bridge synaptic neural network and the cellular neural network is used for determining a cellular neural network template in an edge extraction mode; the cellular neural network template is used for identifying target elements through pixel iteration.
As an embodiment of the present invention: the dual-mode memristive bridge synaptic neural network comprises a plurality of dual-mode memristive bridge synaptic circuits, different pulse signals are generated through the dual-mode memristive bridge synaptic circuits, and various cell neural network templates are extracted from the edge through the different pulse signals.
As an embodiment of the present invention: and the cell neural network template extracted from the edge is used for comparing and identifying the characteristics of the target elements in the input image to determine the target elements.
The invention has the beneficial effects that:
the invention provides a novel dual-mode memristor bridge synaptic circuit for simulating synaptic weights, and the dual-mode memristor bridge synaptic circuit is combined with a cellular neural network to construct a dual-mode memristor bridge synaptic neural network. And applies the network to edge extraction of the image. The method can better solve the weight simulation error caused by the nonlinear memristor, thereby accelerating the speed of edge extraction and improving the acquisition precision of the cellular neural network template.
The method firstly performs mathematical modeling on the linear memristor and the nonlinear memristor and performs corresponding simulation. In particular, the nonlinear memristor model is closer to the actual memristor. Then, a novel memristive bridge synaptic circuit and a dual-mode memristive bridge synaptic circuit are provided. The two circuits respectively inhibit the nonlinearity of the synapse weight and the output voltage, and reduce the weight simulation error. And finally, passing through a dual-mode memristive bridge neural network. The image is processed accordingly.
The implementation of the method is premised on determining the memristor change of different types of linear and nonlinear memristor models, and the method can verify that the memristor change of the nonlinear memristor models is closer to the physical characteristic. The invention provides for linearizing synaptic weights by varying an input voltage of a non-linear memristive bridge synaptic circuit. The invention further comprises a method for accurately inputting the voltage based on the simulated annealing method, so that the change of the synaptic circuit of the nonlinear memristor bridge is more linear. However, when the input voltage changes, the corresponding output voltage also changes, and the nonlinearity is intensified, so the dual-mode memristive bridge synaptic circuit is adopted in the invention, and because the influence of symmetrical positive and negative pulses on the synaptic weight is minimum, the output voltage and the synaptic weight are linearized. Finally, the neural network proposed herein is formed using dual-mode memristive bridge-synaptic circuits for extracting the edges of the image. The response time and the weight of the dual-mode memristive bridge synaptic neural network are faster and more accurate than those of the traditional method. Therefore, the dual-mode memristive bridge synaptic neural network is expected to be more real-time and solve more complex image processing.
Additional features and advantages of the invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. The objectives and other advantages of the invention will be realized and attained by the structure particularly pointed out in the written description and drawings.
The technical solution of the present invention is further described in detail by the accompanying drawings and embodiments.
Drawings
The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the principles of the invention and not to limit the invention. In the drawings:
fig. 1 is a flowchart of a method for processing an image based on a dual-mode memristive bridge synaptic circuit according to an embodiment of the present disclosure.
FIG. 2 is a diagram of a memristor architecture in an embodiment of the present disclosure;
FIG. 3 is a linear schematic diagram of a memristor model with linear and nonlinear charge control based on a memristance vs.Q in an embodiment of the present disclosure;
FIG. 4 is a diagram illustrating the synaptic weights between the nonlinear and linear models by varying the input pair according to an embodiment of the present invention;
FIG. 5 is a graph of memristance comparisons between linear and nonlinear memristor models with and without the use of the modified simulated annealing method in an embodiment of the present disclosure;
FIG. 6 is a graph of output voltage variation for linear and nonlinear memristor models with and without the use of a modified simulated annealing method in an embodiment of the present disclosure;
FIG. 7 is a diagram of a synaptic circuit of a dual-mode memristive bridge generating positive and negative pulses and controlling synaptic weights in an embodiment of the present disclosure;
FIG. 8 is a graph comparing output voltages of a linear memristor circuit, a nonlinear memristor circuit employing an improved simulated annealing method, and a dual-mode memristor bridge synapse circuit in an embodiment of the present disclosure;
FIG. 9 is a circuit diagram of a dual-mode memristive bridge synaptic neural network in an embodiment of the present disclosure;
FIG. 10 is a graph of (a) edge extraction results using a conventional non-linear memristive bridge synaptic neural network, in an embodiment of the present disclosure;
FIG. 11 is a graph of edge extraction results using a dual-mode memristive bridge synaptic neural network in an embodiment of the present disclosure.
Detailed Description
The preferred embodiments of the present invention will be described in conjunction with the accompanying drawings, and it will be understood that they are described herein for the purpose of illustration and explanation and not limitation.
In the present invention:
a memristor is a synaptic element with nanometer dimensions and continuously variable memristance. A bridge type synaptic circuit formed by memristors is simple in structure and accurate in control. In practice, due to the nonlinear characteristic of the memristor, a certain influence is generated on the control of the synaptic circuit, and a weight error is caused.
The invention relates to an image processing method based on a dual-mode memristive bridge synaptic circuit, which is shown in the attached figure 1 and comprises the following steps:
step 100: constructing a dual-mode memristive bridge synaptic circuit through a linear memristor bridge synaptic circuit and a nonlinear memristor bridge synaptic circuit and an output memristive bridge synaptic circuit based on positive and negative pulses;
step 101: performing simulation weight according to the dual-mode memristor bridge synaptic circuit, and constructing a dual-mode memristor bridge neural network through the simulation weight;
step 102: and combining the dual-mode memristor bridge synaptic neural network with the cellular neural network, and performing edge extraction on the input image through the combined neural network to identify target elements in the input image.
The working principle of the technical scheme is as follows: the invention provides a novel dual-mode memristor bridge synaptic circuit for simulating synaptic weights, and the dual-mode memristor bridge synaptic circuit is combined with a cellular neural network to construct a dual-mode memristor bridge synaptic neural network. And applies the network to edge extraction of the image. The method can better solve the weight simulation error caused by the nonlinear memristor, thereby accelerating the speed of edge extraction and improving the acquisition precision of the cellular neural network template.
The method firstly performs mathematical modeling on the linear memristor and the nonlinear memristor and performs corresponding simulation. In particular, the nonlinear memristor model is closer to the actual memristor. Then, a novel memristive bridge synaptic circuit and a dual-mode memristive bridge synaptic circuit are provided. The two circuits respectively inhibit the nonlinearity of the synapse weight and the output voltage, and reduce the weight simulation error. And finally, passing through a dual-mode memristive bridge neural network. The image is processed accordingly.
The implementation of the method is premised on determining the memristor change of different types of linear and nonlinear memristor models, and the method can verify that the memristor change of the nonlinear memristor models is closer to the physical characteristic. The invention provides for linearizing synaptic weights by varying an input voltage of a non-linear memristive bridge synaptic circuit. The invention further comprises a method for accurately inputting the voltage based on the simulated annealing method, so that the change of the synaptic circuit of the nonlinear memristor bridge is more linear. However, when the input voltage changes, the corresponding output voltage also changes, and the nonlinearity is intensified, so the dual-mode memristive bridge synaptic circuit is adopted in the invention, and because the influence of symmetrical positive and negative pulses on the synaptic weight is minimum, the output voltage and the synaptic weight are linearized. Finally, the neural network proposed herein is formed using dual-mode memristive bridge-synaptic circuits for extracting the edges of the image.
The beneficial effects of the above technical scheme are: the response time and the weight of the dual-mode memristive bridge synaptic neural network are faster and more accurate than those of the traditional method. Therefore, the dual-mode memristive bridge synaptic neural network is expected to be more real-time and solve more complex image processing.
In one practical embodiment: bird strike is a huge threat to modern civil aviation, and brings huge loss to the life and economy of people. Therefore, researchers and related departments are urgently looking for finding a proper bird repelling method at present. At present, birds are generally repelled by means of radar to obtain bird information, and then corresponding bird repelling equipment is used. However, the information displayed on the radar can only be related to the approximate bird flight direction, and the type and number of birds cannot be identified. If more detailed bird information can be obtained, a more effective and more targeted bird repelling method can be adopted. Therefore, the image processing method of the invention can solve the problems of bird species identification and bird counting according to bird images around or passing through an airport.
As an embodiment of the present invention: the method further comprises the following steps:
determining a physical structure of a memristor;
constructing a memristor bridge structure based on simulated synapse through a physical structure of the memristor;
according to the memristor bridge structure of the simulated synapse, a mathematical model of the memristor is built through charge control; wherein,
the memristor model includes: a mathematical model of a linear memristor and a mathematical model of a nonlinear memristor.
The working principle of the technical scheme is as follows: the invention is based on the synaptic circuit of the dual-mode memristive bridge, so the construction of the synaptic circuit of the dual-mode memristive bridge can be firstly carried out, but the construction of the synaptic circuit of the dual-mode memristive bridge is not limited by a memristor, because a mathematical model of the memristor is constructed in advance according to the memristor and a point and control, the construction of the neural network of the synaptic circuit of the dual-mode memristive bridge is further realized.
The beneficial effects of the above technical scheme are: the electric charge quantity is controllable by establishing a mathematical model of the linear memristor and a mathematical model of the nonlinear memristor, the electric charge quantity also represents that synapses are controllable, the synapses are controllable, and the synapses are controllable, so that the synapses of the nonlinear memristor model are close to the synapses of the linear memristor model, and the synapses required by image recognition are reached more quickly.
As an embodiment of the present invention: the physical structure of the memristor includes: a metal platinum electrode and two layers of titanium dioxide films; wherein,
the two layers of titanium dioxide films are clamped between the metal platinum motors;
the two titanium dioxide films comprise a non-doped layer and a low-resistance doped layer;
the non-doped layer has high memristance;
the low-resistance doped layer has semiconductor characteristics.
The above techniqueThe working principle of the scheme is as follows: the method is characterized in that controllable memristors are needed for establishing a synaptic circuit of the dual-mode memristor bridge, and the controllable memristors are divided into charge controllable memristors and magnetic flux control memristors; but because charge-controllable memristors have good nonlinear and linear characteristics; therefore, it is consistent with the present invention to implement synapses, and thus the present invention employs charge-controllable memristors as shown in fig. 2, where 1 is a memristor model and 2 is a memristor symbol. The memristor is formed by two layers of titanium dioxide (TiO) between two metal platinum electrodes2) And (3) film composition. One layer of TiO2The thin film is a non-doped layer and has high memristance. Another layer of TiO2The thin film is a doped layer having semiconductor characteristics and corresponding to low resistance. The resistance value of the high-voltage memristor is M (t).
Where ω (t) is the thickness of the now doped layer and D is two TiO s2The total thickness of the film. MOFFAnd MONAnd the limiting resistance values, i.e., the maximum resistance value and the minimum resistance value, of the D memristor, ω (t) ═ 0 and ω (t) ═ D, respectively.
The beneficial effects of the above technical scheme are: because nonlinear and linear data analysis is needed, the invention determines the memristor in advance, and the physical memristor determined by the invention is a circuit element with a nanometer size, so that the memristor generates a huge electric field when a tiny voltage is applied. Thus, the ion motion inside the memristor has a significant nonlinear characteristic.
As an embodiment of the present invention: through charge control, build the mathematical model of recalling the resistance ware, include:
determining a mathematical equation of the memristor based on the physical structure and charge control of the memristor:
wherein Q (t) represents a charge quantity function; i (t) represents the current of the memristor at time t; determining a mathematical model of a linear memristor according to a mathematical equation of the memristor:
M1(t)＝M(0)+k1Q(t)
wherein M (0) represents the initial memristance;
MONan ultimate resistance value of the memristor representing a thickness of the current doped layer; mOFFRepresents two TiO2The limiting resistance value of the memristor for the total thickness of the thin film; d represents two TiO2The total thickness of the film;
determining a mathematical model of a nonlinear memristor according to a mathematical equation of the memristor:
wherein Δ M ═ MOFF-MON；
The working principle of the technical scheme is as follows: because the controllable memristor is needed for establishing the dual-mode memristor bridge circuit, two models of the controllable memristor are established; respectively a mathematical model of a linear memristor and a mathematical model of a nonlinear memristor; as shown in FIG. 3, the mathematical model of the memristor according to the present invention shows the relationship between the memristance change and the charge or flux for different types of linear and nonlinear memristor models, for which the memristance change reaches the threshold earlier than the nonlinear memristor; when the charge reaches or exceeds the upper or lower limit, the memristance will reach and maintain a maximum or minimum value. k is a radical of1And k2The parameter is expressed without practical meaning and is defined by specific limiting impedance value and TiO2The total thickness of the film was determined. EtavIs a constant of memristor ion mobility in uniform field, etav≈10-14m2·s-1·V-1。
The beneficial effects of the above technical scheme are: according to the method, the dual-mode memristor bridge circuit can be judged to realize synapse weight control through the design of the mathematical model. The linearization of synaptic weights and output voltages may also be facilitated by the control of synaptic weights through the relationship between the nonlinear and linear models.
As an embodiment of the present invention: the method further comprises the following steps:
setting parameters through a simulated annealing algorithm according to the linear memristor bridge synaptic circuit and the nonlinear memristor bridge synaptic circuit;
the setting steps are as follows:
step 2: t ═ jT, j is the descent rate, j ∈ (0,1), and n is the number of iterations. An evaluation function is obtained for each iteration,
for each memristor at the current input voltage, the evaluation function is the linear memristor and the nonlinear memristor at the time tTThe memristance difference between;
and step 3: randomly selecting t within a set voltage rangeTNew input voltage V at a momentin. Then calculating the memristance and evaluation function of the nonlinear memristor
And 4, step 4: if it is not
Then
Otherwise, returning to the step 3 if n is n + 1;
and 5: if it is
Accepting the changed input voltage, otherwise judging whether to accept the voltage according to probability
Let n be n + 1;
step 6: if it is not
And 7: if n is less than L, returning to the step 3;
and 8: if T is<TzThe algorithm is terminated, otherwise step 2 is returned.
The working principle of the technical scheme is as follows: the invention adopts a simulated annealing method to obtain more accurate input voltage, and then controls the change of the synaptic weight by the voltage; by varying the input V, as shown in FIG. 4inA memristive bridge synaptic circuit based on improved simulated annealing of a nonlinear memristor model is constructed. Therefore, it is necessary to find the change input VinInfluencing the entire circuit, e.g. increasing or decreasing input Vin. Initial input V assuming linear and nonlinear memristor modelsinAt 1v, the input is increased and decreased by 0.0005v, respectively. It can be found that by reducing the input VinThe difference between linear and non-linear is larger and larger, and by increasing the input, the synaptic weight of the non-linear model can follow the synaptic weight V of the linear model within a certain rangein. Then through the above steps, the invention can determine that the nonlinear memristor and the linear memristor bridge synaptic circuit set the same memristor parameters, as shown in FIGS. 5 and 6. FIG. 5 is a memristance M between linear and nonlinear memristor models with and without the modified simulated annealing method1And (6) comparing. As shown in FIG. 5, when the simulated annealing method selects the exact input voltage, it comes from the non-linearityMemristance M of memristive bridge synaptic circuit1Can be significantly linearized. FIG. 6 is the output voltage variation of linear and nonlinear memristor models with and without the modified simulated annealing method. As can be seen from fig. 6, with the improved simulated annealing method, the output voltage of the synaptic circuit of the nonlinear memristive bridge is not linear, and the absolute value of the output voltage is much larger than that of the synaptic circuit of the linear memristive bridge, for example, due to the nonlinearity of the input voltage, the time is further.
The beneficial effects of the above technical scheme are: selecting proper input V by adopting simulated annealing method random search algorithmintSimple structure and high search speed
As an embodiment of the present invention: the method further comprises the following steps:
acquiring a memristive bridge synaptic circuit;
connecting the memristor bridge synaptic circuit with a switch circuit switched in a dual mode to form an output memristor bridge synaptic circuit with positive and negative pulses; wherein,
the switch circuit for dual-mode switching at least comprises a positive pulse switch interface, a negative pulse switch interface and a pulse conversion control signal interface.
The principle of the technical scheme is as follows: when an input voltage or current is applied to the memristor, the memristance may change. Thus, the present invention reduces the effect of the input voltage applied to the memristor by applying positive and negative pulses across the memristor by constructing the synaptic electrical circuit that can switch between the two modes and generate positive and negative pulses. As shown in FIG. 7, the dual mode switching is through a single pole double throw switch S1And (4) realizing. Switch S1With two interfaces K1And K2And a pulse conversion control signal interface. Control signal interface by signal PAImposition of S1At K1And K2For control and output modes. Positive and negative pulse single-pole double-throw switch S2、S3And single-pole single-throw switch S4A handover occurs. Switch S2With two interfaces K3、K4Switch S3With two interfaces K5、K6Are respectively lettersNumber PBAnd PCThe applied double pulse switches the control signal interface. Single-pole single-throw switch S4With control of signal P by pulsesDThe applied pulses switch the control signal interface. The circuit part for obtaining the output voltage is composed of a MOS transistor T1And T2And (4) forming. MOS transistor T1And T2All turn-on voltages of VTThe gate voltages of the two transistors are both V1. In the control phase, to obtain zero output voltage, let V1<VT. Otherwise, let V in the output stage1>VT. The specific effect is shown in fig. 8, where the red line represents the non-linear output voltage, since only the control mode is active. If so, the line of output voltage becomes more linear, as shown by the orange line. The dual-mode memristor bridge synaptic circuit is verified to realize that both synaptic weight and output voltage can be linearized, and the weight obtaining speed is accelerated. In FIG. 8, output voltage comparisons of a linear memristor circuit (blue line), a nonlinear memristor circuit using a modified simulated annealing method (red line), and a dual-mode memristive bridge synapse circuit (orange line)
The beneficial effects of the above technical scheme are: the speed of acquiring the weight can be increased.
As an embodiment of the present invention: the dual-mode memristive bridge synapse circuit is also connected with a differential pair circuit and an effective load circuit; wherein,
the dual-mode memristive bridge synaptic circuit is used for generating symmetrical positive and negative pulses and simulation weights, and determining an expected template operator through the positive and negative pulses and the simulation weights;
the differential pair circuit comprises a differential amplifier composed of transistors, and is used for being electrically connected with the output side of the dual-mode memristive bridge synapse circuit and converting the output voltage of the dual-mode memristive bridge synapse circuit into current;
the effective load circuit is electrically connected with the output side of the differential pair circuit and converts the current of the differential pair circuit into a voltage signal through a resistance load.
The principle of the technical scheme is as follows: in the design process of the dual-mode memristive bridge synaptic circuit, the synaptic weight needs to change in a linear trend, so that a load band circuit and a differential pair circuit are required to support. As shown in fig. 9, the circuit is mainly divided into a dual-mode memristive bridge synapse circuit a, a differential pair circuit B and an effective load circuit C;
the beneficial effects of the above technical scheme are: the synaptic circuit in the control mode of the dual-mode memristive bridge synaptic circuit A changes the input voltage, so that the synaptic weight changes in a linear trend. In the output mode, by generating symmetrical positive and negative pulses, the influence on the synaptic weight of the synaptic circuit of the memristive bridge is minimum, and the expected template operator can be obtained more quickly. The whole neural network consists of a plurality of dual-mode memristive bridge synaptic circuits, and the synaptic circuits are mutually independent.
The output voltage of the dual-mode memristive bridge synaptic circuit is converted into a current through the differential pair circuit B.
Each dual-mode memristive bridge synaptic circuit is connected in series through a payload circuit C. The circuit converts current from the differential pair circuit through a resistive load into a voltage signal.
As an embodiment of the present invention: the neural network combining the dual-mode memristive bridge synaptic neural network and the cellular neural network is used for determining a cellular neural network template in an edge extraction mode; the cellular neural network template is used for identifying target elements through pixel iteration.
As an embodiment of the present invention: the dual-mode memristive bridge synaptic neural network comprises a plurality of dual-mode memristive bridge synaptic circuits, different pulse signals are generated through the dual-mode memristive bridge synaptic circuits, and various cell neural network templates are extracted from the edge through the different pulse signals.
The principle and the beneficial effects of the technical scheme are as follows: the invention combines the dual-mode memristor bridge synaptic neural network with the cellular neural network template to realize the edge extraction of the pictures. A common cellular neural network template is 3 × 3, and thus is implemented herein using a plurality of dual-mode memristive bridge synaptic neural networks. In each circuit, by inputting different pulse signals, a corresponding output value can be obtained. As shown in figure 9:
the template based on edge extraction comprises
The parameters of the memristor in fig. 9 are shown in the formula below. The MOS transistor in FIG. 9 has a parameter VT＝3v。
The pixels of the M × N image are inserted into a cellular neural network of a template size of 3 × 3, and the pixels are iterated according to the following equation.
Wherein v isxIs the state value, vyIs the output value, vuIs an input value, i is more than or equal to 1 and less than or equal to M, j is more than or equal to 1 and less than or equal to N, P1(i，j)＝{c(k，l)|max{|k-i|，|l-j|}≤1，1≤k≤M，1≤l≤N}
P1(i,j)＝{c(k,l)|max{|k-i|,|l-j|}≤1,1≤k≤M,1≤l≤N}。
In one embodiment, as shown in FIGS. 10 and 11: FIG. 10 shows the results of edge extraction using a nonlinear memristive bridge synaptic circuit. The bird image edge extraction result of the dual-mode memristive bridge synaptic neural network is realized by using the template, as shown in fig. 11. The edge extraction response speed of the dual-mode memristive bridge synaptic neural network is faster than that of a nonlinear memristive circuit. The invention carries out quantitative comparison on different edge extraction algorithms by using performance indexes FOM (figure Of Merit), wherein the FOM equation is as follows.
Wherein QiIs the original templateThe number of edge pixels of (c). QtIs the number of edge pixels actually detected. A is a compensation coefficient, a is 1/4, diThe shortest distance between the detected edge point and the edge point of the original template. FOM is in the range of [0,1 ]]The larger the FOM, the better the effect. Bird image edge results from different methods were evaluated and the FOM values are shown in table 1.
TABLE 1 two memristor bridge synaptic neural network index comparisons
| Conventional methods | Methods of the disclosure | |
| FOM | 0.9557 | 0.9865 |
| Number of |
90 | 78 |
As can be seen from Table 1, the FOM value obtained by the method proposed herein is greater than that of the conventional method, indicating that the method has better edge extraction effect. The number of iterations of the method is also smaller than that of the conventional method. Therefore, the dual-mode memristive-bridge synaptic neural network has more efficient performance in terms of edge extraction than a traditional non-linear memristive-bridge synaptic neural network.
As an embodiment of the present invention: and the cell neural network template extracted from the edge is used for comparing and identifying the characteristics of the target elements in the input image to determine the target elements.
The method analyzes the memristor changes of different types of linear and nonlinear memristor models, and verifies that the memristor changes of the nonlinear memristor models are closer to physical characteristics. Then, it is proposed herein to linearize synaptic weights by varying the input voltage of a non-linear memristive bridge synaptic circuit. Therefore, an accurate input voltage method based on a simulated annealing method is designed, so that the change of the synaptic circuit of the nonlinear memristive bridge is more linear. However, as the input voltage changes, the corresponding output voltage also changes, and the nonlinearity is exacerbated. Therefore, a dual-mode memristive bridge synaptic circuit is proposed, because symmetric positive and negative pulses have minimal influence on synaptic weights, so that the output voltage and synaptic weights are linearized. Finally, the neural network proposed herein is formed using dual-mode memristive bridge-synaptic circuits for extracting the edges of bird images. According to simulation results, the response time and the weight of the dual-mode memristive bridge synaptic neural network are faster and more accurate than those of the traditional method. Therefore, the dual-mode memristive bridge synaptic neural network is expected to be more real-time and solve more complex image processing.
It will be apparent to those skilled in the art that various changes and modifications may be made in the present invention without departing from the spirit and scope of the invention. Thus, if such modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include such modifications and variations.
Claims (10)
1. An image processing method based on a dual-mode memristive bridge synaptic circuit is characterized by comprising the following steps:
constructing a dual-mode memristive bridge synaptic circuit through a linear memristor bridge synaptic circuit and a nonlinear memristor bridge synaptic circuit and an output memristive bridge synaptic circuit based on positive and negative pulses;
performing simulation weight according to the dual-mode memristor bridge synaptic circuit, and constructing a dual-mode memristor bridge neural network through the simulation weight;
and combining the dual-mode memristor bridge synaptic neural network with the cellular neural network, and performing edge extraction on the input image through the combined neural network to identify target elements in the input image.
2. The method of image processing based on a dual-mode memristive bridge synaptic electrical circuit of claim 1, wherein the method further comprises:
determining a physical structure of a memristor;
constructing a memristor bridge structure based on simulated synapse through a physical structure of the memristor;
according to the memristor bridge structure of the simulated synapse, a mathematical model of the memristor is built through charge control; wherein,
the memristor model includes: a mathematical model of a linear memristor and a mathematical model of a nonlinear memristor.
3. The method of claim 2, in which the physical structure of the memristor comprises: a metal platinum electrode and two layers of titanium dioxide films; wherein,
the two layers of titanium dioxide films are clamped between the metal platinum motors;
the two titanium dioxide films comprise a non-doped layer and a low-resistance doped layer;
the non-doped layer has high memristance;
the low-resistance doped layer has semiconductor characteristics.
4. The method for processing the image based on the synaptic circuit of the dual-mode memristor bridge as claimed in claim 2, wherein the building of the mathematical model of the memristor through charge control comprises:
determining a mathematical equation of the memristor based on the physical structure and charge control of the memristor:
wherein Q (t) represents a charge quantity function; i (t) represents the current of the memristor at time t;
determining a mathematical model of a linear memristor according to a mathematical equation of the memristor:
M1(t)＝M(0)+k1Q(t)
wherein M (0) represents the initial memristance;
MONan ultimate resistance value of the memristor representing a thickness of the current doped layer; mOFFRepresents two TiO2The limiting resistance value of the memristor for the total thickness of the thin film; d represents two TiO2The total thickness of the film;
determining a mathematical model of a nonlinear memristor according to a mathematical equation of the memristor:
wherein Δ M ═ MOFF-MON；
5. The method of image processing based on a dual-mode memristive bridge synaptic electrical circuit of claim 1, wherein the method further comprises:
setting parameters through a simulated annealing algorithm according to the linear memristor bridge synaptic circuit and the nonlinear memristor bridge synaptic circuit;
the setting steps are as follows:
step 1 setting an initial input voltage V by measurementintTemperature T, end temperature TzMaximum number of iterations L per temperature T, interval time TsAnd a maximum value at which the input voltage changes at each instant, the input voltage being selected at each instant between the initial input voltage and the maximum voltage;
step 2: t ═ jT, j is the descent rate, j ∈ (0,1), and n is the number of iterations. Obtaining an evaluation function for each iteration，
For each memristor at the current input voltage, the evaluation function is the linear memristor and the nonlinear memristor at the time tTThe memristance difference between;
and step 3: randomly selecting t within a set voltage rangeTNew input voltage V at a momentin. Then calculating the memristance and evaluation function of the nonlinear memristor
And 4, step 4: if it is not
Then
Otherwise, returning to the step 3 if n is n + 1;
and 5: if it is
Accepting the changed input voltage, otherwise judging whether to accept the voltage according to probability
Let n be n + 1;
step 6: if it is not
The algorithm is terminated;
and 7: if n is less than L, returning to the step 3;
and 8: if T is<TzThe algorithm is terminated, otherwise step 2 is returned.
6. The method of image processing based on a dual-mode memristive bridge synaptic electrical circuit of claim 1, wherein the method further comprises:
acquiring a memristive bridge synaptic circuit;
connecting the memristor bridge synaptic circuit with a switch circuit switched in a dual mode to form an output memristor bridge synaptic circuit with positive and negative pulses; wherein,
the switch circuit for dual-mode switching at least comprises a positive pulse switch interface, a negative pulse switch interface and a pulse conversion control signal interface.
7. The method for processing the image based on the dual-mode memristive bridge-synaptic circuit according to claim 1, wherein a differential pair circuit and a payload circuit are further connected to the dual-mode memristive bridge-synaptic circuit; wherein,
the dual-mode memristive bridge synaptic circuit is used for generating symmetrical positive and negative pulses and simulation weights, and determining an expected template operator through the positive and negative pulses and the simulation weights;
the differential pair circuit comprises a differential amplifier composed of transistors, and is used for being electrically connected with the output side of the dual-mode memristive bridge synapse circuit and converting the output voltage of the dual-mode memristive bridge synapse circuit into current;
the effective load circuit is electrically connected with the output side of the differential pair circuit and converts the current of the differential pair circuit into a voltage signal through a resistance load.
8. The method for processing the image based on the dual-mode memristive bridge synaptic electrical circuit according to claim 1, wherein the neural network of the dual-mode memristive bridge synaptic neural network in combination with the cellular neural network is used for determining the cellular neural network template by means of edge extraction; the cellular neural network template is used for identifying target elements through pixel iteration.
9. The method as claimed in claim 8, wherein the dual-mode memristive bridge synaptic neural network comprises a plurality of dual-mode memristive bridge synaptic circuits, different pulse signals are generated by the dual-mode memristive bridge synaptic circuits, and a plurality of cell neural network templates are edge-extracted by the different pulse signals.
10. The method as claimed in claim 8, wherein the edge-extracted cellular neural network template is used for performing contrast recognition with features of a target element in an input image to determine the target element.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110825319.8A CN113516138B (en) | 2021-07-21 | 2021-07-21 | Image processing method based on dual-mode memristor bridge synaptic circuit |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110825319.8A CN113516138B (en) | 2021-07-21 | 2021-07-21 | Image processing method based on dual-mode memristor bridge synaptic circuit |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113516138A true CN113516138A (en) | 2021-10-19 |
| CN113516138B CN113516138B (en) | 2022-07-29 |
Family
ID=78068532
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110825319.8A Active CN113516138B (en) | 2021-07-21 | 2021-07-21 | Image processing method based on dual-mode memristor bridge synaptic circuit |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113516138B (en) |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20090129547A1 (en) * | 2007-11-16 | 2009-05-21 | General Electric Company | Portable dual-mode digital x-ray detector and methods of operation of same |
| CN107016200A (en) * | 2017-04-14 | 2017-08-04 | 江西理工大学 | A kind of circuit design method of the smooth cell neural network based on magnetic control memristor |
| CN107085628A (en) * | 2017-03-21 | 2017-08-22 | 东南大学 | A kind of adjustable weights modular simulation method of cell neural network |
| CN109460818A (en) * | 2018-09-25 | 2019-03-12 | 电子科技大学 | A kind of multilayer neural network design method based on memristor bridge and array |
| CN110991633A (en) * | 2019-12-04 | 2020-04-10 | 电子科技大学 | Residual error neural network model based on memristor network and application method thereof |
| CN111815640A (en) * | 2020-07-21 | 2020-10-23 | 江苏经贸职业技术学院 | Memristor-based RBF neural network medical image segmentation algorithm |
| CN111833278A (en) * | 2020-07-13 | 2020-10-27 | 西南大学 | Picture processing method based on memristor forgetting characteristic |
-
2021
- 2021-07-21 CN CN202110825319.8A patent/CN113516138B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20090129547A1 (en) * | 2007-11-16 | 2009-05-21 | General Electric Company | Portable dual-mode digital x-ray detector and methods of operation of same |
| CN107085628A (en) * | 2017-03-21 | 2017-08-22 | 东南大学 | A kind of adjustable weights modular simulation method of cell neural network |
| CN107016200A (en) * | 2017-04-14 | 2017-08-04 | 江西理工大学 | A kind of circuit design method of the smooth cell neural network based on magnetic control memristor |
| CN109460818A (en) * | 2018-09-25 | 2019-03-12 | 电子科技大学 | A kind of multilayer neural network design method based on memristor bridge and array |
| CN110991633A (en) * | 2019-12-04 | 2020-04-10 | 电子科技大学 | Residual error neural network model based on memristor network and application method thereof |
| CN111833278A (en) * | 2020-07-13 | 2020-10-27 | 西南大学 | Picture processing method based on memristor forgetting characteristic |
| CN111815640A (en) * | 2020-07-21 | 2020-10-23 | 江苏经贸职业技术学院 | Memristor-based RBF neural network medical image segmentation algorithm |
Non-Patent Citations (1)
| Title |
|---|
| 冯广等: "双极性脉冲忆阻桥电子突触神经网络及图像处理", 《中国科学：信息科学》 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113516138B (en) | 2022-07-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Liu et al. | Implementation of memristive neural network with full-function Pavlov associative memory | |
| Nossek | Design and learning with cellular neural networks | |
| CN110826602A (en) | Image classification method and system based on membrane potential regulation and control pulse neural network | |
| CN110289317B (en) | Ferroelectric graphene transistor, complementary type synapse device based on ferroelectric graphene transistor and regulation and control method | |
| CN112257815A (en) | Model generation method, target detection method, device, electronic device, and medium | |
| CN104573238A (en) | Circuit design method for memory resisting cell neural network | |
| CN109840595B (en) | Knowledge tracking method based on group learning behavior characteristics | |
| CN105243454A (en) | Big data-based electrical load prediction system | |
| CN110363799B (en) | Multi-motion human target tracking method based on vision under man-machine coexistence environment | |
| CN113516138B (en) | Image processing method based on dual-mode memristor bridge synaptic circuit | |
| CN109901064B (en) | ICA-LVQ-based high-voltage circuit breaker fault diagnosis method | |
| Hussain et al. | Delay learning architectures for memory and classification | |
| Donahue et al. | Design and analysis of neuromemristive echo state networks with limited-precision synapses | |
| Afshari et al. | Unsupervised learning in hexagonal boron nitride memristor-based spiking neural networks | |
| Hajiabadi et al. | Memristor-based synaptic plasticity and unsupervised learning of spiking neural networks | |
| Bavandpour et al. | Spiking neuro-fuzzy clustering system and its memristor crossbar based implementation | |
| Bodyanskiy et al. | Evolving cascaded neural network based on multidimensional Epanechnikov’s kernels and its learning algorithm | |
| CN109376803A (en) | Multiple neural networks classifier fusion method and device based on fuzzy complex sets value integral | |
| CN111582461B (en) | Neural network training method and device, terminal equipment and readable storage medium | |
| Liu et al. | A multi-terminal ion-controlled transistor with multifunctionality and wide temporal dynamics for reservoir computing | |
| Mladenov et al. | Learning of an Artificial Neuron with Resistor-Memristor Synapses | |
| Wang et al. | Dual-mode memristor synaptic circuit design and application in image processing | |
| CN113537398A (en) | Color value evaluation model training method and component, and color value evaluation method and component | |
| Liu et al. | Non-local mean filtering algorithm based on deep learning | |
| CN111309145B (en) | Electrostatic force touch rendering method based on physiological and physical modeling |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |