Remote sensing data classification method, system, electronic equipment and storage medium

Feature extraction and dimensionality reduction through vortex optimization neural network and collaborative constraint auto-encoding network, combined with asynchronous update support vector machine for classification, the accuracy and adaptability of massive remote sensing data classification is solved, and efficient data classification is achieved.

CN118411564BActive Publication Date: 2024-11-19GUANGDONG PROVINCIAL ACADEMY OF ENVIRONMENTAL SCI
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410552550.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-05-07
Publication Date
2024-11-19
Estimated Expiration
2044-05-07

AI Technical Summary

Technical Problem

It is difficult for the prior art to accurately identify and classify massive, high-latitude, and high-complex remote sensing data, resulting in insufficient accuracy and adaptability of remote sensing data classification.

Method used

Feature extraction is performed using vortex optimization neural network algorithm, data dimensionality reduction is achieved based on collaborative constraints, and classification is performed through support vector machines with asynchronous update mechanism, improving the accuracy of feature extraction and dimensionality reduction and the adaptability of classification .

Benefits of technology

It effectively improves the accuracy and adaptability of remote sensing data classification, retains key information of the data, and improves the ability to adapt to different data distributions.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118411564B_ABST
    Figure CN118411564B_ABST
Patent Text Reader

Abstract

The present application discloses a remote sensing data classification method, system, electronic device and storage medium, the method comprising: obtaining a remote sensing data set to be classified; extracting features from the remote sensing data set to be classified by a preset feature extraction model to obtain remote sensing feature data; wherein the preset feature extraction model is obtained by training an eddy optimization neural network algorithm; inputting the remote sensing feature data into a preset feature dimension reduction model to perform data dimension reduction to obtain remote sensing dimension reduction data; wherein the preset feature dimension reduction model is obtained by training an autoencoder neural network algorithm based on collaborative constraints; classifying the remote sensing dimension reduction data by a preset classifier to obtain a remote sensing data classification result; wherein the preset classifier is obtained by training a support vector machine classification algorithm based on an asynchronous update mechanism. The embodiments of the present application can effectively improve the accuracy of remote sensing data classification and improve the adaptability to diversified data distribution. The present application can be widely used in the field of remote sensing data processing technology.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the technical field of remote sensing data processing, and in particular to a remote sensing data classification method, system, electronic equipment and storage medium. Background Art

[0002] With the rapid development of remote sensing technology, humans have acquired new means to observe the earth's surface, such as satellite remote sensing and drone aerial photography. Remote sensing technology can provide a large number of high-resolution images of the earth's surface, which are crucial for many fields such as environmental monitoring, urban planning, agriculture, forestry, hydrology, and disaster management. However, as the amount of data obtained continues to expand, it is difficult to extract useful information from a large amount of remote sensing data. In related technologies, the classification of remote sensing data mainly relies on manual analysis by experts or automatic processing programs based on simple rules. It performs well when processing small-scale data sets, but when faced with the current massive, high-latitude and highly complex remote sensing data, it is often difficult to accurately identify and classify diverse surface features.

[0003] In summary, the technical problems existing in the relevant technologies need to be improved. Summary of the invention

[0004] The main purpose of the embodiments of the present application is to propose a remote sensing data classification method, system, electronic device and storage medium, which can effectively improve the accuracy of remote sensing data classification and improve the adaptability to diversified data distribution.

[0005] To achieve the above purpose, one aspect of an embodiment of the present application provides a remote sensing data classification method, the method comprising:

[0006] Obtain remote sensing data sets to be classified;

[0007] Extracting features from the remote sensing data set to be classified by a preset feature extraction model to obtain remote sensing feature data; wherein the preset feature extraction model is obtained by training a vortex optimization neural network algorithm;

[0008] Inputting the remote sensing feature data into a preset feature dimensionality reduction model to perform data dimensionality reduction, thereby obtaining remote sensing dimensionality reduction data; wherein the preset feature dimensionality reduction model is obtained by training an autoencoder neural network algorithm based on collaborative constraints;

[0009] The remote sensing dimension reduction data is classified by a preset classifier to obtain a remote sensing data classification result; wherein the preset classifier is obtained by training a support vector machine classification algorithm based on an asynchronous update mechanism.

[0010] In some embodiments, before performing the step of extracting features from the remote sensing data set to be classified by using a preset feature extraction model to obtain remote sensing feature data, the method further includes:

[0011] Build a feature extraction neural network model;

[0012] Inputting a preset remote sensing training data set into the feature extraction neural network model to perform feature extraction to obtain training feature data;

[0013] Determine target parameter data through the eddy optimization neural network algorithm according to the training feature data;

[0014] The feature extraction neural network model is assigned parameters according to the target parameter data to construct the preset feature extraction model.

[0015] In some embodiments, determining the target parameter data by using the vortex optimization neural network algorithm according to the training feature data includes:

[0016] Constructing a first vortex set in a parameter space; wherein each vortex in the first vortex set includes a corresponding parameter configuration;

[0017] A loss function is calculated according to the training feature data to obtain a loss value, and the performance of the parameter configuration of each vortex is determined by the loss value, so as to sort the vortices according to the performance to obtain vortex sorting information;

[0018] Perform vortex dynamic update according to the vortex sorting information to obtain a second vortex set;

[0019] Performing a preset vortex operation on the second vortex set according to a preset vortex threshold to obtain a third vortex set; wherein the preset vortex operation includes vortex merging and vortex splitting;

[0020] Selecting vortices that meet preset performance conditions in the third vortex set for local search to adjust parameters, thereby obtaining a fourth vortex set;

[0021] The resources of the vortices in the fourth vortex set whose life value is lower than the preset life threshold are released, and the loss value obtained by calculating the loss function according to the training feature data is returned, and the performance of the parameter configuration of each vortex is determined by the loss value, so as to sort the vortices according to the performance and obtain the vortex sorting information, until the number of iterations is greater than the preset iteration threshold, and the target parameter data is obtained; wherein, the life value is determined by a preset attenuation rule.

[0022] In some embodiments, before executing the step of inputting the preset remote sensing training data set into the feature extraction neural network model to extract features to obtain training feature data, the method further includes:

[0023] Collect original remote sensing data and annotate the data to obtain the original data set;

[0024] The original data set is expanded by using a heuristic generative adversarial network model to obtain the preset remote sensing training data set; wherein the heuristic generative adversarial network model is obtained by training a generative adversarial network algorithm based on heuristic games.

[0025] In some embodiments, before performing the step of augmenting the original data set by using a heuristic generative adversarial network model to obtain the preset remote sensing training data set, the method further includes:

[0026] Constructing a preset generative adversarial network model; wherein the preset generative adversarial network model includes a generator, a discriminator, and a first heuristic strategy parameter;

[0027] Generate remote sensing data samples through the generator, and evaluate the authenticity of the remote sensing data samples through the discriminator to obtain feedback data;

[0028] Dynamically adjusting the first heuristic strategy parameter according to the feedback data and the data diversity information of the remote sensing data sample to obtain a second heuristic strategy parameter;

[0029] Optimizing the composite loss function according to the second heuristic strategy parameter to obtain a preset loss function; wherein the composite loss function includes content loss, style loss and discrimination loss;

[0030] Dynamically adjust the learning rate according to the variance of the feedback data to obtain a learning rate parameter;

[0031] The parameters of the preset generative adversarial network model are updated according to the learning rate parameter and the preset loss function to obtain the heuristic generative adversarial network model.

[0032] In some embodiments, before executing the step of inputting the remote sensing feature data into a preset feature dimension reduction model to perform data dimension reduction to obtain remote sensing dimension reduction data, the method further includes:

[0033] Constructing a first autoencoding neural network model; wherein the first autoencoding neural network model includes an encoder and a decoder;

[0034] Performing feature compression on the preset training data by the encoder, and reconstructing the data by the decoder to obtain reconstructed data;

[0035] Performing loss calculation according to the reconstruction data and the preset training data to obtain preset loss data; wherein the preset loss data includes reconstruction loss data and collaborative constraint loss data;

[0036] Adjusting parameters of the first autoencoder neural network model according to the inverse reinforcement learning strategy to obtain a second autoencoder neural network model;

[0037] The model parameters of the second autoencoder neural network model are updated through a back propagation algorithm according to the preset loss data to obtain the preset feature dimensionality reduction model.

[0038] In some embodiments, before performing the step of classifying the remote sensing dimensionality reduction data by a preset classifier to obtain a remote sensing data classification result, the method further includes:

[0039] Initializing classifier parameters to construct a first classifier; wherein the classifier parameters include a weight vector, a bias term, and a penalty parameter;

[0040] The parameters of the first classifier are updated according to the random selection mechanism and the time delay factor to obtain the preset classifier.

[0041] To achieve the above object, another aspect of the embodiment of the present application provides a remote sensing data classification system, the system comprising:

[0042] The first module is used to obtain the remote sensing data set to be classified;

[0043] The second module is used to extract features from the remote sensing data set to be classified through a preset feature extraction model to obtain remote sensing feature data; wherein the preset feature extraction model is obtained by training the eddy optimization neural network algorithm;

[0044] The third module is used to input the remote sensing feature data into a preset feature dimensionality reduction model to perform data dimensionality reduction to obtain remote sensing dimensionality reduction data; wherein the preset feature dimensionality reduction model is obtained by training an autoencoder neural network algorithm based on collaborative constraints;

[0045] The fourth module is used to classify the remote sensing dimension reduction data through a preset classifier to obtain a remote sensing data classification result; wherein the preset classifier is obtained by training a support vector machine classification algorithm based on an asynchronous update mechanism.

[0046] To achieve the above objective, another aspect of an embodiment of the present application provides an electronic device, the electronic device comprising:

[0047] at least one processor;

[0048] at least one memory for storing at least one program;

[0049] When the at least one program is executed by the at least one processor, the at least one processor implements the above method.

[0050] To achieve the above objective, another aspect of an embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and the computer program implements the above method when executed by a processor.

[0051] The embodiments of the present application include at least the following beneficial effects: The present application provides a remote sensing data classification method, system, electronic device and storage medium. The scheme obtains a remote sensing data set to be classified, and extracts features from the remote sensing data set to be classified through a preset feature extraction model to obtain remote sensing feature data. Among them, the embodiment of the present invention obtains a preset feature extraction model through eddy optimization neural network algorithm training to alleviate gradient problems and local optimal solution problems, and improves the accuracy and stability of feature extraction. Then, the embodiment of the present invention inputs the remote sensing feature data into a preset feature dimensionality reduction model for data dimensionality reduction to obtain remote sensing dimensionality reduction data. Among them, the embodiment of the present invention obtains a preset feature dimensionality reduction model through the training of an autoencoder neural network algorithm based on collaborative constraints to improve the feature dimensionality reduction performance, which can effectively reduce the data dimension and maintain the core features of the data. Finally, the embodiment of the present invention classifies the remote sensing dimensionality reduction data through a preset classifier to obtain a remote sensing data classification result. Among them, the embodiment of the present invention obtains a preset classifier through the support vector classification algorithm of the asynchronous update mechanism, which can improve the model's adaptability to data and effectively improve the accuracy of classification. It is easy to understand that the embodiment of the present invention effectively retains the key information of remote sensing data through feature extraction and data dimensionality reduction, and then classifies the data through the preset classifier trained by the asynchronous update mechanism, thereby improving the adaptability to different data distributions, thereby effectively improving the accuracy of remote sensing data classification. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 is a flow chart of a remote sensing data classification method provided in an embodiment of the present application;

[0053] Figure 2 is a schematic diagram of a feature extraction network structure provided in an embodiment of the present application;

[0054] Figure 3 is a structural diagram of a remote sensing data classification system provided in an embodiment of the present application;

[0055] Figure 4 It is a schematic diagram of the hardware structure of the electronic device provided in the embodiment of the present application. DETAILED DESCRIPTION

[0056] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application is further described in detail below in conjunction with the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the embodiments of the present application. They are only examples of devices and methods consistent with some aspects of the embodiments of the present application detailed in the attached claims.

[0057] It is understood that the terms "first", "second", etc. used in this application can be used to describe various concepts in this article, but unless otherwise specified, these concepts are not limited by these terms. These terms are only used to distinguish one concept from another. For example, without departing from the scope of the embodiments of the present application, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the words "if" and "if" as used herein can be interpreted as "at the time of" or "when" or "in response to determination".

[0058] The terms "at least one", "multiple", "each", "any", etc. used in this application, at least one includes one, two or more, multiple includes two or more, each refers to each of the corresponding multiple, and any refers to any one of the multiple.

[0059] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0060] Before describing the embodiments of the present application in detail, some nouns and terms involved in the embodiments of the present application are first described. The nouns and terms involved in the embodiments of the present application are subject to the following explanations.

[0061] Remote sensing data: refers to the information collected by remote sensing sensors, which includes the electromagnetic wave reflection and radiation characteristics of surface objects. Remote sensing data can be obtained through different types of remote sensing platforms, such as satellites, aircraft, etc.

[0062] Neural network model: It is a model that simulates biological nervous systems for calculation and learning. It is composed of a large number of nodes (neurons) and connections between nodes. Each neuron in the neural network receives input signals from other neurons, and adjusts the connection weights according to the strength of these signals through specific algorithms (such as back propagation algorithms) to generate output signals, which can be used for classification, prediction, pattern recognition and other tasks.

[0063] Support vector machine: It is a generalized linear classifier that performs binary classification of data in a supervised learning manner. Its decision boundary is the maximum margin hyperplane solved for the learning samples.

[0064] With the rapid development of remote sensing technology, humans have acquired new means to observe the earth's surface, such as satellite remote sensing and drone aerial photography. Remote sensing technology can provide a large number of high-resolution images of the earth's surface, which are crucial for many fields such as environmental monitoring, urban planning, agriculture, forestry, hydrology, and disaster management. However, as the amount of data obtained continues to expand, it is difficult to extract useful information from a large amount of remote sensing data. In related technologies, the classification of remote sensing data mainly relies on manual analysis by experts or automatic processing programs based on simple rules. It performs well when processing small-scale data sets, but when faced with the current massive, high-latitude and highly complex remote sensing data, it is often difficult to accurately identify and classify diverse surface features.

[0065] In view of this, a remote sensing data classification method, system, electronic device and storage medium are provided in the embodiment of the present application. The scheme obtains a remote sensing data set to be classified, and extracts features from the remote sensing data set to be classified through a preset feature extraction model to obtain remote sensing feature data. Among them, the embodiment of the present invention obtains a preset feature extraction model through eddy optimization neural network algorithm training to alleviate gradient problems and local optimal solution problems, and improves the accuracy and stability of feature extraction. Then, the embodiment of the present invention inputs the remote sensing feature data into a preset feature dimensionality reduction model for data dimensionality reduction to obtain remote sensing dimensionality reduction data. Among them, the embodiment of the present invention obtains a preset feature dimensionality reduction model through the training of an autoencoder neural network algorithm based on collaborative constraints to improve the feature dimensionality reduction performance, which can effectively reduce the data dimension and maintain the core features of the data. Finally, the embodiment of the present invention classifies the remote sensing dimensionality reduction data through a preset classifier to obtain a remote sensing data classification result. Among them, the embodiment of the present invention obtains a preset classifier through the support vector classification algorithm of the asynchronous update mechanism, which can improve the model's adaptability to data and effectively improve the accuracy of classification.

[0066] The remote sensing data classification method provided in the embodiment of the present application relates to the field of data processing technology. The remote sensing data classification method provided in the embodiment of the present application can be applied to a terminal, can be applied to a server, or can be software running in a terminal or a server. In some embodiments, the terminal can be a smart phone, a tablet computer, a laptop computer, a desktop computer, a smart speaker, a smart watch, and a car terminal, etc., but is not limited to this; the server side can be configured as an independent physical server, or it can be configured as a server cluster or distributed system composed of multiple physical servers, and can also be configured to provide cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms and other basic cloud computing services. The cloud server, the server can also be a node server in the blockchain network; the software can be an application that implements the remote sensing data classification method, etc., but is not limited to the above forms.

[0067] The present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc. The present application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application can also be practiced in distributed computing environments, in which tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

[0068] Figure 1 is an optional flow chart of the remote sensing data classification method provided in the embodiment of the present application. Figure 1 The method may include but is not limited to steps S110 to S140.

[0069] Step S110, obtaining a remote sensing data set to be classified.

[0070] Step S120, extracting features from the remote sensing data set to be classified by using a preset feature extraction model to obtain remote sensing feature data, wherein the preset feature extraction model is obtained by training the eddy optimization neural network algorithm.

[0071] Step S130, inputting the remote sensing feature data into a preset feature dimension reduction model to perform data dimension reduction to obtain remote sensing dimension reduction data. The preset feature dimension reduction model is obtained by training an autoencoder neural network algorithm based on collaborative constraints.

[0072] Step S140, classifying the remote sensing dimension reduction data by a preset classifier to obtain a remote sensing data classification result, wherein the preset classifier is obtained by training a support vector machine classification algorithm based on an asynchronous update mechanism.

[0073] In this specific embodiment, the embodiment of the present invention first obtains a remote sensing data set to be classified, and extracts features from the remote sensing data set to be classified through a preset feature extraction model to obtain remote sensing feature data. Specifically, the embodiment of the present invention first obtains remote sensing data samples that need to be classified, that is, remote sensing data sets to be classified. Then, the embodiment of the present invention inputs these data samples into a preset feature extraction model to extract features and obtain corresponding remote sensing feature data. Among them, the embodiment of the present invention obtains a preset feature extraction model through eddy optimization neural network algorithm training, so as to optimize the neural network parameters through the eddy optimization neural network algorithm, thereby alleviating the problems of gradient disappearance and explosion, and effectively improving the stability of training and model performance. Then, the embodiment of the present invention inputs the extracted remote sensing feature data into a preset feature dimensionality reduction model for data extraction, and obtains remote sensing dimensionality reduction data. Specifically, the embodiment of the present invention obtains a preset feature dimensionality reduction model through training of an autoencoder neural network algorithm based on collaborative constraints. Among them, the embodiment of the present invention adopts a collaborative constraint mechanism so that the encoder can take into account the requirements of the decoder reconstruction quality while compressing the data features, so that it can not only reduce the dimension of the data in the dimensionality reduction process, but also retain the key information of the metadata to the maximum extent. Accordingly, the embodiment of the present invention performs data dimensionality reduction on the remote sensing feature data through the preset feature extraction model obtained by training to reduce the data dimension, and can effectively retain the key information of the data, obtain remote sensing dimensionality reduction data, and provide support for remote sensing data classification. Finally, the embodiment of the present invention classifies the remote sensing dimensionality reduction data through a preset classifier to obtain the remote sensing data classification result. Specifically, the support vector machine of the asynchronous update mechanism in the embodiment of the present invention trains the preset classifier by asynchronous updating and adaptively adjusting the penalty parameter, which effectively improves the model training speed and the adaptability to new data. Accordingly, the categories classified in the embodiment of the present invention include "water body", "grassland", "shrubland", "broad-leaved farmland", "savanna", "evergreen broad-leaved forest", "deciduous broad-leaved forest", "evergreen coniferous forest", "deciduous coniferous forest", "non-vegetation covered land", "urban and construction land", etc.

[0074] In some embodiments of the present invention, before performing the step of extracting features from the remote sensing data set to be classified by using a preset feature extraction model to obtain remote sensing feature data, the method provided by the embodiment of the present invention further includes but is not limited to the following steps:

[0075] Build a feature extraction neural network model.

[0076] The preset remote sensing training data set is input into the feature extraction neural network model for feature extraction to obtain training feature data.

[0077] The target parameter data is determined by the eddy optimization neural network algorithm based on the training feature data.

[0078] The feature extraction neural network model is assigned parameters according to the target parameter data to construct a preset feature extraction model.

[0079] In this specific embodiment, before extracting features from a remote sensing data set to be classified using a preset feature extraction model, the embodiment of the present invention first obtains a preset feature extraction model by training a vortex optimization neural network algorithm. Specifically, Figure 2As shown, the embodiment of the present invention first constructs a feature extraction neural network model, i.e., a feature extraction module, and inputs a preset remote sensing training data set into the feature extraction neural network model for feature extraction to obtain training feature data. Then, the embodiment of the present invention determines the corresponding target parameter data through the vortex optimization neural network algorithm according to the extracted training feature data. Among them, a 3-layer fully connected neural network is used for feature extraction in the embodiment of the present invention. Since some neural network structures may have problems of gradient disappearance, gradient explosion, or fall into a local optimal solution when extracting features through a neural network model, the stability of the training and the performance of the model are affected. Therefore, the embodiment of the present invention analyzes the training feature data through the vortex optimization neural network algorithm to determine the target parameter data of the model. Accordingly, the embodiment of the present invention optimizes the parameters (weight w and bias b) in the neural network by simulating the formation, evolution, and dissipation process of the vortex. In addition, in fluid dynamics, vortices can increase the chaos and complexity of the fluid, thereby guiding the search process in the parameter space toward a better solution. By simulating this natural process, the vortex optimization neural network aims to find parameter configurations that can minimize the neural network loss function, without relying on the gradient-based neural network parameter optimization method, thereby effectively alleviating the gradient problem and the local optimal solution problem, and effectively improving the accuracy and reliability of feature extraction. Furthermore, the embodiment of the present invention assigns parameters to the feature extraction neural network model according to the target parameter data to construct a preset feature extraction model. The embodiment of the present invention searches for the optimal solution of the model parameters of the model in the parameter space by simulating the vortex phenomenon, that is, the target parameter data, and then assigns the target parameter data to the feature extraction neural network model to construct a preset feature extraction model.

[0080] In some embodiments of the present invention, target parameter data is determined by using a vortex optimization neural network algorithm based on training feature data, including but not limited to:

[0081] A first vortex set is constructed in the parameter space, wherein each vortex in the first vortex set includes a corresponding parameter configuration.

[0082] The loss function is calculated based on the training feature data to obtain the loss value, and the performance of the parameter configuration of each vortex is determined by the loss value, so that the vortices are sorted by performance to obtain vortex sorting information.

[0083] The vortexes are dynamically updated according to the vortex sorting information to obtain a second vortex set.

[0084] The second vortex set is subjected to a preset vortex operation according to a preset vortex threshold value to obtain a third vortex set, wherein the preset vortex operation includes vortex merging and vortex splitting.

[0085] A vortex that meets a preset performance condition in the third vortex set is selected for local search to adjust parameters, thereby obtaining a fourth vortex set.

[0086] The vortices whose life value is lower than the preset life threshold in the fourth vortex set are released, and the loss value is obtained by calculating the loss function according to the training feature data, and the performance of the parameter configuration of each vortex is determined by the loss value, so as to sort the vortices by performance and obtain the vortex sorting information, until the number of iterations is greater than the preset iteration threshold, and the target parameter data is obtained. The life value is determined by a preset attenuation rule.

[0087] In this specific embodiment, the embodiment of the present invention first constructs a first vortex set in the parameter space. Specifically, each vortex in the first vortex set in the embodiment of the present invention includes a corresponding parameter configuration, such as a weight w and a bias b. Among them, each vortex in the embodiment of the present invention represents a possible parameter configuration. Exemplarily, the embodiment of the present invention defines a neural network parameter vector P i =(w i ,b i ), where i is the vortex index. During initialization, the parameter vector P of each vortex i It is randomly generated in the parameter space and conforms to the uniform distribution. The total number of initial vortices is set to N, and the initialization process is shown in the following formula (1):

[0088]

[0089] in, represents uniform distribution, a and b are the lower bound and upper bound of the parameter space, and are preset artificially in some embodiments of the present invention.

[0090] Next, the embodiment of the present invention calculates the loss function according to the training feature data, and determines the performance of the parameter configuration of each vortex through the corresponding loss value obtained by calculation, so as to sort the vortices according to the performance and obtain vortex sorting information. Specifically, Figure 2 As shown, after the feature extraction module outputs the feature data after feature extraction in the embodiment of the present invention, the loss function L(P i ) to evaluate the turbine performance to obtain the performance of each vortex. After the performance evaluation is completed, the embodiment of the present invention sorts the turbines according to the performance of each turbine, that is, the value of the loss function, as shown in the following formula (2):

[0091] L(P1)≤L(P2)≤…≤L(P N )(2)

[0092] It should be noted that, in some embodiments of the present invention, the loss function L(P i ) can be the cross entropy loss function.

[0093] Furthermore, the embodiment of the present invention dynamically updates the vortices according to the vortex ranking information to obtain a second vortex set. Specifically, when performing the vortex dynamic update, the embodiment of the present invention dynamically adjusts the position and strength of the vortices according to the ranking of the vortices and the distance between them to simulate the interaction and evolution of the vortices. Exemplarily, during the vortex dynamic update stage, the embodiment of the present invention adjusts the position of the vortex in the parameter space by simulating the interaction of the vortices, and sets the distance between vortices i and j to d. ij =∥P i -P j ∥, then the update amount of the i-th vortex ΔP i As shown in the following formula (3):

[0094]

[0095] Among them, γ is a hyperparameter that adjusts the distance effect, F(d ij ) is a distance-based adjustment function, λ b is the adaptive vortex factor, ξ i It is a regulatory factor based on environmental feedback.

[0096] In some embodiments of the present invention, the distance-based adjustment function F(d ij ) is calculated by using an exponential decay model to reflect the weakening of the interaction force between vortices as the distance increases, as shown in the following formula (4):

[0097] F(d ij )=α q exp(-β q d ij )(4)

[0098] Among them, α q and β q is a hyperparameter that controls the magnitude and decay rate of the force. For example, in the embodiment of the present invention, parameter α q is set to 0.5, β q is set to 0.1. In addition, in the embodiment of the present invention, d ij represents the distance between vortex i and vortex j, which is calculated based on the Euclidean distance of the parameter vectors representing the two vortices, as shown in the following formula (5):

[0099]

[0100] Where D is the dimension of the parameter vector, P ik and P jk are the parameter values ​​of vortex i and vortex j in the kth dimension respectively.

[0101] In addition, the adaptive swirl factor λ in the embodiment of the present invention is b The information entropy H of the vortex distribution is calculated by adaptive vortex guided learning and is shown in the following formula (6):

[0102]

[0103] Among them, p i is the proportion of the i-th vortex in all vortices, which is calculated by the inverse of the vortex loss function, as shown in the following formula (7):

[0104]

[0105] Accordingly, the embodiment of the present invention adapts the vortex factor λ according to the change of information entropy. b The search strategy for adjusting the vortex is calculated as shown in the following formula (8):

[0106]

[0107] Among them, κ is a preset maximum regulation strength, H max It is the maximum information entropy value that can be achieved theoretically, indicating a completely uniform distribution state. i is the momentum adjustment factor, M i is the average momentum vector of vortex i in the most recent T iterations.

[0108] At the same time, the momentum adjustment factor μ in the embodiment of the present invention is i and the average momentum vector M i It is obtained by time series momentum adjustment. Among them, the time series momentum adjustment aims to adjust the exploration trend of the vortex by analyzing its historical movement trend in the parameter space, considering the movement direction and speed of the vortex in the historical iteration, calculating its momentum, and then using this momentum information to guide future parameter updates. For example, calculate the average momentum vector M of vortex i in the last T iterations i As shown in the following formula (9):

[0109]

[0110] Among them, P i () is the parameter vector of vortex i after the tth iteration, P i (-1) is the parameter vector after the t-1th iteration.

[0111] Accordingly, in the embodiment of the present invention, the momentum adjustment factor μ i It is used to measure the size and consistency of momentum, as shown in the following formula (10):

[0112] μi =∥M i ∥·cos(θ i )(10)

[0113] Among them, θ i It is M i and P i () -P i (-1) The angle between them is used to measure the consistency between the latest moving direction and the average momentum direction. The smaller the angle, the more consistent the momentum.

[0114] In addition, the calculation method of the adjustment factor based on environmental feedback in the embodiment of the present invention can be shown in the following formula (11):

[0115] ξ i =Sig(E(P i ))(11)

[0116] Where Sig() is the Sigmoid mapping function, which maps the environmental feedback to the (0, 1) interval. i ) is the environmental feedback function, which is used to evaluate the characteristics of the environment around vortex i, and the calculation method is shown in the following formula (12):

[0117]

[0118] Among them, is the Hessian matrix of the loss function at the vortex i position (used to evaluate local concavity), is the magnitude of the gradient (used to assess steepness in parameter space).

[0119] Furthermore, the embodiment of the present invention performs a preset vortex operation on the second vortex set according to a preset vortex threshold to obtain a third vortex set. Specifically, the preset vortex operation in the embodiment of the present invention includes vortex merging and vortex splitting. The embodiment of the present invention promotes spatial diversity and explores new parameter areas by simulating the merging and splitting phenomena of vortices in nature. Exemplarily, the embodiment of the present invention first sets a preset vortex threshold, such as a merging threshold T m and the splitting threshold T s When the distance between the two vortices is less than T m When the strength of a vortex exceeds T s When , it will split into two vortices. Accordingly, the merging operation in the embodiment of the present invention can be expressed as the following formula (13):

[0120]

[0121] Among them, L(Pi ) and L(P j ) is the loss function value corresponding to vortex i and vortex j.

[0122] Accordingly, the splitting operation in the embodiment of the present invention can be expressed as the following formula (14):

[0123]

[0124] Among them, δ q is a random perturbation vector based on the parameter vector of vortex i, used to generate the parameters of two new vortices.

[0125] Accordingly, in the embodiment of the present invention, the random disturbance vector δ is generated based on the local gradient information of the current vortex position to increase the randomness and exploratory nature of the search process. The calculation method can be expressed as the following formula (15):

[0126]

[0127] Among them, ζ q and ε q is a hyperparameter that controls the gradient direction and random perturbation intensity, and g represents the vortex position P i The gradient information calculated at The mean is 0 and the covariance matrix is For example, in some embodiments of the present invention, q is set to 1.0, ε q is set to 0.1.

[0128] Furthermore, the embodiment of the present invention selects vortices that meet the preset performance conditions in the third vortex set for local search to adjust parameters and obtain a fourth vortex set. Specifically, the embodiment of the present invention selects vortices whose performance meets the preset performance conditions in the third vortex set, such as the vortex with the best performance, and performs local search to fine-tune its parameters to simulate the local absorption and energy concentration phenomenon of the vortex in the fluid. Exemplarily, for the best-performing vortex, the embodiment of the present invention performs a local search to further fine-tune its parameters, which is achieved through a domain-based search strategy. For example, in the embodiment of the present invention, in the domain-based search strategy, the local search method can be expressed as the following formula (16):

[0129]

[0130] Among them, η m is the learning rate, is 0 as the mean, Σ loc is a Gaussian distribution with covariance matrix, used to generate random perturbations of the parameters.

[0131] Accordingly, the covariance matrix Σ loc According to the historical movement direction and speed of the vortex in the parameter space, it is dynamically adjusted to adapt to different search stages and characteristics. Its update strategy can be expressed as the following formula (17):

[0132]

[0133] Among them, is the adjusted covariance matrix, is the covariance matrix before adjustment, ρ n is a learning rate parameter used to control the influence of historical information on the current search. For example, in some embodiments of the present invention, ρ n is set to 0.2.

[0134] Furthermore, the embodiment of the present invention releases resources for the vortices whose life values ​​are lower than a preset life threshold in the fourth vortex set. Specifically, the embodiment of the present invention simulates the dissipation process of vortex energy in nature by performing vortex attenuation, gradually eliminates vortices with poor performance, and releases resources to newly generated or better performing vortices. Accordingly, the attenuation process in the embodiment of the present invention is achieved by adjusting the life value of the vortex. Vortices with life values ​​lower than a certain threshold will be eliminated. The life value of vortex i is set to L i , the preset attenuation rule is shown in the following formula (18):

[0135]

[0136] Among them, is the health value after decay, δ L It is the amount of health reduction.

[0137] Then, the embodiment of the present invention returns to the step of calculating the loss value by performing the loss function calculation according to the training feature data, and determining the performance of the parameter configuration of each vortex by the loss value, so as to sort the vortex by performance and obtain the vortex sorting information, until the number of iterations is greater than the preset iteration threshold, and the target parameter data is obtained. Among them, after each round of iteration, the embodiment of the present invention updates the state of the vortex and re-evaluates their performance to ensure that the algorithm progresses toward the optimal solution. For example, in an embodiment of the present invention, the maximum number of iterations is set to 1000 times.

[0138] In some embodiments of the present invention, before executing the step of inputting a preset remote sensing training data set into a feature extraction neural network model to extract features and obtain training feature data, the remote sensing data classification method provided by the embodiment of the present invention further includes but is not limited to the following steps:

[0139] Collect original remote sensing data and annotate the data to obtain the original data set.

[0140] The original data set is augmented by a heuristic generative adversarial network model to obtain a preset remote sensing training data set. The heuristic generative adversarial network model is trained by a generative adversarial network algorithm based on heuristic games.

[0141] In this specific embodiment, the embodiment of the present invention first collects raw remote sensing data and performs data annotation to obtain a raw data set. Specifically, the embodiment of the present invention obtains detailed images of the surface through high-precision remote sensing technology, and accurately classifies and annotates these images to obtain a raw data set. The embodiment of the present invention obtains remote sensing data through remote sensing satellites or aerial photography drones. These data have high resolution and multi-spectral characteristics, and the data format is mainly multi-dimensional image data. Among them, each image consists of multiple spectral bands and can be represented as a three-dimensional array ax i ,ay j ,az k , where i, j, k represent the length, width and number of spectral bands of the image respectively. Then, the embodiment of the present invention labels the collected data. For example, the labeling method in the embodiment of the present invention is manual labeling. Accordingly, the labeling categories in the embodiment of the present invention include 11 categories: "water body", "grassland", "shrubland", "broad-leaved farmland", "savanna", "evergreen broad-leaved forest", "deciduous broad-leaved forest", "evergreen coniferous forest", "deciduous coniferous forest", "non-vegetated land", and "urban and construction land". Exemplarily, the embodiment of the present invention can collect remote sensing data of forest coverage areas in a certain area. The collected remote sensing data has a spatial resolution of 30 cm, and the spectral range covers the visible light to near-infrared bands. The data format is ax i ,ay j ,az k On this basis, the embodiment of the present invention forms a remote sensing data set for model training, that is, an original data set, by labeling 11 categories, namely, "water body", "grassland", "shrubland", "broad-leaved farmland", "savanna", "evergreen broad-leaved forest", "deciduous broad-leaved forest", "evergreen coniferous forest", "deciduous coniferous forest", "non-vegetated land", and "urban and construction land".

[0142] Furthermore, the embodiment of the present invention performs data expansion on the original data set through a heuristic generative adversarial network model to obtain a preset remote sensing training data set. Specifically, the heuristic generative adversarial network model in the embodiment of the present invention is obtained by training a generative adversarial network algorithm based on a heuristic game. Since the acquisition, labeling and preprocessing of training data are time-consuming and labor-intensive, and insufficient training samples are likely to lead to poor generalization ability of the model, while affecting the accuracy of the model. Therefore, the embodiment of the present invention adopts a multi-strategy loss function to achieve effective expansion of remote sensing data. Among them, the traditional generative adversarial network relies on the game process of a generator and a discriminator. The embodiment of the present invention introduces a heuristic strategy so that the generator can more accurately simulate the distribution characteristics of remote sensing data when generating data, allowing the generator to gradually learn and adjust during the game process, thereby generating more diverse and real data samples. In addition, in order to further improve the quality of data expansion, the embodiment of the present invention adopts a multi-strategy loss function, which combines multiple different loss terms to optimize the authenticity and diversity of generated data from multiple dimensions, especially while maintaining the characteristics of remote sensing data while increasing the generalization ability of samples.

[0143] In some embodiments of the present invention, before performing the step of expanding the original data set by using a heuristic generative adversarial network model to obtain a preset remote sensing training data set, the remote sensing data classification method provided by the embodiment of the present invention further includes but is not limited to the following steps:

[0144] Construct a preset generative adversarial network model, wherein the preset generative adversarial network model includes a generator, a discriminator, and a first heuristic strategy parameter.

[0145] The remote sensing data samples are generated by the generator, and the authenticity of the remote sensing data samples is evaluated by the discriminator to obtain feedback data.

[0146] The first heuristic strategy parameters are dynamically adjusted according to the data diversity information of the feedback data and the remote sensing data samples to obtain the second heuristic strategy parameters.

[0147] The composite loss function is optimized according to the second heuristic strategy parameters to obtain a preset loss function, wherein the composite loss function includes content loss, style loss and discrimination loss.

[0148] The learning rate is dynamically adjusted according to the variance of the feedback data to obtain the learning rate parameter.

[0149] The parameters of the preset generative adversarial network model are updated according to the learning rate parameters and the preset loss function to obtain a heuristic generative adversarial network model.

[0150] In this specific embodiment, the embodiment of the present invention first constructs a preset generative adversarial network model. Specifically, the preset generative adversarial network model in the embodiment of the present invention includes a generator, a discriminator, and a first heuristic strategy parameter. The embodiment of the present invention first initializes the network parameters of the generator and the discriminator, and sets the first heuristic strategy parameter to provide preliminary generation guidance for the generator. Exemplarily, in the initialization stage, the embodiment of the present invention sets the parameters θ of the generator G and the discriminator D G and θ D , heuristic strategy parameter ψ, then the initialization methods of the generator G and the discriminator D can be expressed as the following equations (19) and (20) respectively:

[0151]

[0152] Accordingly, the initialization method of the first heuristic strategy parameter in the embodiment of the present invention can be expressed as the following formula (21):

[0153]

[0154] Among them, represents a normal distribution, represents uniform distribution, I is the identity matrix, and a and b are the upper and lower limits of the first heuristic strategy parameters, respectively.

[0155] Next, the embodiment of the present invention generates remote sensing data samples through the generator, and performs authenticity evaluation on the remote sensing data samples through the discriminator to obtain feedback data. Specifically, the generator and the discriminator enter a loop iteration process. In each iteration process, the generator first generates remote sensing data samples according to the current strategy parameters, and then the discriminator evaluates the authenticity of these samples. Accordingly, the embodiment of the present invention adjusts the parameters and strategies of the generator according to the feedback of the discriminator to improve the quality of the generated samples. Exemplarily, in the loop iteration process, the data generation process of the generator can be expressed as the following formula (22):

[0156] x fake =G(z;θ G )(twenty two)

[0157] Where z is the value from the prior distribution p z The noise extracted from .

[0158] Accordingly, the authenticity evaluation process of the discriminator in the embodiment of the present invention can be expressed as the following formula (23):

[0159]

[0160] Furthermore, the embodiment of the present invention dynamically adjusts the first heuristic strategy parameter according to the data diversity information of the feedback data and the remote sensing data sample to obtain the second heuristic strategy parameter. Specifically, in the iterative process, according to the quality and diversity of the generated data, that is, the feedback data and the data diversity information, the heuristic strategy parameter is dynamically adjusted to continuously optimize the generation strategy and improve the effect of data expansion. Exemplarily, the method of dynamically adjusting the heuristic strategy parameter ψ in the embodiment of the present invention can be expressed as the following formula (24):

[0161]

[0162] Wherein, R represents the reward function based on generated data and real data, and δ is the strategy update rate. Accordingly, in some embodiments of the present invention, the calculation method of the reward function R can be expressed as the following formula (25):

[0163] R(x fake ,x real )=α R ∥x fake -x real ∥+β R Entropy fake )(25)

[0164] Among them, Entropy (D fake ) represents the entropy of generated data, which is used to measure the diversity of generated data, α R and β R is an adjustment factor used to balance the distance term and the diversity term.

[0165] Furthermore, the embodiment of the present invention optimizes the composite loss function according to the second heuristic strategy parameters to obtain a preset loss function. Specifically, the composite loss function in the embodiment of the present invention includes content loss, style loss, and discrimination loss. Accordingly, the embodiment of the present invention refines the characteristics of the generated data by optimizing the composite loss function including multiple loss terms to ensure that the generated remote sensing data has both a high degree of authenticity and can cover a wide range of data distribution. Exemplarily, the multi-strategy loss function (composite loss function) in the embodiment of the present invention is composed of content loss L content , style loss L style and the discriminative loss L discrim The calculation method can be expressed as the following formula (26):

[0166] L total =αL content (x fake ,x real )+βL style (x fake ,x real )+γLdiscrim (D fake ,D real )(26)

[0167] Among them, α, β, and γ are weight coefficients, which adjust the relative importance of the loss items. For example, in the embodiment of the present invention, the loss weight α is set to 0.5, the style loss weight β is set to 0.4, and the discrimination loss weight γ is set to 0.1.

[0168] In addition, the content loss L in the embodiment of the present invention is content The calculation method can be expressed as the following formula (27):

[0169]

[0170] Wherein, F represents a feature extraction function, and N is the number of samples in a batch. In the embodiment of the present invention, the feature extraction function is a single-layer neural network feature extraction function.

[0171] Accordingly, the style loss L in the embodiment of the present invention is style The calculation method can be expressed as the following formula (28):

[0172]

[0173] Among them, G l represents the Gram matrix of the lth layer, w l is the weight coefficient of each layer, is the Frobenius norm.

[0174] At the same time, the discrimination loss L in the embodiment of the present invention is discrim The calculation method of can be expressed as the following formula (29):

[0175] L discrim (D fake ,D real )=-log(D real )-log(1-D fake )(29)

[0176] Furthermore, the embodiment of the present invention dynamically adjusts the learning rate according to the variance of the feedback data to obtain a learning rate parameter. Specifically, the embodiment of the present invention adopts an adaptive adjustment mechanism and a feedback mechanism to balance the stability of the model and the diversity of the generated data, so as to ensure that while improving the quality of the generated samples, the model is prevented from converging to a local optimal solution too early. Exemplarily, the method of dynamically adjusting the learning rate in the embodiment of the present invention can be expressed as the following formula (30):

[0177]

[0178] Wherein, VAR represents variance, and f() is an adjustment function. For example, in the embodiment of the present invention, the learning rate λ G and λ D Set to 0.0002 and 0.0001 respectively.

[0179] In addition, the calculation method of the adjustment function f() in the embodiment of the present invention can be expressed as the following formula (31):

[0180]

[0181] Wherein, VAR(D) is the variance of the discriminator output, τ is the target variance threshold, and κ is the adjustment sensitivity parameter. For example, in the embodiment of the present invention, the adjustment sensitivity parameter κ is set to 0.05. The embodiment of the present invention adjusts the learning rate by comparing the difference between the actual variance and the target variance, in order to achieve the best balance between model stability and generated data diversity.

[0182] Next, the embodiment of the present invention updates the parameters of the preset generative adversarial model according to the learning rate parameter and the preset loss function to obtain a heuristic generative adversarial network model. Specifically, the parameter update method in the embodiment of the present invention can be expressed as the following formula (32):

[0183]

[0184] Among them, L G and L D are the loss functions of the generator and the discriminator, respectively, G and λ D is the learning rate.

[0185] Accordingly, during the training process, the calculation method of the heuristic strategy parameter ψ can be expressed as follows (33):

[0186] ψ=a+(ba)·rand(0,1)(33)

[0187] Wherein, rand(0,1) generates a uniform random number between [0,1], and a and b are the lower and upper bounds of the policy parameter, respectively. For example, in the embodiment of the present invention, the upper bound b can be set to 1, and the lower bound a can be set to -1.

[0188] Accordingly, when the number of iterations reaches the preset maximum number of iterations, the training ends and the heuristic generative adversarial network model is constructed. In the embodiment of the present invention, the trained generator outputs the expanded data, thereby constructing an expanded remote sensing data set, that is, a preset remote sensing training data set.

[0189] In some embodiments of the present invention, before executing the step of inputting remote sensing feature data into a preset feature dimension reduction model to perform data dimension reduction to obtain remote sensing dimension reduction data, the remote sensing data classification method provided by the embodiment of the present invention further includes but is not limited to the following steps:

[0190] A first autoencoding neural network model is constructed, wherein the first autoencoding neural network model includes an encoder and a decoder.

[0191] The preset training data is feature compressed by the encoder and reconstructed by the decoder to obtain reconstructed data.

[0192] Loss calculation is performed based on the reconstruction data and the preset training data to obtain preset loss data, wherein the preset loss data includes reconstruction loss data and collaborative constraint loss data.

[0193] The parameters of the first autoencoder neural network model are adjusted according to the inverse reinforcement learning strategy to obtain a second autoencoder neural network model.

[0194] The model parameters of the second autoencoder neural network model are updated through a back-propagation algorithm according to the preset loss data to obtain a preset feature dimensionality reduction model.

[0195] In this specific embodiment, the embodiment of the present invention uses a collaborative constraint mechanism to enable the encoder to take into account the reconstruction quality requirements of the decoder while compressing the data features, so that the dimension of the data can be reduced during the dimension reduction process, and the key information of the original data can be retained to the maximum extent. At the same time, the embodiment of the present invention uses the technology of inverse reinforcement learning to enable the autoencoder to automatically adjust its parameters during the training process, optimize the performance of feature dimension reduction, and allow the model to learn how to effectively reduce the data dimension by maximizing the expected reward without a direct goal, while maintaining the core features of the data. Accordingly, the embodiment of the present invention first constructs a first autoencoder neural network model, including an encoder and a decoder. Specifically, the embodiment of the present invention first initializes the parameters of the autoencoder, including the weights and biases of the encoder and the decoder. For example, the embodiment of the present invention sets eW and eb to represent the weights and biases of the encoder, respectively, and dW and db to represent the weights and biases of the decoder, respectively. Among them, the method of initializing the parameters of the autoencoder in the embodiment of the present invention is random initialization.

[0196] Next, the embodiment of the present invention performs feature compression on the preset training data through the encoder, and reconstructs it through the decoder to obtain reconstructed data. Specifically, in the embodiment of the present invention, the input data (preset training data) is feature compressed through the encoder, and then reconstructed through the decoder to obtain reconstructed data. Exemplarily, the encoding process in the embodiment of the present invention is shown in the following formula (34):

[0197] z=Sig(eW·x+eb)(34)

[0198] Among them, x is the input remote sensing data feature vector, z is the encoded feature representation, and Sig() is the Sigmoid activation function.

[0199] Accordingly, the decoding process in the embodiment of the present invention is shown in the following formula (35):

[0200]

[0201] Among them, is the reconstructed remote sensing data feature vector.

[0202] Accordingly, the calculation method of the Sigmoid function in the embodiment of the present invention can be expressed as the following formula (36):

[0203]

[0204] Furthermore, the embodiment of the present invention performs loss calculation based on the reconstruction data and the preset training data to obtain the preset loss data. Specifically, the preset loss data in the embodiment of the present invention includes reconstruction loss data and collaborative constraint loss data. The embodiment of the present invention calculates the difference between the reconstructed data and the original data while considering the collaborative constraint loss to ensure the retention of key information in the feature dimensionality reduction process. Exemplarily, the calculation method of the reconstruction loss in the embodiment of the present invention can be expressed as the following formula (37):

[0205]

[0206] Where n is the number of samples, is the eigenvector of the i-th reconstructed remote sensing data, x i is the i-th original remote sensing data feature vector, that is, the remote sensing data feature vector of the preset training data.

[0207] Accordingly, in the embodiment of the present invention, the calculation formula of the collaborative constraint loss is shown in the following formula (38):

[0208]

[0209] Where λ is the regularization parameter, which controls the strength of the collaborative constraint, ∥∥ F represents the Frobenius norm. For example, in the embodiment of the present invention, the regularization parameter λ y is set to 0.001.

[0210] In addition, the Frobenius norm in the embodiment of the present invention is and The calculation formulas are shown in equations (39) and (40) below:

[0211]

[0212] Where m and n represent the number of rows and columns of the weight matrix respectively, eW ij and dW ij are the elements in the encoder and decoder weight matrices respectively.

[0213] Furthermore, the embodiment of the present invention adjusts the parameters of the first autoencoder neural network model according to the inverse reinforcement learning strategy to obtain a second autoencoder neural network model. Specifically, the embodiment of the present invention adjusts the model parameters to maximize the expected reward and optimize the feature dimensionality reduction performance according to the inverse reinforcement learning strategy. In the embodiment of the present invention, inverse reinforcement learning aims to adjust the values ​​of eW, eb, dW, db by optimizing the reward R, thereby minimizing the reconstruction loss. Exemplarily, the reward function in the embodiment of the present invention is defined as shown in the following formula (41):

[0214] R=-L recon (41)

[0215] Furthermore, the embodiment of the present invention updates the model parameters of the second autoencoder neural network model through the back propagation algorithm according to the preset loss data to obtain a preset feature dimension reduction model. Specifically, the embodiment of the present invention updates the model parameters through the back propagation algorithm and adjusts the learning rate using a dynamic learning rate adjustment strategy. Exemplarily, the embodiment of the present invention uses the gradient descent method to update the parameters, where the learning rate is α y , the parameter update method can be expressed as the following formula (42):

[0216]

[0217] Wherein, eW′, eb′, dW′, and db′ are respectively the updated encoder weight, encoder bias, decoder weight, and decoder bias. For example, in the embodiment of the present invention, the learning rate α y is set to 0.0005.

[0218] Accordingly, taking the gradient calculation of the encoder weight eW as an example, the calculation method in the embodiment of the present invention can be expressed as the following formula (43):

[0219]

[0220] Among them, Reconstruction data The partial derivatives of the encoder weights eW are further expanded using the chain rule.

[0221] It is easy to understand that the embodiment of the present invention constructs a preset feature dimensionality reduction model by repeating the above steps until a preset number of iterations is reached, such as a preset number of iterations of 1000 times.

[0222] In some embodiments of the present invention, before performing the step of classifying the remote sensing dimensionality reduction data by a preset classifier to obtain the remote sensing data classification result, the remote sensing data classification method provided by the embodiment of the present invention further includes but is not limited to the following steps:

[0223] Initialize the classifier parameters and construct the first classifier, wherein the classifier parameters include a weight vector, a bias term, and a penalty parameter.

[0224] The parameters of the first classifier are updated according to the random selection mechanism and the time delay factor to obtain a preset classifier.

[0225] In this specific embodiment, the embodiment of the present invention inputs the reduced-dimensional data into the classifier to train the classifier. Since the traditional support vector machine algorithm usually adopts a synchronous update method when updating parameters, that is, the update of all support vectors is completed within one iteration cycle. In the embodiment of the present invention, the support vector machine allows the parameters of the support vector to be updated independently at different times through an asynchronous update mechanism, thereby speeding up the training speed and improving the model's adaptability to new data. In addition, in the traditional support vector machine algorithm, the penalty parameter C is fixed to balance the weight of the classification interval and the classification error, while the embodiment of the present invention adopts an adaptive adjustment mechanism to dynamically adjust the penalty parameter C according to the classification error rate during the training process, so as to more flexibly adapt to the characteristics of the data and improve the accuracy of classification. Specifically, the embodiment of the present invention first initializes the classifier parameters to construct a first classifier. Among them, the classifier parameters in the embodiment of the present invention include a weight vector, a bias term, and a penalty parameter. Exemplarily, in the embodiment of the present invention, the reduced-dimensional data set is in The feature vector representing the reduced-dimensional data, y i is the corresponding category label. The goal of the embodiment of the present invention is to learn a classification model f(p) = sign(w·p+b), where w is a weight vector and b is a bias term. Next, when the classifier parameters are initialized, the formula for initializing w and b is shown in the following formula (44):

[0226]

[0227] Furthermore, the embodiment of the present invention updates the parameters of the first classifier according to the random selection mechanism and the time delay factor to construct a preset classifier. Specifically, the embodiment of the present invention adopts an asynchronous update strategy in each training cycle to randomly select one or more support vectors for parameter update, rather than waiting for the entire data set to be processed. Exemplarily, in each iteration, the embodiment of the present invention selects a sample (p j ,y j ) to update the parameters, and the update rule is based on the gradient descent method. At the same time, the embodiment of the present invention adopts an asynchronous strategy, using a random selection mechanism and a time delay factor Δt to enhance the generalization ability of the model and the ability to adapt to different data distributions, which can be expressed as the following formula (45):

[0228]

[0229] Where L is the loss function, η s is the learning rate, α s is a hyperparameter that controls the effect of time delay, Δw and Δb are the time delay adjustment items for weight and bias, respectively. For example, in this embodiment of the present invention, the learning rate η is set to 0.01, α s is set to 0.1.

[0230] Accordingly, in the embodiment of the present invention, the time delay adjustment terms Δw and Δb are simulated by using attenuation factors. When the fixed time delay factor α s , then the calculation method of Δw and Δb can be expressed as the following formula (46):

[0231]

[0232] Among them, L prev Refers to the gradient of the loss function of the previous iteration.

[0233] Based on this, α s It not only serves as a regulating factor for time delay, but also regulates the influence of the previous gradient on the current update. Furthermore, the embodiment of the present invention regularizes the loss function, that is, adopts a regularized loss function to prevent overfitting and improve the generalization ability of the model, as shown in the following formula (47):

[0234]

[0235] Where C is the regularization parameter, which controls the balance between the error term and the regularization term.

[0236] In addition, when solving the gradient of weight w When the regularization term Its gradient is calculated as w. For the error term max(0,1-y i(w·p i +b)) 2 , when 1-y i (w·p i +b)>0, the gradient is -2y i p i (1-y i (w·p i + b)); otherwise, the gradient is 0. The calculation method of can be expressed as the following formula (48):

[0237]

[0238] In addition, when solving the gradient of bias b For the error term, when 1-y i (w·p i +b)>0, the gradient is -2y i (1-y i (w·p i +b)); otherwise, the gradient is 0. Since the regularization term does not depend on b, The calculation method of can be expressed as the following formula (49):

[0239]

[0240] Next, the embodiment of the present invention dynamically adjusts the penalty parameter C according to the current classification error rate to optimize the classification performance. In the embodiment of the present invention, if the classification error rate increases, the penalty parameter is reduced. Accordingly, the embodiment of the present invention repeatedly performs the above steps until a preset maximum number of iterations is reached, such as a preset maximum number of iterations of 1000, thereby constructing a preset classifier.

[0241] It is easy to understand that after the training of each model is completed, the embodiment of the present invention processes new remote sensing data samples through the trained models and classifies these samples. For example, the embodiment of the present invention extracts features from the remote sensing data set to be classified by using a preset feature extraction model obtained by training the eddy optimization neural network algorithm, and performs data dimension reduction on the extracted remote sensing feature data by using a preset feature dimension reduction model obtained by training the autoencoder neural network algorithm based on collaborative constraints to obtain remote sensing dimension reduction data. Then, the embodiment of the present invention inputs the obtained remote sensing feature data into a preset classifier obtained by training the autoencoder neural network algorithm based on collaborative constraints to classify the remote sensing data and obtain remote sensing data classification results. Accordingly, during the training process, the embodiment of the present invention obtains detailed images of the surface through high-precision remote sensing technology, and accurately classifies and annotates these images. At the same time, since the collection, acquisition, annotation and preprocessing of training data are time-consuming and labor-intensive, and insufficient training samples are likely to lead to poor model generalization ability, while affecting the accuracy of the model. Therefore, the embodiment of the present invention obtains a heuristic generative adversarial network model through training of a generative adversarial network algorithm based on heuristic games, so as to perform data expansion on the labeled data set through the network model to obtain a preset remote sensing training data set. Next, the embodiment of the present invention inputs the preset remote sensing training data set into the feature extraction neural network model to train the feature extraction model, and obtains a preset feature extraction model. Next, the embodiment of the present invention inputs the features after feature extraction into the first autoencoder neural network model to train the feature dimensionality reduction model, and obtains a preset feature dimensionality reduction model. Finally, the embodiment of the present invention inputs the reduced dimensionality data into a classifier to train the classifier, thereby obtaining a preset classifier.

[0242] Accordingly, the embodiment of the present invention introduces a heuristic strategy so that the generator can more accurately simulate the distribution of remote sensing data, improve the diversity and authenticity of the generated samples, and improve the generalization ability and classification accuracy of the model. At the same time, the neural network parameters are optimized by the eddy optimization neural network algorithm to solve the gradient vanishing and explosion problems, and improve the stability of training and model performance. In addition, the embodiment of the present invention uses inverse reinforcement learning to optimize the dimensionality reduction process based on the autoencoder with collaborative constraints, which can simultaneously consider the reconstruction quality and feature compression, ensure the retention of key information, and improve classification accuracy. Furthermore, the support vector machine with an asynchronous update mechanism in the embodiment of the present invention improves the training speed and the model's adaptability to new data by asynchronously updating and adaptively adjusting the penalty parameters.

[0243] See also Figure 3 The present application also provides a remote sensing data classification system, which can implement the above remote sensing data classification method. The system includes:

[0244] The first module 210 is used to obtain a remote sensing data set to be classified.

[0245] The second module 220 is used to extract features from the unclassified remote sensing data set through a preset feature extraction model to obtain remote sensing feature data. The preset feature extraction model is obtained by training the eddy optimization neural network algorithm.

[0246] The third module 230 is used to input the remote sensing feature data into a preset feature dimension reduction model to perform data dimension reduction to obtain remote sensing dimension reduction data. The preset feature dimension reduction model is obtained by training an autoencoder neural network algorithm based on collaborative constraints.

[0247] The fourth module 240 is used to classify the remote sensing dimension reduction data by using a preset classifier to obtain a remote sensing data classification result. The preset classifier is obtained by training a support vector machine classification algorithm based on an asynchronous update mechanism.

[0248] It can be understood that the contents of the above method embodiments are all applicable to the present system embodiments, the functions specifically implemented by the present system embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0249] The embodiment of the present application also provides an electronic device, the electronic device includes a memory and a processor, the memory stores a computer program, and the processor implements the above remote sensing data classification method when executing the computer program. The electronic device can be any intelligent terminal including a tablet computer, a car computer, etc.

[0250] It can be understood that the contents of the above method embodiments are all applicable to the present device embodiments, the functions specifically implemented by the present device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0251] See also Figure 4 , Figure 4 The hardware structure of an electronic device of another embodiment is illustrated, and the electronic device includes:

[0252] The processor 310 may be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present application;

[0253] The memory 320 can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 320 can store an operating system and other application programs. When the technical solution provided in the embodiment of this specification is implemented by software or firmware, the relevant program code is stored in the memory 320, and the processor 310 calls and executes the remote sensing data classification method of the embodiment of this application;

[0254] Input / output interface 330, used to implement information input and output;

[0255] Communication interface 340, used to realize communication interaction between the device and other devices, which can be realized through wired mode (such as USB, network cable, etc.) or wireless mode (such as mobile network, WIFI, Bluetooth, etc.);

[0256] bus 350 , which transmits information between the various components of the device (e.g., processor 310 , memory 320 , input / output interface 330 , and communication interface 340 );

[0257] The processor 310 , the memory 320 , the input / output interface 330 and the communication interface 340 are connected to each other in communication within the device via the bus 350 .

[0258] An embodiment of the present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the remote sensing data classification method described above is implemented.

[0259] It can be understood that the contents of the above method embodiments are all applicable to the present storage medium embodiments, the functions specifically implemented by the present storage medium embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0260] The memory, as a non-transient computer-readable storage medium, can be used to store non-transient software programs and non-transient computer executable programs. In addition, the memory may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory may optionally include a memory remotely disposed relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0261] The embodiments described in the embodiments of the present application are intended to more clearly illustrate the technical solutions of the embodiments of the present application and do not constitute a limitation on the technical solutions provided in the embodiments of the present application. Those skilled in the art will appreciate that with the evolution of technology and the emergence of new application scenarios, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.

[0262] Those skilled in the art will appreciate that the technical solutions shown in the figures do not constitute a limitation on the embodiments of the present application, and may include more or fewer steps than shown in the figures, or a combination of certain steps, or different steps.

[0263] The device embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separated, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0264] Those skilled in the art will appreciate that all or some of the steps in the methods disclosed above, and the functional modules / units in the systems and devices may be implemented as software, firmware, hardware, or a suitable combination thereof.

[0265] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. 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 that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0266] It should be understood that in the present application, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0267] 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. For example, the device embodiments described above are only schematic. For example, the division of the above 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 integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0268] The units described above as separate components may or may not be physically separated, and the components shown 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 solution of this embodiment.

[0269] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0270] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including multiple instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, referred to as ROM), random access memory (Random Access Memory, referred to as RAM), disk or optical disk and other media that can store programs.

[0271] The preferred embodiments of the present invention are described above with reference to the accompanying drawings, but the scope of the rights of the present invention is not limited thereto. Any modification, equivalent substitution and improvement made by a person skilled in the art without departing from the scope and essence of the present invention should be within the scope of the rights of the present invention.

Claims

1. A remote sensing data classification method, characterized in that: The method comprises: Obtain remote sensing data sets to be classified; Extracting features from the remote sensing data set to be classified by a preset feature extraction model to obtain remote sensing feature data; wherein the preset feature extraction model is obtained by training a vortex optimization neural network algorithm; Inputting the remote sensing feature data into a preset feature dimensionality reduction model to perform data dimensionality reduction, thereby obtaining remote sensing dimensionality reduction data; wherein the preset feature dimensionality reduction model is obtained by training an autoencoder neural network algorithm based on collaborative constraints; The remote sensing dimension reduction data is classified by a preset classifier to obtain a remote sensing data classification result; wherein the preset classifier is obtained by training a support vector machine classification algorithm based on an asynchronous update mechanism; Before performing the step of extracting features from the remote sensing data set to be classified by using a preset feature extraction model to obtain remote sensing feature data, the method further includes: Build a feature extraction neural network model; Inputting a preset remote sensing training data set into the feature extraction neural network model to perform feature extraction to obtain training feature data; Determine target parameter data through the eddy optimization neural network algorithm according to the training feature data; Assigning parameters to the feature extraction neural network model according to the target parameter data to construct the preset feature extraction model; Wherein, determining the target parameter data by using the eddy optimization neural network algorithm according to the training feature data includes: Constructing a first vortex set in a parameter space; wherein each vortex in the first vortex set includes a corresponding parameter configuration; A loss function is calculated according to the training feature data to obtain a loss value, and the performance of the parameter configuration of each vortex is determined by the loss value, so as to sort the vortices according to the performance to obtain vortex sorting information; Perform vortex dynamic update according to the vortex sorting information to obtain a second vortex set; Performing a preset vortex operation on the second vortex set according to a preset vortex threshold to obtain a third vortex set; wherein the preset vortex operation includes vortex merging and vortex splitting; Selecting vortices that meet preset performance conditions in the third vortex set for local search to adjust parameters, thereby obtaining a fourth vortex set; The vortices whose life values ​​in the fourth vortex set are lower than a preset life threshold are released, and the loss value obtained by calculating the loss function according to the training feature data is returned, and the performance of the parameter configuration of each vortex is determined by the loss value, so as to sort the vortices by the performance and obtain the vortex sorting information, until the number of iterations is greater than the preset iteration threshold, and the target parameter data is obtained; wherein the life value is determined by a preset attenuation rule; Before executing the step of inputting the remote sensing feature data into a preset feature dimension reduction model to perform data dimension reduction to obtain remote sensing dimension reduction data, the method further includes: Constructing a first autoencoding neural network model; wherein the first autoencoding neural network model includes an encoder and a decoder; Performing feature compression on the preset training data by the encoder, and reconstructing the data by the decoder to obtain reconstructed data; Performing loss calculation according to the reconstruction data and the preset training data to obtain preset loss data; wherein the preset loss data includes reconstruction loss data and collaborative constraint loss data; Adjusting parameters of the first autoencoder neural network model according to the inverse reinforcement learning strategy to obtain a second autoencoder neural network model; According to the preset loss data, the model parameters of the second autoencoder neural network model are updated by a back propagation algorithm to obtain the preset feature dimensionality reduction model; Wherein, before executing the step of classifying the remote sensing dimension reduction data by a preset classifier to obtain a remote sensing data classification result, the method further includes: Initializing classifier parameters to construct a first classifier; wherein the classifier parameters include a weight vector, a bias term, and a penalty parameter; The parameters of the first classifier are updated according to the random selection mechanism and the time delay factor to obtain the preset classifier.

2. The method according to claim 1, characterized in that Before executing the step of inputting the preset remote sensing training data set into the feature extraction neural network model to extract features to obtain training feature data, the method further includes: Collect original remote sensing data and annotate the data to obtain the original data set; The original data set is expanded by using a heuristic generative adversarial network model to obtain the preset remote sensing training data set; wherein the heuristic generative adversarial network model is obtained by training a generative adversarial network algorithm based on heuristic games.

3. The method according to claim 2, characterized in that Before executing the step of performing data expansion on the original data set by using the heuristic generative adversarial network model to obtain the preset remote sensing training data set, the method further includes: Constructing a preset generative adversarial network model; wherein the preset generative adversarial network model includes a generator, a discriminator, and a first heuristic strategy parameter; Generate remote sensing data samples through the generator, and evaluate the authenticity of the remote sensing data samples through the discriminator to obtain feedback data; Dynamically adjusting the first heuristic strategy parameter according to the feedback data and the data diversity information of the remote sensing data sample to obtain a second heuristic strategy parameter; Optimizing the composite loss function according to the second heuristic strategy parameter to obtain a preset loss function; wherein the composite loss function includes content loss, style loss and discrimination loss; Dynamically adjust the learning rate according to the variance of the feedback data to obtain a learning rate parameter; The parameters of the preset generative adversarial network model are updated according to the learning rate parameter and the preset loss function to obtain the heuristic generative adversarial network model.

4. A remote sensing data classification system, characterized in that: The system comprises: The first module is used to obtain the remote sensing data set to be classified; The second module is used to extract features from the remote sensing data set to be classified through a preset feature extraction model to obtain remote sensing feature data; wherein the preset feature extraction model is obtained by training the eddy optimization neural network algorithm; The third module is used to input the remote sensing feature data into a preset feature dimensionality reduction model to perform data dimensionality reduction to obtain remote sensing dimensionality reduction data; wherein the preset feature dimensionality reduction model is obtained by training an autoencoder neural network algorithm based on collaborative constraints; The fourth module is used to classify the remote sensing dimension reduction data through a preset classifier to obtain a remote sensing data classification result; wherein the preset classifier is obtained by training a support vector machine classification algorithm based on an asynchronous update mechanism; Before performing the step of extracting features from the remote sensing data set to be classified by using a preset feature extraction model to obtain remote sensing feature data, the method further includes: Build a feature extraction neural network model; Inputting a preset remote sensing training data set into the feature extraction neural network model to perform feature extraction to obtain training feature data; Determine target parameter data through the eddy optimization neural network algorithm according to the training feature data; Assigning parameters to the feature extraction neural network model according to the target parameter data to construct the preset feature extraction model; Wherein, determining the target parameter data by using the eddy optimization neural network algorithm according to the training feature data includes: Constructing a first vortex set in a parameter space; wherein each vortex in the first vortex set includes a corresponding parameter configuration; A loss function is calculated according to the training feature data to obtain a loss value, and the performance of the parameter configuration of each vortex is determined by the loss value, so as to sort the vortices according to the performance to obtain vortex sorting information; Perform vortex dynamic update according to the vortex sorting information to obtain a second vortex set; Performing a preset vortex operation on the second vortex set according to a preset vortex threshold to obtain a third vortex set; wherein the preset vortex operation includes vortex merging and vortex splitting; Selecting vortices that meet preset performance conditions in the third vortex set for local search to adjust parameters, thereby obtaining a fourth vortex set; The vortices whose life values ​​in the fourth vortex set are lower than a preset life threshold are released, and the loss value obtained by calculating the loss function according to the training feature data is returned, and the performance of the parameter configuration of each vortex is determined by the loss value, so as to sort the vortices by the performance and obtain the vortex sorting information, until the number of iterations is greater than the preset iteration threshold, and the target parameter data is obtained; wherein the life value is determined by a preset attenuation rule; Before executing the step of inputting the remote sensing feature data into a preset feature dimension reduction model to perform data dimension reduction to obtain remote sensing dimension reduction data, the method further includes: Constructing a first autoencoding neural network model; wherein the first autoencoding neural network model includes an encoder and a decoder; Performing feature compression on the preset training data by the encoder, and reconstructing the data by the decoder to obtain reconstructed data; Performing loss calculation according to the reconstruction data and the preset training data to obtain preset loss data; wherein the preset loss data includes reconstruction loss data and collaborative constraint loss data; Adjusting parameters of the first autoencoder neural network model according to the inverse reinforcement learning strategy to obtain a second autoencoder neural network model; According to the preset loss data, the model parameters of the second autoencoder neural network model are updated by a back propagation algorithm to obtain the preset feature dimensionality reduction model; Before executing the step of classifying the remote sensing dimension reduction data by a preset classifier to obtain the remote sensing data classification result, the method further includes: Initializing classifier parameters to construct a first classifier; wherein the classifier parameters include a weight vector, a bias term, and a penalty parameter; The parameters of the first classifier are updated according to the random selection mechanism and the time delay factor to obtain the preset classifier.

5. An electronic device, characterized in that: include: at least one processor; at least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements the method according to any one of claims 1 to 3.

6. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 3 is implemented.