A pseudo-neural network classification method and system based on hypersurface

Abstract

The present invention relates to a pseudo-neural network classification method and system based on a hypersurface, aiming to address the problems of gradient vanishing, gradient exploding, and local minimum points inherent in existing optimization algorithms. The method comprises: generating first electromagnetic wave information carrying the true category; processing the first electromagnetic wave information using an initial network model to obtain second electromagnetic wave information; processing the second electromagnetic wave information using a decision function to obtain an initial predicted category of the target to be classified; determining whether the initial predicted category is the same as the true category to obtain a judgment result; performing a multi-level synchronous iterative update of the initial neural network model based on the judgment result to obtain a result neural network model; and using the result neural network model to predict the result category of the first electromagnetic wave information. The present invention improves the parameter optimization method from layer-by-layer optimization to simultaneous optimization of all layers, greatly improving the model training speed.

Description

Technical Field

[0001] The present invention belongs to the field of classification technology, and in particular relates to a pseudo-neural network classification method and system based on a hypersurface. Background Art

[0002] Neural network algorithms are a key branch of artificial intelligence. Neural networks mimic the human nervous system and consist of multiple neurons interconnected in a hierarchical structure. They extract features from data and then perform predictions or decision-making tasks. Currently, neural network technology has been widely applied across various industries and has achieved remarkable results in many fields.

[0003] In the field of classification technology, integrating neural network algorithms into classification technology can relatively accurately determine the category of the target object. In the existing technology, the optimization algorithm for deep learning classification models usually uses the gradient back propagation algorithm or other variant algorithms.

[0004] The existing technology has at least the following problems:

[0005] 1. The existing gradient backpropagation algorithm not only has potential problems such as gradient vanishing, gradient exploding, and local minimum points, but also has the disadvantages of slow calculation speed and large data requirements.

[0006] 2. The existing neural network classification model training process uses CPU and GPU servers to train and load positioning models, which is costly and consumes a lot of resources. Summary of the Invention

[0007] The present invention provides a pseudo-neural network classification method and system based on a hypersurface, aiming to solve the technical problems in the above-mentioned prior art of the gradient backpropagation algorithm, such as slow calculation speed, large data volume requirement, and potential shortcomings such as gradient vanishing, gradient explosion, and local minimum points; it also solves the technical problems in the prior art of using CPU and GPU servers to train and carry positioning models during the training process of neural network classification models, which has high costs and large resource consumption.

[0008] The present invention solves the above-mentioned technical problems with the following technical solutions: A pseudo-neural network classification method based on a hypersurface, comprising:

[0009] S1: Randomly select a target to be classified from the classification target sample set, mark the true category of the target to be classified, and generate first electromagnetic wave information after the marking is completed;

[0010] Based on multiple metasurface plates, a multi-level initial neural network model is constructed using a neural network method, and an objective function of the initial neural network model is defined;

[0011] S2: Processing the first electromagnetic wave information using the initial network model to obtain second electromagnetic wave information;

[0012] S3: Processing the second electromagnetic wave information through a decision function to obtain an initial predicted category of the target to be classified;

[0013] S4: Determine whether the initial predicted category is the same as the true category, and obtain a determination result; train the initial neural network model based on the determination result, and perform multi-level synchronous iterative updating on the initial neural network model by reselecting the target to be classified from the classification target sample set, to obtain a result neural network model;

[0014] S5: Use the result neural network model to process the first electromagnetic wave information to obtain third electromagnetic wave information, and process the third electromagnetic wave information through a decision function to obtain a result prediction category.

[0015] The beneficial effects of the present invention are: in terms of algorithm, the present invention uses a non-gradient neural network optimization algorithm to optimize the classification model. This algorithm improves the parameter optimization method from layer-by-layer optimization (i.e., serial optimization) to simultaneous optimization of all layers (i.e., parallel optimization), which greatly improves the model training speed; in terms of hardware, a metasurface classification device is constructed. The metasurface classification device includes multiple metasurface plates for simulating artificial neural networks. The metasurface classification device uses an intelligent metasurface instead of the currently popular CPU and GPU servers to carry the classification model, which has the advantages of low cost, fast speed, and strong real-time performance.

[0016] On the basis of the above technical solution, the present invention can also be improved as follows.

[0017] Furthermore, the above-mentioned initial neural network model is composed of multiple programmable metasurface layers, each layer of programmable metasurface layer corresponds to a metasurface board, and the programmable neurons of the programmable metasurface layer are composed of metasurface neurons on the metasurface board.

[0018] The beneficial effects of adopting this further solution are: in the present invention, multiple metasurface plates are placed vertically in parallel, with the distance between adjacent metasurface plates being the same, and the number of metasurface neurons on each metasurface plate can be the same or different. The present invention controls the electromagnetic wave transmittance of the metasurface neurons through a field-programmable gate array (FPGA), thereby increasing programmable flexibility.

[0019] Furthermore, in the above S1, the objective function of the multi-level initial neural network model is shown in formula (1):

[0020]

[0021] Among them, Wl-1 is the spatial attenuation coefficient matrix of the lth programmable metasurface layer, T l-1 is the complex transmission coefficient matrix of the lth programmable metasurface layer, specifically, l=1,2,3,...,L, L is the total number of programmable metasurface layers; {W l} is the set of spatial attenuation coefficient matrices of all programmable metasurface layers; {T l} is the set of complex transmission coefficient matrices of all programmable metasurface layers, is the loss function, y is the predicted category, is the true category, f(E L ) is the decision function, E L is the electromagnetic wave output matrix of the Lth programmable metasurface layer; E0 is the input of the first programmable metasurface layer, and e is the Hadamard product.

[0022] Furthermore, in the above S1, the expression of the objective function can be changed to be as shown in formula (2):

[0023]

[0024] Among them, α l-1 represents the Lagrange multiplier of the lth programmable metasurface layer, β l-1 is the l-th layer constant matrix, which is used to characterize the l-th layer spatial attenuation coefficient matrix. Since the spatial attenuation coefficient matrix W l-1 It is a parameter that is difficult to adjust. The distance between the two metasurface plates and the propagation medium will affect it. In the present invention, since the distance between the two adjacent metasurface plates is fixed, the constant matrix β is used to adjust the distance between the two adjacent metasurface plates. l-1 Instead of the spatial attenuation coefficient matrix W l-1 , E L-i is the electromagnetic wave output matrix of the Li-th programmable metasurface layer, specifically, i=0, 1, 2, 3, ..., L-1.

[0025] The beneficial effect of adopting the above further scheme is that the present invention uses the spatial attenuation coefficient matrix and the complex transmission coefficient matrix as important parameters of the initial neural network model. Since it is difficult to accurately control the numerical size of the spatial attenuation coefficient matrix artificially, the complex transmission coefficient matrix is ​​selected as the optimized parameter during the optimization process.

[0026] Furthermore, the above S4 is specifically:

[0027] S4.1: Determine whether the initial predicted category is the same as the actual category. If they are different, execute S4.2; if they are the same, execute S4.3;

[0028] S4.2: Synchronously adjust the complex transmission coefficient matrix used to control the intensity of the electromagnetic wave when it propagates to the next layer in each programmable metasurface layer of the initial neural network model. After the adjustment is completed, execute S4.3;

[0029] S4.3: Randomly select a target to be classified from the classification target sample set, repeat S1-S4 until the complex transmission coefficient matrix of each layer of the programmable metasurface layer reaches the convergence condition, and obtain the resulting neural network model.

[0030] The beneficial effect of adopting the above further solution is that the present invention adjusts the transmittance of the metasurface panel, that is, adjusts the complex transmission coefficient matrix, when the predicted category is inaccurate, thereby achieving iterative update of the neural network model.

[0031] Furthermore, the complex transmission coefficient matrix for controlling the intensity of electromagnetic waves when they are transmitted to the next layer in each programmable metasurface layer of the initial neural network model synchronously adjusted in S4.2 is specifically:

[0032] S4.2.1: Obtain an electromagnetic wave output matrix obtained after the first electromagnetic wave information passes through all programmable metasurface layers, and a complex transmission coefficient matrix of each programmable metasurface layer; wherein the electromagnetic wave output matrix is ​​the second electromagnetic wave information;

[0033] S4.2.2: Preset a complex adjustment initial matrix acting on the complex transmission coefficient matrix in each programmable metasurface layer, and adjust the complex adjustment initial matrix using the electromagnetic wave real matrix marked as the real category and the electromagnetic wave output matrix determined to be the predicted category, to obtain a complex adjustment change matrix for each programmable metasurface layer;

[0034] S4.2.3: In each programmable metasurface layer, calculations are performed based on the respective complex adjustment change matrices and the respective complex transmission coefficient matrices to obtain the complex transmission coefficient change matrix of each layer; wherein the complex transmission coefficient change matrix of each layer is shown in formula (3):

[0035]

[0036] in, represents the complex transmission coefficient change matrix of the lth layer, C l-1 represents the complex adjustment change matrix of the lth layer;

[0037] S4.2.4: In each layer of the programmable metasurface, the initial neural network model is updated synchronously at multiple levels based on the respective complex transmission coefficient change matrices, and the complex transmission coefficient matrix of each layer of the editable metasurface is updated to the corresponding complex transmission coefficient change matrix.

[0038] The beneficial effect of adopting the above further scheme is that the present invention reduces the electromagnetic wave intensity of the initial prediction category-related area and enhances the electromagnetic wave intensity of the real category-related area, thereby achieving the effect of adjusting the transmittance of the metasurface plate through parameter adjustment.

[0039] In a second aspect, in order to solve the above technical problems, the present invention further provides a pseudo-neural network classification system based on a hypersurface, comprising:

[0040] Electromagnetic wave information generation module: used to randomly select a target to be classified from the classification target sample set, mark the true category of the target to be classified, and generate the first electromagnetic wave information after the marking is completed;

[0041] Based on multiple metasurface plates, a multi-level initial neural network model is constructed using a neural network method, and an objective function of the initial neural network model is defined;

[0042] An electromagnetic wave information processing module is configured to process the first electromagnetic wave information using the initial network model to obtain second electromagnetic wave information;

[0043] Prediction module: used for processing the second electromagnetic wave information through a decision function to obtain an initial predicted category of the target to be classified;

[0044] Update module: used to determine whether the initial predicted category is the same as the true category and obtain a judgment result; train the initial neural network model based on the judgment result, and perform multi-level synchronous iterative update on the initial neural network model by reselecting the to-be-classified targets from the classification target sample set to obtain a result neural network model;

[0045] Classification module: used to use the result neural network model to process the first electromagnetic wave information to obtain third electromagnetic wave information, and process the third electromagnetic wave information through a decision function to obtain a result prediction category.

[0046] In the third aspect, in order to solve the above-mentioned technical problems, the present invention also provides an electronic device, which includes a memory, a processor, and a computer program stored in the memory and runnable on the processor. When the processor executes the computer program, the hypersurface-based pseudo-neural network classification method of the present application is implemented.

[0047] In a fourth aspect, in order to solve the above-mentioned technical problems, the present invention also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the hypersurface-based pseudo-neural network classification method of the present application is implemented.

[0048] Additional aspects and advantages of the present application will be given in part in the following description, which will become apparent from the following description, or will be learned through practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 A schematic diagram of a flow chart of a pseudo-neural network classification method based on a hypersurface provided by one embodiment of the present invention;

[0050] Figure 2 A schematic diagram of the structure of a pseudo-neural network classification system based on a hypersurface provided by one embodiment of the present invention;

[0051] Figure 3 A schematic structural diagram of an electronic device provided by one embodiment of the present invention;

[0052] Figure 4 A structural diagram of a metasurface classification device provided for one embodiment of the present invention. DETAILED DESCRIPTION

[0053] The principles and features of the present invention are described below. The examples given are only used to explain the present invention and are not used to limit the scope of the present invention.

[0054] The following describes in detail the technical solution of the present invention and how the technical solution of the present invention solves the above-mentioned technical problems using specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The following embodiments of the present invention are described in conjunction with the accompanying drawings.

[0055] The embodiment of the present invention provides a pseudo-neural network classification method based on a hypersurface, such as Figure 4 As shown, a metasurface classification device is first constructed, including:

[0056] Signal receiving and transmitting module: It consists of two parts: one is the electromagnetic wave signal receiver, which is used to receive external electromagnetic wave information; the other is the electromagnetic wave signal transmitter, which is used to send the received electromagnetic wave information to the internal intelligent metasurface module. In this example, the electromagnetic wave intensity information is mainly used.

[0057] Smart metasurface module: Contains five smart metasurface panels with programmable artificial neurons of 4×4, 8×8, 12×12, 16×16, and 20×20, respectively. The transmittance, or complex transmission coefficient, of each programmable artificial neuron can be controlled via FPGA.

[0058] Processor module: Contains a 20×20 small electromagnetic wave intensity sensor array and a small processor. The task of the electromagnetic wave intensity sensor is to receive the electromagnetic wave intensity after being processed by the intelligent metasurface module. The task of the small processor is to analyze the electromagnetic wave intensity and adjust the complex transmission coefficient of the programmable artificial neurons of the metasurface board in the intelligent metasurface module through FPGA.

[0059] On this basis, if Figure 1 As shown, this embodiment provides a pseudo-neural network classification method based on a hypersurface, comprising the following steps:

[0060] S1: Randomly select a target to be classified from the classification target sample set, mark the true category of the target to be classified, and generate first electromagnetic wave information after the marking is completed;

[0061] Based on multiple metasurface plates, a multi-level initial neural network model is constructed using a neural network method, and an objective function of the initial neural network model is defined;

[0062] S2: Processing the first electromagnetic wave information using the initial network model to obtain second electromagnetic wave information;

[0063] S3: Processing the second electromagnetic wave information through a decision function to obtain an initial predicted category of the target to be classified;

[0064] S4: Determine whether the initial predicted category is the same as the true category, and obtain a determination result; train the initial neural network model based on the determination result, and perform multi-level synchronous iterative updating on the initial neural network model by reselecting the target to be classified from the classification target sample set, to obtain a result neural network model;

[0065] S5: Use the result neural network model to process the first electromagnetic wave information to obtain third electromagnetic wave information, and process the third electromagnetic wave information through a decision function to obtain a result prediction category.

[0066] Among them, in terms of algorithm, the present invention uses a non-gradient neural network optimization algorithm to optimize the classification model. This algorithm improves the parameter optimization method from layer-by-layer optimization (i.e., serial optimization) to simultaneous optimization of all layers (i.e., parallel optimization), which greatly improves the model training speed; in terms of hardware, a metasurface classification device is constructed. The metasurface classification device includes multiple metasurface plates for simulating artificial neural networks. The metasurface classification device uses an intelligent metasurface instead of the currently popular CPU and GPU servers to carry the classification model, which has the advantages of low cost, fast speed, and strong real-time performance.

[0067] Optionally, the initial neural network model is composed of multiple programmable metasurface layers, each programmable metasurface layer corresponds to a metasurface board, and the programmable neurons of the programmable metasurface layer are composed of metasurface neurons on the metasurface board.

[0068] In this invention, multiple metasurface panels are placed vertically in parallel, with the distance between adjacent metasurface panels being the same. The number of metasurface neurons on each metasurface panel can be the same or different. This invention uses a field-programmable gate array (FPGA) to control the electromagnetic wave transmittance of the metasurface neurons, improving programmable flexibility.

[0069] Optionally, in this embodiment, the total number of layers L=5.

[0070] In S1, the expression of the multi-level initial neural network model is shown in formula (1):

[0071]

[0072] Among them, W l-1 is the spatial attenuation coefficient matrix of the lth programmable metasurface layer, T l-1 is the complex transmission coefficient matrix of the lth programmable metasurface layer, specifically, l=1, 2, 3, 4, 5, L is the total number of programmable metasurface layers; {W l} is the set of spatial attenuation coefficient matrices of all programmable metasurface layers; {T l} is the set of complex transmission coefficient matrices of all programmable metasurface layers, is the loss function, y is the predicted category, is the true category, f(E L ) is the decision function, E5 is the electromagnetic wave output matrix of the fifth programmable metasurface layer; E0 is the input of the first programmable metasurface layer, and e is the Hadamard product.

[0073] The expression of the initial neural network model is transformed to obtain a first transformation formula, which is shown in formula (2):

[0074]

[0075] Among them, E i is the electromagnetic wave output matrix of the i-th layer;

[0076] Based on the Lagrange multiplier method, the first transformation formula is calculated to obtain a second transformation formula, which is shown in formula (3):

[0077]

[0078] Among them, α represents the Lagrange multiplier, and the subscript of α is its layer number. i-1 is the Lagrange multiplier of the i-th layer;

[0079] The second conversion formula is expanded to obtain a first expanded formula, which is shown in formula (4):

[0080]

[0081] The first expansion formula is transformed to obtain the objective function of the initial neural network model, which is shown in formula (5):

[0082]

[0083] Among them, β l-1 is the l-th layer constant matrix, used to characterize the l-th layer spatial attenuation coefficient matrix, E L-i is the electromagnetic wave output matrix of the Li-th programmable metasurface layer, specifically, i=0, 1, 2, 3, 4.

[0084] In this embodiment, there are two main model parameters in formula (4), namely the spatial attenuation coefficient matrix W and the complex transmission coefficient matrix T. Since it is difficult to precisely control the numerical value of the spatial attenuation coefficient matrix W, it is kept unchanged during the optimization process, and the complex transmission coefficient matrix T is mainly optimized.

[0085] In this embodiment, the complex transmission coefficient matrix T of each layer is randomly set, and then the Lagrange multiplier method is changed. The Lagrange multiplier α of each layer is i-1 It is directly set as a hyperparameter. After fixing the position of the smart metasurface panel, the electromagnetic wave signal is received to obtain the spatial attenuation coefficient matrix W of each layer, i.e. β.

[0086] Optionally, S4 is:

[0087] S4.1: Determine whether the initial predicted category is the same as the actual category. If they are different, execute S4.2; if they are the same, execute S4.3;

[0088] S4.2: Synchronously adjust the complex transmission coefficient matrix used to control the intensity of the electromagnetic wave when it propagates to the next layer in each programmable metasurface layer of the initial neural network model. After the adjustment is completed, execute S4.3;

[0089] S4.3: Randomly select a target to be classified from the classification target sample set, repeat S1-S4 until the complex transmission coefficient matrix of each layer of the programmable metasurface layer reaches the convergence condition, and obtain the resulting neural network model.

[0090] Among them, the present invention adjusts the transmittance of the metasurface plate when the predicted category is inaccurate, that is, adjusts the complex transmission coefficient matrix, thereby achieving iterative update of the neural network model.

[0091] Optionally, the complex transmission coefficient matrix for controlling the intensity of the electromagnetic wave when it propagates to the next layer in each programmable metasurface layer of the initial neural network model is synchronously adjusted in S4.2 as follows:

[0092] S4.2.1: Obtain an electromagnetic wave output matrix obtained after the first electromagnetic wave information passes through all programmable metasurface layers, and a complex transmission coefficient matrix of each programmable metasurface layer; wherein the electromagnetic wave output matrix is ​​the second electromagnetic wave information;

[0093] S4.2.2: Preset a complex adjustment initial matrix acting on the complex transmission coefficient matrix in each programmable metasurface layer, and adjust the complex adjustment initial matrix using the electromagnetic wave real matrix marked as the real category and the electromagnetic wave output matrix determined to be the predicted category, to obtain a complex adjustment change matrix for each programmable metasurface layer;

[0094] S4.2.3: In each programmable metasurface layer, calculations are performed based on the respective complex adjustment change matrices and the respective complex transmission coefficient matrices to obtain the complex transmission coefficient change matrix of each layer; wherein the complex transmission coefficient change matrix of each layer is shown in formula (3):

[0095]

[0096] in, represents the complex transmission coefficient change matrix of the lth layer, C l-1 represents the complex adjustment change matrix of the lth layer;

[0097] S4.2.4: In each layer of the programmable metasurface, the initial neural network model is updated synchronously at multiple levels based on the respective complex transmission coefficient change matrices, and the complex transmission coefficient matrix of each layer of the editable metasurface is updated to the corresponding complex transmission coefficient change matrix.

[0098] The present invention reduces the electromagnetic wave intensity of the initial predicted category-related area and enhances the electromagnetic wave intensity of the real category-related area, thereby adjusting the transmittance of the metasurface plate through parameter adjustment.

[0099] Optionally, in this embodiment, the obtained electromagnetic wave output matrix and the complex transmission coefficient matrix of each layer are as follows; wherein the electromagnetic wave output matrix is The complex transmission coefficient matrix of the i-th layer is

[0100] Wherein, e represents electromagnetic wave information, the superscript of e represents its layer number, the first digit of the subscript of e represents the true category, and the second digit of the subscript of e represents the serial number of the refined true category, t represents the complex transmission coefficient, the superscript of t represents its layer number, the first digit of the subscript of t represents the true category, and the second digit of the subscript of t represents the serial number of the refined true category;

[0101] Among them, f(·) in formula (5) can be expressed as the electromagnetic wave output matrix E L In the area with large value and compactness, if The electromagnetic wave intensity in the region is significantly larger and denser than that in other regions, then the prediction category of the initial neural network model

[0102] When the true category y=1, the initial predicted category When n is not equal to 1, the complex transmission coefficient matrix used to control the intensity of the electromagnetic wave when it is transmitted to the next layer in each programmable metasurface layer of the initial neural network model is synchronously adjusted; specifically,

[0103] In this embodiment, the change matrix C is adjusted by complex numbers. i , enhance the complex transmission coefficient matrix T of each layer of the initial neural network i Parameter information of the area related to the true category 1 And reduce the parameter information of the area related to the initial prediction category n The complex transmission coefficient change matrix of each layer is obtained, and the increase and decrease amplitudes of the parameter information in the corresponding areas of each layer are not the same.

[0104] In this embodiment, due to the complex adjustment change matrix C i is the complex transmission coefficient matrix T i Independent of each other, so the complex transmission coefficient change matrix of each layer Calculations can be performed simultaneously, that is, the parameters of the entire pseudo-neural network can be calculated in parallel, which has a qualitative improvement in training speed compared with the traditional serial calculation of gradient back propagation; another advantage of this optimization algorithm is that the number of programmable artificial neurons in each layer of the metasurface plate can be different, while the traditional metasurface device based on the reverse gradient propagation algorithm generally needs to keep the number of programmable artificial neurons in each layer consistent.

[0105] Based on Figure 1 The present invention also provides a pseudo-neural network classification system based on a hypersurface, such as Figure 2 As shown in , the hypersurface-based pseudo-neural network classification system may include:

[0106] Electromagnetic wave information generation module: used to randomly select a target to be classified from the classification target sample set, mark the true category of the target to be classified, and generate the first electromagnetic wave information after the marking is completed;

[0107] Based on multiple metasurface plates, a multi-level initial neural network model is constructed using a neural network method, and an objective function of the initial neural network model is defined;

[0108] An electromagnetic wave information processing module is configured to process the first electromagnetic wave information using the initial network model to obtain second electromagnetic wave information;

[0109] Prediction module: used for processing the second electromagnetic wave information through a decision function to obtain an initial predicted category of the target to be classified;

[0110] Update module: used to determine whether the initial predicted category is the same as the true category and obtain a judgment result; train the initial neural network model based on the judgment result, and perform multi-level synchronous iterative update on the initial neural network model by reselecting the to-be-classified targets from the classification target sample set to obtain a result neural network model;

[0111] Classification module: used to use the result neural network model to process the first electromagnetic wave information to obtain third electromagnetic wave information, and process the third electromagnetic wave information through a decision function to obtain a result prediction category.

[0112] The hypersurface-based quasi-neural network classification system of the embodiment of the present invention can execute the hypersurface-based quasi-neural network classification method provided by the embodiment of the present invention, and the implementation principle is similar. The actions performed by each module and unit in the hypersurface-based quasi-neural network classification system in each embodiment of the present invention correspond to the steps in the hypersurface-based quasi-neural network classification method in each embodiment of the present invention. For the detailed functional description of each module of the hypersurface-based quasi-neural network classification system, please refer to the description of the corresponding hypersurface-based quasi-neural network classification method shown in the previous text, which will not be repeated here.

[0113] Among them, the above-mentioned pseudo-neural network classification system based on the hypersurface can be a computer program (including program code) running in a computer device, for example, the pseudo-neural network classification system based on the hypersurface is an application software; the system can be used to execute the corresponding steps in the method provided in the embodiment of the present invention.

[0114] In some embodiments, the metasurface-based pseudo-neural network classification system provided by the embodiments of the present invention can be implemented in a combination of software and hardware. As an example, the metasurface-based pseudo-neural network classification system provided by the embodiments of the present invention can be a processor in the form of a hardware decoding processor, which is programmed to execute the metasurface-based pseudo-neural network classification method provided by the embodiments of the present invention. For example, the processor in the form of a hardware decoding processor can adopt one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs) or other electronic components.

[0115] In other embodiments, the pseudo-neural network classification system based on the hypersurface provided by the embodiments of the present invention can be implemented in software. Figure 2 A pseudo-neural network classification system based on a hypersurface stored in a memory is shown, which can be software in the form of programs and plug-ins, and includes a series of modules, including a signal acquisition module, a neural network model building module, an analysis module, an update module, and a classification module, for implementing the method provided in an embodiment of the present invention.

[0116] The modules involved in the embodiments of the present invention may be implemented in software or hardware, wherein the name of a module does not necessarily limit the module itself.

[0117] Based on the same principle as the method shown in the embodiments of the present invention, an electronic device is also provided in the embodiments of the present invention, which may include but is not limited to: a processor and a memory; the memory is used to store computer programs; the processor is used to execute the method shown in any embodiment of the present invention by calling the computer program.

[0118] In an alternative embodiment, an electronic device is provided, such as Figure 3 As shown, Figure 3 The electronic device shown includes a processor and a memory. The processor and the memory are connected, for example, via a bus. Optionally, the electronic device may further include a transceiver, which can be used for data exchange between the electronic device and other electronic devices, such as data transmission and / or data reception. It should be noted that in actual applications, there is not limited to one transceiver, and the structure of the electronic device does not constitute a limitation on the embodiments of the present invention.

[0119] The processor can be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the present disclosure. Processor 4001 can also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc.

[0120] A bus may include a path for transmitting information between the above components. The bus may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus. The bus may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 3 Only one thick line is used in the diagram, but this does not mean that there is only one bus or one type of bus.

[0121] The memory may be a ROM (Read Only Memory) or other type of static storage device that can store static information and instructions, a RAM (Random Access Memory) or other type of dynamic storage device that can store information and instructions, or an EEPROM (Electrically Erasable Programmable Read Only Memory), a CD-ROM (Compact Disc Read Only Memory) or other optical disk storage, optical disc storage (including compact disc, laser disc, optical disc, digital versatile disc, Blu-ray disc, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited to these.

[0122] The memory is used to store application code (computer program) for executing the solution of the present invention, and the processor controls the execution of the application code. The processor is used to execute the application code stored in the memory to implement the content shown in the above method embodiment.

[0123] Among them, the electronic device can also be a terminal device, Figure 3 The electronic device shown is only an example and should not limit the functions and scope of use of the embodiments of the present invention.

[0124] An embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer-readable storage medium is run on a computer, the computer can execute the corresponding contents of the aforementioned method embodiment.

[0125] According to another aspect of the present invention, a computer program product or computer program is provided, comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the methods provided in the various implementations described above.

[0126] Computer program code for performing the operations of the present invention may be written in one or more programming languages, or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on the remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., through the Internet using an Internet service provider).

[0127] It should be understood that the flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the methods and computer program products according to various embodiments of the present invention. In this regard, each box in the flowchart or block diagram can represent a module, program segment, or a part of code, and the module, program segment, or a part of code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order than the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of the boxes in the block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or can be implemented using a combination of dedicated hardware and computer instructions.

[0128] The computer-readable storage medium provided by the embodiments of the present invention may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or component, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to, an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present invention, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, device or device.

[0129] The computer-readable storage medium carries one or more programs. When the one or more programs are executed by the electronic device, the electronic device executes the method shown in the above embodiment.

[0130] The above description is merely a preferred embodiment of the present invention and an illustration of the technical principles employed. Those skilled in the art should understand that the scope of disclosure involved in the present invention is not limited to the technical solutions formed by the specific combination of the above-mentioned technical features, but also includes other technical solutions formed by any combination of the above-mentioned technical features or their equivalents without departing from the above-mentioned disclosed concepts. For example, a technical solution formed by replacing the above-mentioned features with (but not limited to) technical features with similar functions disclosed in the present invention.

Claims

1. A pseudo-neural network classification method based on hypersurface, characterized in that: include: S1: Randomly select a target to be classified from the classification target sample set, mark the true category of the target to be classified, and generate first electromagnetic wave information after the marking is completed; Based on multiple metasurface plates, a multi-level initial neural network model is constructed using a neural network method, and an objective function of the initial neural network model is defined; S2: Processing the first electromagnetic wave information using an initial neural network model to obtain second electromagnetic wave information; S3: Processing the second electromagnetic wave information through a decision function to obtain an initial predicted category of the target to be classified; S4: Determine whether the initial predicted category is the same as the true category, and obtain a determination result; The initial neural network model is trained based on the judgment result, and the initial neural network model is iteratively updated with multi-level synchronization by reselecting the target to be classified from the classification target sample set to obtain a result neural network model; Obtaining an electromagnetic wave output matrix obtained after the first electromagnetic wave information passes through all programmable metasurface layers, and a complex transmission coefficient matrix of each programmable metasurface layer; wherein the electromagnetic wave output matrix is ​​the second electromagnetic wave information; Presetting a complex adjustment initial matrix acting on the complex transmission coefficient matrix in each layer of the programmable metasurface layer, and adjusting the complex adjustment initial matrix by using the electromagnetic wave real matrix marked as the real category and the electromagnetic wave output matrix judged as the predicted category, to obtain a complex adjustment change matrix for each layer of the programmable metasurface layer; In each programmable metasurface layer, calculations are performed based on the respective complex adjustment change matrix and the respective complex transmission coefficient matrix to obtain the complex transmission coefficient change matrix of each layer; wherein the complex transmission coefficient change matrix of each layer is shown in formula (3): in, represents the complex transmission coefficient change matrix of the lth layer, T l-1 is the complex transmission coefficient matrix of the lth programmable metasurface layer, C l-1 represents the complex adjustment change matrix of the lth layer; In each programmable metasurface layer, the initial neural network model is synchronously updated at multiple levels based on the respective complex transmission coefficient change matrices, and the complex transmission coefficient matrix of each programmable metasurface layer is updated to the corresponding complex transmission coefficient change matrix; S5: Use the result neural network model to process the first electromagnetic wave information to obtain third electromagnetic wave information, and process the third electromagnetic wave information through a decision function to obtain a result prediction category.

2. The method of classification based on a pseudo-neural network of a hypersurface according to claim 1, characterized in that: The initial neural network model is composed of multiple programmable metasurface layers, each programmable metasurface layer corresponds to a metasurface board, and the programmable neurons of the programmable metasurface layer are composed of metasurface neurons on the metasurface board.

3. The method of classification based on a pseudo-neural network of a hypersurface according to claim 2, characterized in that: In S1, the objective function of the multi-level initial neural network model is shown in formula (1): Among them, W l-1 is the spatial attenuation coefficient matrix of the lth programmable metasurface layer, T l-1 is the complex transmission coefficient matrix of the lth programmable metasurface layer, specifically, l=1,2,3,...,L, L is the total number of programmable metasurface layers; {W l } is the set of spatial attenuation coefficient matrices of all programmable metasurface layers; {T l } is the set of complex transmission coefficient matrices of all programmable metasurface layers, is the loss function, y is the predicted category, is the true category, f(.) is the decision function, E L is the electromagnetic wave output matrix of the Lth programmable metasurface layer; E0 is the input of the first programmable metasurface layer, and ⊙ is the Hadamard product.

4. The method of classification based on a pseudo-neural network of a hypersurface according to claim 3, characterized in that: In S1, the expression of the objective function can be changed to the following formula (2): Among them, α l-1 represents the Lagrange multiplier of the lth programmable metasurface layer, β l-1 is the l-th layer constant matrix, used to characterize the l-th layer spatial attenuation coefficient matrix; E L-i is the electromagnetic wave output matrix of the Li-th programmable metasurface layer, specifically, i=0, 1, 2, 3, ..., L-1.

5. A pseudo-neural network classification system based on a hypersurface, characterized in that: include: Electromagnetic wave information generation module: used to randomly select a target to be classified from the classification target sample set, mark the true category of the target to be classified, and generate the first electromagnetic wave information after the marking is completed; Based on multiple metasurface plates, a multi-level initial neural network model is constructed using a neural network method, and an objective function of the initial neural network model is defined; Electromagnetic wave information processing module: used to process the first electromagnetic wave information using an initial neural network model to obtain second electromagnetic wave information; Prediction module: used for processing the second electromagnetic wave information through a decision function to obtain an initial predicted category of the target to be classified; Update module: used to determine whether the initial predicted category is the same as the true category and obtain a judgment result; The initial neural network model is trained based on the judgment result, and the initial neural network model is iteratively updated with multi-level synchronization by reselecting the target to be classified from the classification target sample set to obtain a result neural network model; The update module is specifically: Obtaining an electromagnetic wave output matrix obtained after the first electromagnetic wave information passes through all programmable metasurface layers, and a complex transmission coefficient matrix of each programmable metasurface layer; wherein the electromagnetic wave output matrix is ​​the second electromagnetic wave information; Presetting a complex adjustment initial matrix acting on the complex transmission coefficient matrix in each layer of the programmable metasurface layer, and adjusting the complex adjustment initial matrix by using the electromagnetic wave real matrix marked as the real category and the electromagnetic wave output matrix judged as the predicted category, to obtain a complex adjustment change matrix for each layer of the programmable metasurface layer; In each programmable metasurface layer, calculations are performed based on the respective complex adjustment change matrix and the respective complex transmission coefficient matrix to obtain the complex transmission coefficient change matrix of each layer; wherein the complex transmission coefficient change matrix of each layer is shown in formula (3): in, represents the complex transmission coefficient change matrix of the lth layer, T l-1 is the complex transmission coefficient matrix of the lth programmable metasurface layer, C l-1 represents the complex adjustment change matrix of the lth layer; In each programmable metasurface layer, the initial neural network model is synchronously updated at multiple levels based on the respective complex transmission coefficient change matrices, and the complex transmission coefficient matrix of each programmable metasurface layer is updated to the corresponding complex transmission coefficient change matrix; Classification module: used to use the result neural network model to process the first electromagnetic wave information to obtain third electromagnetic wave information, and process the third electromagnetic wave information through a decision function to obtain a result prediction category.

6. An electronic device, characterized in that: The method comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method according to any one of claims 1 to 4 when executing the computer program.

7. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 4 is implemented.

Images