CN114492789B - A method and device for constructing a neural network model of data samples - Google Patents

Abstract

The invention discloses a neural network model construction method and device for a data sample, comprising the following steps: the method comprises the steps that a preset neural network model is obtained, the preset neural network model comprises an input layer, a projection network and a learning network, the input layer is used for obtaining input sample signals and mapping the input sample signals into neuron outputs, the projection network is used for reducing dimensions of high-dimensional information output by the input layer, and the learning network is used for learning the information subjected to dimension reduction by the projection network and outputting a learning result; determining a connection weight of each layer of the preset neural network model based on a target task corresponding to the input sample signal; training the preset neural network model according to the connection weight of each layer of the preset neural network model to obtain a target neural network model. According to the invention, the projection network is constructed, so that the projection study on the data sample can be facilitated, the information can be rapidly transmitted in the network, and the efficient study on the sample can be realized.

Description

A method and device for constructing a neural network model of data samples

Technical Field

The present invention relates to the field of data processing technology, and in particular to a method and device for constructing a neural network model of data samples.

Background technique

Artificial neural network is an important research direction in the field of artificial intelligence. It is a simulation of the human brain neural network from the perspective of information processing. Artificial neural network can arbitrarily approximate any nonlinear function, thereby realizing the distinction between different signal patterns. Artificial neural network has been widely used in many fields such as signal processing, pattern recognition, automatic control, artificial intelligence, etc. There are many models of artificial neural network, such as BP model, Hopfield model, extreme learning machine, deep learning model, etc. These models usually consist of input layer, hidden layer and output layer. Information is processed and transmitted between layers. In each layer, neurons perform nonlinear processing on input signals through activation functions to realize feature extraction. Each layer consists of multiple neuron nodes. Neuron nodes in adjacent layers are connected to each other, but neuron nodes in the same layer and across layers are usually not connected. The connections between neurons have weights, and the learning process is the process of constantly modifying the weights of the connections between neurons. BP neural network adjusts weights by forward transmission of information and reverse transmission of errors.

Based on the BP neural network, deep learning has reactivated the research of neural networks. It has pushed artificial neural networks from shallow layers to deep layers, opening a new era of deep neural networks (DNN). Deep neural networks have more hidden layers. The first hidden layer extracts basic features from the original data, and the subsequent hidden layers combine the basic features into a higher-order abstract feature. Deep learning can automatically extract the features required for classification without human participation, and has achieved great success in speech recognition, image processing, pattern recognition and other fields.

However, deep learning, like previous neural networks, requires a lot of time to complete network training. Although it is the era of big data, expert knowledge samples are scarce. On the other hand, computing power is often limited in practical applications. Limited computing resources restrict the performance of DNN. At the same time, saving computing resources can save energy and bring benefits. Therefore, it is necessary to study artificial neural networks with strong generalization ability and fast learning speed.

Summary of the invention

In order to solve the above problems, the present invention provides a method and device for constructing a neural network model of data samples, which realizes the rapid transmission of information within the network and improves the learning efficiency of sample data.

In order to achieve the above object, the present invention provides the following technical solutions:

A method for constructing a neural network model of a data sample, comprising:

Obtain a preset neural network model, the preset neural network model includes an input layer, a projection network and a learning network, the input layer is used to obtain an input sample signal and map the input sample signal to a neuron output, the projection network is used to reduce the dimension of the high-dimensional information output by the input layer, and the learning network is used to learn the information after the dimension reduction of the projection network and output the learning result;

Determining a connection weight with each layer of the preset neural network model based on a target task corresponding to the input sample signal;

The preset neural network model is trained according to the connection weights of each layer of the preset neural network model to obtain a target neural network model.

Optionally, it also includes:

Obtain target training samples;

Based on the length of the target training sample, the number of neurons in each layer of the preset neural network model is determined.

Optionally, the input layer of the preset neural network model is composed of a single-layer neural network; the projection network is composed of several layers of neural networks, the several layers of neural networks of the projection network have specific connection weights, and each layer of the neural network is composed of several neurons, and the projection network neurons are only connected to a limited number of neurons in the previous layer; the learning network is composed of a shallow neural network, the first layer of the neural network is connected to the first layer of the projection network, the shallow neural network is a fully connected network, and the neurons between different layers of the shallow neural network are connected through weights.

Optionally, determining the connection weights with each layer of the preset neural network model based on the target task corresponding to the input sample signal includes:

Acquire a target training sample corresponding to the input sample signal;

Inputting the target training sample into the input layer, and obtaining the output of the neurons in the input layer;

Inputting the output of the input layer neurons into the projection network to obtain an output signal of the projection network;

Inputting the output signal of the projection network into the learning network, and obtaining the learning result of the learning network;

Based on the comparison value between the target task and the learning result, the connection weights of the neurons are adjusted to obtain the connection weights of each layer of the preset neural network model.

Optionally, the method further comprises:

When determining the output information of each layer of the preset neural network model, determining the activation function of each layer of the neural network model, and performing normalization processing on the input signal of each layer;

The output signal of each layer is determined based on the normalized input signal and the corresponding activation function.

A device for constructing a neural network model of a data sample, comprising:

An acquisition unit is used to acquire a preset neural network model, wherein the preset neural network model includes an input layer, a projection network and a learning network, wherein the input layer is used to acquire an input sample signal and map the input sample signal to a neuron output, the projection network is used to reduce the dimension of the high-dimensional information output by the input layer, and the learning network is used to learn the information after the dimension reduction of the projection network and output the learning result;

A determination unit, configured to determine a connection weight with each layer of the preset neural network model based on a target task corresponding to the input sample signal;

A training unit is used to train the preset neural network model according to the connection weights of each layer of the preset neural network model to obtain a target neural network model.

Optionally, it also includes:

A neuron number determination unit, used to obtain target training samples;

Based on the length of the target training sample, the number of neurons in each layer of the preset neural network model is determined.

Optionally, the input layer of the preset neural network model is composed of a single-layer neural network; the projection network is composed of several layers of neural networks, the several layers of neural networks of the projection network have specific connection weights, and each layer of the neural network is composed of several neurons, and the projection network neurons are only connected to a limited number of neurons in the previous layer; the learning network is composed of a shallow neural network, the first layer of the neural network is connected to the first layer of the projection network, the shallow neural network is a fully connected network, and the neurons between different layers of the shallow neural network are connected through weights.

Optionally, the determining unit is specifically configured to:

Acquire a target training sample corresponding to the input sample signal;

Inputting the target training sample into the input layer, and obtaining the output of the neurons in the input layer;

Inputting the output of the input layer neurons into the projection network to obtain an output signal of the projection network;

Inputting the output signal of the projection network into the learning network, and obtaining the learning result of the learning network;

Based on the comparison value between the target task and the learning result, the connection weights of the neurons are adjusted to obtain the connection weights of each layer of the preset neural network model.

Optionally, the device further comprises:

An output signal determination unit, used to determine the activation function of each layer of the neural network model when determining the output information of each layer of the preset neural network model, and to perform normalization processing on the input signal of each layer;

The output signal of each layer is determined based on the normalized input signal and the corresponding activation function.

Compared with the prior art, the present invention provides a method and device for constructing a neural network model of a data sample, including: obtaining a preset neural network model, the preset neural network model includes an input layer, a projection network and a learning network, the input layer is used to obtain an input sample signal and map the input sample signal to a neuron output, the projection network is used to reduce the dimension of the high-dimensional information output by the input layer, and the learning network is used to learn the information after the dimension reduction of the projection network, and output the learning result; based on the target task corresponding to the input sample signal, determine the connection weight with each layer of the preset neural network model; according to the connection weight of each layer of the preset neural network model, train the preset neural network model to obtain the target neural network model. The present invention can facilitate the projection learning of data samples by constructing a projection network, realize the rapid transmission of information within the network, and then realize efficient learning of samples.

BRIEF DESCRIPTION OF THE DRAWINGS

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

FIG1 is a schematic diagram of a flow chart of a method for constructing a neural network model of a data sample provided by an embodiment of the present invention;

FIG2 is a schematic diagram of the structure of a projection learning model provided by an embodiment of the present invention;

FIG3 is a schematic diagram of the structure of a device for constructing a neural network model of a data sample provided in an embodiment of the present invention.

Detailed ways

The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

The terms "first" and "second" and the like in the specification and claims of the present invention and the above drawings are used to distinguish different objects, rather than to describe a specific order. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device including a series of steps or units is not limited to the listed steps or units, but may include steps or units that are not listed.

In an embodiment of the present invention, a method for constructing a neural network model of a data sample is provided. Referring to FIG1 , the method may include the following steps:

S101, obtaining a preset neural network model.

Among them, the preset neural network model includes an input layer, a projection network and a learning network. The input layer is used to obtain input sample signals and map the input sample signals to neuron outputs. The projection network is used to reduce the dimension of the high-dimensional information output by the input layer. The learning network is used to learn the information after the dimension reduction of the projection network and output the learning results.

Specifically, the input layer of the preset neural network model is composed of a single-layer neural network; the projection network is composed of several layers of neural networks, the several layers of neural networks of the projection network have specific connection weights, and each layer of the neural network is composed of several neurons, and the projection network neurons are only connected to a limited number of neurons in the previous layer; the learning network is composed of a shallow neural network, the first layer of the neural network is connected to the first layer of the projection network, the shallow neural network is a fully connected network, and the neurons between different layers of the shallow neural network are connected through weights.

S102: Determine a connection weight with each layer of the preset neural network model based on a target task corresponding to the input sample signal.

The target task may be any target task that is desired to be achieved during the training process based on the target training sample corresponding to the input sample signal, and there is no limitation on this.

Each layer in the preset neural network model can be processed based on the output signal of the previous layer to obtain the final output signal, and then the output signal is compared with the expected output signal of the target task, and the connection weights of each layer are adjusted based on the comparison result to obtain the connection weights of each layer corresponding to the target task.

In one implementation of the embodiment of the present application, determining the connection weights with each layer of the preset neural network model based on the target task corresponding to the input sample signal includes:

Acquire a target training sample corresponding to the input sample signal;

Inputting the target training sample into the input layer, and obtaining the output of the neurons in the input layer;

Inputting the output of the input layer neurons into the projection network to obtain an output signal of the projection network;

Inputting the output signal of the projection network into the learning network, and obtaining the learning result of the learning network;

Based on the comparison value between the target task and the learning result, the connection weights of the neurons are adjusted to obtain the connection weights of each layer of the preset neural network model.

Furthermore, the method further comprises:

When determining the output information of each layer of the preset neural network model, determining the activation function of each layer of the neural network model, and performing normalization processing on the input signal of each layer;

The output signal of each layer is determined based on the normalized input signal and the corresponding activation function.

S103, training the preset neural network model according to the connection weights of each layer of the preset neural network model to obtain a target neural network model.

After obtaining the connection weights of each layer, the associated weight values between the preset neural network models are adjusted and trained to obtain the target neural network model. The target neural network can be used to process data in subsequent application scenarios similar to the target task.

The embodiment of the present application provides a method for constructing a neural network model of a data sample, including: obtaining a preset neural network model, the preset neural network model includes an input layer, a projection network and a learning network, the input layer is used to obtain an input sample signal and map the input sample signal to a neuron output, the projection network is used to reduce the dimension of the high-dimensional information output by the input layer, and the learning network is used to learn the information after the dimension reduction of the projection network, and output the learning result; based on the target task corresponding to the input sample signal, determine the connection weight with each layer of the preset neural network model; train the preset neural network model according to the connection weight of each layer of the preset neural network model to obtain a target neural network model. The present invention can facilitate the projection learning of data samples by constructing a projection network, realize the rapid transmission of information within the network, and then realize efficient learning of samples.

The preset neural network model provided in the embodiment of the present application is a projection learning model, including an input layer, a projection network and a learning network.

The input layer is composed of a single-layer neural network, which is used to receive input sample signals and map the input sample signals to outputs of neurons; the single-layer neural network is composed of a number of neurons, and there is no connection between the neurons in this layer.

The projection network is composed of several layers of neural networks, which are used to quickly project the output of the input layer to the learning network to achieve dimensionality reduction of the high-dimensional information of the input layer; the several layers of neural networks in the projection network have specific connection weights, which do not need to be modified during the learning process, and each layer of the neural network is composed of several neurons; the projection network neurons have a local field of view, that is, the projection network neurons are only connected to a limited number of neurons in the previous layer; the local field of view can be dynamically adjusted.

The learning network consists of a shallow neural network, which is used to learn and memorize the information after dimensionality reduction of the projection network, and output the results of learning and memorization; the first layer of the neural network of the learning network is connected to the last layer of the neural network of the projection network; the shallow neural network refers to a neural network with a number of layers generally not exceeding five; the shallow neural network is a fully connected network; the neurons between different layers of the shallow neural network are connected through weights, and the learning process is the process of modifying the connection weights of the shallow neural network; the last layer of the shallow neural network outputs the learning results.

The following describes the embodiment of the present application by taking the process of learning one-dimensional data samples. For example, taking CWRU data as an example, CWRU data is data measured by the Bearing Data Center of Case Western Reserve University in the United States. It should be noted that processing can also be performed based on other data.

First, transform the CWRU bearing data in the time domain into the frequency domain to establish a training sample library. According to the length of the sample, construct a projection learning network model; the sample length selected here is 2048, and the constructed projection learning model is shown in Figure 2, where n = 2048, p = d = 16. In this way, the designed projection learning model has an input layer composed of 2048 neurons, a projection network composed of two layers of neural networks, the first projection layer has 2048 neurons, and the second projection layer has 128 neurons. The learning network consists of three layers of neural networks, and the number of neurons in the first learning layer is the same as the number of neurons in the second projection layer, that is, the first learning layer has 128 neurons, the second learning layer has 32 neurons, and the number of neurons in the third learning layer is determined by the number of categories of the samples to be learned.

The input sample data is mapped to the output of the neuron through the input layer and used as the input signal of the projection network. The output of each input layer neuron can be expressed as:

_yi ＝f( _xi ) (1)

Where _xi is the normalized sample signal, f(·) represents the activation function, and the input layer uses the optimized Tanh function as the activation function, which is expressed as Each neuron in the first projection network receives the output signal of 16 input layer neurons, and its output is:

In the formula It is called the action coefficient, and is selected here/> The activation function uses the sigmoid function, and its expression is

Each neuron in the second projection network receives the output signals of 16 neurons in the first projection network, and its output is:

In the formula Also called the action coefficient, also used here/> The selection of activation function is the same as that of the first projection network.

The learning network consists of three layers of neural networks. The first layer of learning network is connected to the second layer of projection network. The neurons of the first layer of learning network receive the output signals of the neurons of the second layer of projection network. The neurons between the three learning layers are connected by weights. The learning process is the process of modifying the connection weights of the three layers of neural networks. The results of learning training are output by the third layer of learning network. It can be seen that the projection learning model constructed by the test of the present invention consists of 6 layers of networks.

The above-mentioned projection learning model was tested using CWRU bearing data and compared with the deep neural network model. For ease of comparison, the compared neural network was designed to be 5 layers, and the compared deep neural network model was 2048-414-114-45-N, where N represents the number of neurons in the output layer. The comparison results are shown in Table 1. It can be seen from the table that the generalization ability of the method of the present invention is much improved than that of the original deep neural network model. The recognition rate of the method of the present invention in the first and third tests reached 100%, while the original method was only 85.31% and 90.45%. The method of the present invention increased the recognition rate by at least 9.55%. The learning time of the two methods is shown in Table 2. It can be seen from the table that the training time of the method of the present invention is greatly reduced, less than 2% of the training time of the original neural network model. This shows that the method of the present invention not only has a strong generalization ability, but also has an efficient learning speed.

Table 1. Comparison results of recognition rate tests

In Table 1, F represents a faulty sample, and N represents a normal sample.

Table 2 Comparison results of learning time test

The method for constructing a neural network model of a data sample provided in the embodiment of the present application can realize rapid learning of data samples, save a lot of time spent on training traditional neural network models, and save computing resources. Compared with traditional neural network models, the method of the present invention has better generalization ability, faster learning and training speed, and has broad industrial application prospects.

In another embodiment of the present application, a device for constructing a neural network model of a data sample is also provided, referring to FIG3 , comprising:

An acquisition unit 301 is used to acquire a preset neural network model, wherein the preset neural network model includes an input layer, a projection network and a learning network, wherein the input layer is used to acquire an input sample signal and map the input sample signal to a neuron output, the projection network is used to reduce the dimension of the high-dimensional information output by the input layer, and the learning network is used to learn the information after the dimension reduction of the projection network and output the learning result;

A determination unit 302, configured to determine a connection weight with each layer of the preset neural network model based on a target task corresponding to the input sample signal;

The training unit 303 is used to train the preset neural network model according to the connection weights of each layer of the preset neural network model to obtain a target neural network model.

Optionally, it also includes:

A neuron number determination unit, used to obtain target training samples;

Based on the length of the target training sample, the number of neurons in each layer of the preset neural network model is determined.

Optionally, the input layer of the preset neural network model is composed of a single-layer neural network; the projection network is composed of several layers of neural networks, the several layers of neural networks of the projection network have specific connection weights, and each layer of the neural network is composed of several neurons, and the projection network neurons are only connected to a limited number of neurons in the previous layer; the learning network is composed of a shallow neural network, the first layer of the neural network is connected to the first layer of the projection network, the shallow neural network is a fully connected network, and the neurons between different layers of the shallow neural network are connected through weights.

Optionally, the determining unit is specifically configured to:

Acquire a target training sample corresponding to the input sample signal;

Inputting the target training sample into the input layer, and obtaining the output of the neurons in the input layer;

Inputting the output of the input layer neurons into the projection network to obtain an output signal of the projection network;

Inputting the output signal of the projection network into the learning network, and obtaining the learning result of the learning network;

Based on the comparison value between the target task and the learning result, the connection weights of the neurons are adjusted to obtain the connection weights of each layer of the preset neural network model.

Optionally, the device further comprises:

An output signal determination unit, used to determine the activation function of each layer of the neural network model when determining the output information of each layer of the preset neural network model, and to perform normalization processing on the input signal of each layer;

The output signal of each layer is determined based on the normalized input signal and the corresponding activation function.

The embodiment of the present application provides a neural network model construction device for data samples, including: obtaining a preset neural network model, the preset neural network model includes an input layer, a projection network and a learning network, the input layer is used to obtain an input sample signal and map the input sample signal to a neuron output, the projection network is used to reduce the dimension of the high-dimensional information output by the input layer, and the learning network is used to learn the information after the dimension reduction of the projection network, and output the learning result; based on the target task corresponding to the input sample signal, determine the connection weight with each layer of the preset neural network model; according to the connection weight of each layer of the preset neural network model, train the preset neural network model to obtain the target neural network model. The present invention can facilitate the projection learning of data samples by constructing a projection network, realize the rapid transmission of information within the network, and then realize efficient learning of samples.

Based on the aforementioned embodiments, embodiments of the present application provide a computer-readable storage medium, which stores one or more programs, and the one or more programs can be executed by one or more processors to implement the steps of the method for constructing a neural network model of a data sample as described in any of the above items.

An embodiment of the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of a method for constructing a neural network model of a data sample when executing the program.

It should be noted that the processor or CPU may be at least one of an Application Specific Integrated Circuit (ASIC), a Digital Signal Processor (DSP), a Digital Signal Processing Device (DSPD), a Programmable Logic Device (PLD), a Field Programmable Gate Array (FPGA), a Central Processing Unit (CPU), a controller, a microcontroller, and a microprocessor. It is understandable that the electronic device that implements the above-mentioned processor function may also be other, and the embodiments of the present application are not specifically limited.

It should be noted that the above-mentioned computer storage medium/memory can be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), a magnetic random access memory (FRAM), a flash memory (Flash Memory), a magnetic surface memory, an optical disc, or a compact disc read-only memory (CD-ROM) and other memories; it can also be various terminals including one or any combination of the above-mentioned memories, such as mobile phones, computers, tablet devices, personal digital assistants, etc.

In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored, or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

In addition, all functional units in the embodiments of the present application can be integrated into one processing module, or each unit can be a separate unit, or two or more units can be integrated into one unit; the above integrated unit can be implemented in the form of hardware or in the form of hardware plus software functional units. A person of ordinary skill in the art can understand that all or part of the steps of implementing the above method embodiments can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium, which, when executed, executes the steps of the above method embodiments; and the aforementioned storage medium includes: mobile storage devices, read-only memory (ROM), random access memory (RAM), disks or optical disks, and other media that can store program codes.

The methods disclosed in several method embodiments provided in this application can be arbitrarily combined without conflict to obtain new method embodiments.

The features disclosed in several product embodiments provided in this application can be arbitrarily combined without conflict to obtain new product embodiments.

The features disclosed in several method or device embodiments provided in this application can be arbitrarily combined without conflict to obtain new method embodiments or device embodiments.

The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

The above description of the disclosed embodiments enables those skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims (2)

1. A method for constructing a neural network model of a data sample, characterized by comprising: Obtain a preset neural network model, the preset neural network model includes an input layer, a projection network and a learning network, the input layer is used to obtain an input sample signal and map the input sample signal to a neuron output, the projection network is used to reduce the dimension of the high-dimensional information output by the input layer, and the learning network is used to learn the information after the dimension reduction of the projection network and output the learning result; wherein the number of neurons in each layer of the preset neural network model is determined based on the length of the target training sample; the input layer of the preset neural network model is composed of a single-layer neural network; the projection network is composed of several layers of neural networks, the several layers of neural networks of the projection network have specific connection weights, and each layer of neural network is composed of several neurons, and the projection network neurons are only connected to a limited number of neurons in the previous layer; the learning network is composed of a shallow neural network, the first layer of the neural network is connected to the first layer of the neural network of the projection network, the shallow neural network is a fully connected network, and the neurons between different layers of the shallow neural network are connected by weights; it also includes: when determining the output information of each layer of the preset neural network model, determining the activation function of each layer of the neural network model, and normalizing the input signal of each layer; determining the output signal of each layer according to the normalized input signal and the corresponding activation function; Based on the target task corresponding to the input sample signal, determine the connection weights with each layer of the preset neural network model, wherein the determination of the connection weights with each layer of the preset neural network model based on the target task corresponding to the input sample signal includes: obtaining a target training sample corresponding to the input sample signal; inputting the target training sample into the input layer, and obtaining the output of the input layer neuron; inputting the output of the input layer neuron into the projection network, and obtaining the output signal of the projection network; inputting the output signal of the projection network into the learning network, and obtaining the learning result of the learning network; based on the comparison value of the target task and the learning result, adjusting the connection weights of the neurons to obtain the connection weights with each layer of the preset neural network model; The preset neural network model is trained according to the connection weights of each layer of the preset neural network model to obtain a target neural network model. 2. A device for constructing a neural network model of a data sample, characterized by comprising: An acquisition unit is used to acquire a preset neural network model, wherein the preset neural network model includes an input layer, a projection network and a learning network, wherein the input layer is used to acquire an input sample signal and map the input sample signal to a neuron output, the projection network is used to reduce the dimension of the high-dimensional information output by the input layer, and the learning network is used to learn the information after the dimension reduction of the projection network and output the learning result; wherein the number of neurons in each layer of the preset neural network model is determined based on the length of the target training sample; the input layer of the preset neural network model is composed of a single-layer neural network; the projection network is composed of several layers of neural networks, and the several layers of neural networks of the projection network have specific connection weights, and Each layer of the neural network is composed of a number of neurons, and the projection network neurons are only connected to a limited number of neurons in the previous layer; the learning network is composed of a shallow neural network, the first layer of the neural network is connected to the first layer of the projection network, the shallow neural network is a fully connected network, and the neurons between different layers of the shallow neural network are connected through weights; it also includes: when determining the output information of each layer of the preset neural network model, determining the activation function of each layer of the neural network model, and normalizing the input signal of each layer; determining the output signal of each layer according to the normalized input signal and the corresponding activation function; A determination unit, used to determine the connection weights with each layer of the preset neural network model based on the target task corresponding to the input sample signal; wherein the determination unit is specifically used to obtain a target training sample corresponding to the input sample signal; input the target training sample to the input layer, and obtain the output of the input layer neuron; input the output of the input layer neuron to the projection network, and obtain the output signal of the projection network; input the output signal of the projection network to the learning network, and obtain the learning result of the learning network; based on the comparison value of the target task and the learning result, adjust the connection weights of the neurons to obtain the connection weights with each layer of the preset neural network model; A training unit is used to train the preset neural network model according to the connection weights of each layer of the preset neural network model to obtain a target neural network model.