Water environment quality remote sensing analysis method, system, electronic equipment and storage medium

Abstract

The present application discloses a remote sensing analysis method, system, electronic device and storage medium for water environment quality, the method comprising: obtaining water environment monitoring data; wherein the water environment monitoring data comprises multi-source remote sensing data; inputting the multi-source remote sensing data into a preset dimensionality reduction model for dimensionality reduction processing to obtain dimensionality reduction data; wherein the preset dimensionality reduction model is obtained by training a multi-dimensional adaptive encoder model; inputting the dimensionality reduction data into a preset feature extraction model for feature extraction to obtain feature data; wherein the preset feature extraction model is obtained by training a preset neural network through an ecosystem optimization algorithm; inputting the feature data into a preset classifier for classification to obtain a quality assessment result. The embodiments of the present application can accurately classify and evaluate the quality level of the water environment, and effectively improve the generalization performance and robustness of data processing. The present application can be widely used in the field of water quality monitoring technology.

Description

Technical Field

[0001] The present application relates to the field of water quality monitoring technology, and in particular to a water environment quality remote sensing analysis method, system, electronic equipment and storage medium. Background Art

[0002] With the accelerated development of industrialization and urbanization, water pollution has become increasingly serious, which has had a significant impact on human health and ecosystems. Among related technologies, although machine learning and deep learning technologies have made certain progress in remote sensing water quality monitoring, existing methods still have some shortcomings. For example, the comprehensiveness and accuracy of monitoring results are low. Secondly, when faced with large-scale, high-dimensional remote sensing data, existing data processing methods often have problems such as insufficient generalization ability, limited model stability and robustness. In addition, the efficiency and practicality of water quality monitoring are also limited.

[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 water environment quality remote sensing analysis method, system, electronic device and storage medium, which can accurately classify and evaluate the quality level of the water environment and effectively improve the generalization performance and robustness of data processing.

[0005] To achieve the above purpose, one aspect of an embodiment of the present application provides a water environment quality remote sensing analysis method, the method comprising:

[0006] Acquiring water environment monitoring data; wherein the water environment monitoring data includes multi-source remote sensing data;

[0007] Inputting the multi-source remote sensing data into a preset dimensionality reduction model for dimensionality reduction processing to obtain dimensionality reduction data; wherein the preset dimensionality reduction model is obtained by training a multi-dimensional adaptive encoder model;

[0008] Inputting the dimension reduction data into a preset feature extraction model to extract features and obtain feature data; wherein the preset feature extraction model is obtained by training a preset neural network through an ecosystem optimization algorithm;

[0009] The feature data is input into a preset classifier for classification to obtain a quality assessment result.

[0010] In some embodiments, before executing the step of inputting the multi-source remote sensing data into a preset dimensionality reduction model for dimensionality reduction processing to obtain dimensionality reduction data, the method further includes:

[0011] Constructing the multi-dimensional adaptive encoder model; wherein the multi-dimensional adaptive encoder model includes a preset encoder and a preset decoder;

[0012] Preprocessing a preset remote sensing data set to obtain first remote sensing data; wherein the preset remote sensing data set includes remote sensing source data from a plurality of different sources;

[0013] Integrating the first remote sensing data by a weighted fusion strategy to obtain a first fusion feature vector;

[0014] Dynamically adjusting the network structure and parameters of the preset encoder according to the first fused feature vector to compress and encode the first fused feature vector to obtain a low-dimensional feature;

[0015] The feature weights of the low-dimensional features are evaluated through an attention mechanism to obtain weighted features;

[0016] Dynamically adjusting the network structure and parameters of the preset decoder according to the low-dimensional features to decode the weighted features and reconstruct a second fused feature vector;

[0017] According to the first fused feature vector and the second fused feature vector, the network parameters of the multi-dimensional adaptive encoder model are optimized by a preset optimization algorithm to obtain the preset dimensionality reduction model; wherein the preset optimization algorithm includes back propagation and gradient descent algorithms.

[0018] In some embodiments, before executing the step of inputting the dimension reduction data into a preset feature extraction model to extract features to obtain feature data, the method further includes:

[0019] Constructing the preset neural network;

[0020] Initializing an initial ecosystem; wherein the initial ecosystem includes a plurality of first populations, and the first populations include a plurality of first individuals;

[0021] Calculating the fitness of each of the first individuals in the initial ecosystem to obtain first fitness data;

[0022] Determining the probability of the first individual being selected according to the first fitness data;

[0023] Performing genetic operations according to the probabilities to obtain a plurality of second populations; wherein the genetic operations include crossover operations and mutation operations;

[0024] updating the fitness of each second individual in the second population according to the ecological dynamic information to obtain second fitness data; wherein the ecological dynamic information includes environmental change information and ecological interaction information;

[0025] The parameters of the preset neural network are updated according to the second fitness data and the ecological dynamic information to obtain the preset feature extraction model.

[0026] In some embodiments, before executing the step of inputting the feature data into a preset classifier for classification to obtain a quality assessment result, the method further includes:

[0027] Construct an ELM model;

[0028] The extreme learning machine model is trained by a fractional gradient descent algorithm to optimize the weights from the hidden layer to the output layer in the extreme learning machine model to obtain the preset classifier.

[0029] In some embodiments, before executing the step of preprocessing the preset remote sensing data set to obtain the first remote sensing data, the method further includes:

[0030] Collecting spectral data through several remote sensing platforms, and annotating the spectral data to obtain a spectral feature data set;

[0031] The spectral feature data set is expanded by using a preset generative adversarial network model to obtain the preset remote sensing data set.

[0032] In some embodiments, before performing the step of performing data expansion on the spectral feature dataset by using a preset generative adversarial network model to obtain the preset remote sensing dataset, the method further includes:

[0033] Constructing a first generative adversarial network model; wherein the first generative adversarial network model includes a generator network and a discriminator network;

[0034] The generator network and the discriminator network are optimized through adversarial training according to the loss function, and the network parameters of the first generative adversarial network model are updated by the gradient descent method to obtain the preset generative adversarial network model; wherein the input layer input of the generator network includes control noise, and the loss function of the generator network includes a manifold constraint term.

[0035] In some embodiments, inputting the feature data into a preset classifier for classification to obtain a quality assessment result includes:

[0036] Analyzing the characteristic data by the preset classifier to obtain classification information; wherein the classification information includes water environment quality classification level information;

[0037] A quality assessment is performed on the corresponding water environment according to the classification information to obtain the quality assessment result.

[0038] To achieve the above purpose, another aspect of the embodiment of the present application provides a water environment quality remote sensing analysis system, the system comprising:

[0039] The first module is used to obtain water environment monitoring data; wherein the water environment monitoring data includes multi-source remote sensing data;

[0040] The second module is used to input the multi-source remote sensing data into a preset dimensionality reduction model for dimensionality reduction processing to obtain dimensionality reduction data; wherein the preset dimensionality reduction model is obtained by training a multi-dimensional adaptive encoder model;

[0041] The third module is used to input the dimension reduction data into a preset feature extraction model to extract features and obtain feature data; wherein the preset feature extraction model is obtained by training a preset neural network through an ecosystem optimization algorithm;

[0042] The fourth module is used to input the feature data into a preset classifier for classification to obtain a quality assessment result.

[0043] To achieve the above objective, another aspect of an embodiment of the present application provides an electronic device, the electronic device comprising:

[0044] at least one processor;

[0045] at least one memory for storing at least one program;

[0046] When the at least one program is executed by the at least one processor, the at least one processor implements the above method.

[0047] 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.

[0048] The embodiments of the present application include at least the following beneficial effects: The present application provides a method, system, electronic device and storage medium for remote sensing analysis of water environment quality. The scheme obtains water environment monitoring data, including multi-source remote sensing data, and inputs the multi-source remote sensing data into a preset dimensionality reduction model obtained by training a multi-dimensional adaptive encoder model for dimensionality reduction processing to obtain dimensionality reduction data. Then, the embodiment of the present invention inputs the dimensionality reduction data obtained by dimensionality reduction into a preset feature extraction model obtained by training a preset neural network through an ecosystem optimization algorithm for feature extraction processing to obtain feature data. Finally, the embodiment of the present invention inputs the feature data into a preset classifier for classification to obtain a quality assessment result. It is easy to understand that the embodiment of the present invention can effectively improve the generalization performance by optimizing the dimensionality reduction of multi-source remote sensing data, and can provide a more accurate and robust feature representation for remote sensing analysis of water environment quality through adaptive adjustment. In addition, the embodiment of the present invention can effectively improve the robustness and stability of the model to input data disturbances by performing feature extraction processing on a preset feature extraction model obtained by training a preset neural network through an ecosystem optimization algorithm. At the same time, the embodiment of the present invention performs corresponding dimensionality reduction processing by comprehensively utilizing multi-source remote sensing data, extracts features from the data obtained by dimensionality reduction, and then classifies the extracted feature data, so as to accurately classify and evaluate the quality level of the corresponding water environment, thereby realizing quality assessment and analysis of the water environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 is a flow chart of the water environment quality remote sensing analysis method provided in an embodiment of the present application;

[0050] Figure 2 It is a schematic diagram of the preset feature extraction model training process provided in an embodiment of the present application;

[0051] Figure 3 It is a schematic diagram of the structure of the water environment quality remote sensing analysis system provided in the embodiment of the present application;

[0052] 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

[0053] 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 as detailed in the attached claims.

[0054] 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 concept. For example, without departing from the scope of the embodiment 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".

[0055] 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.

[0056] 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.

[0057] 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.

[0058] Deep learning: It is a branch of machine learning that uses multi-level neural network models, such as deep neural networks, to learn and extract high-level abstract features of data. Deep learning can automatically learn feature representations from data without manually designing feature extractors. It models complex relationships and patterns through multi-level nonlinear transformations, thereby achieving learning and prediction of large-scale data.

[0059] Neural network: It is an artificial intelligence model that imitates the structure of the human brain's neural network. It consists of multiple neurons (nodes) forming a multi-layer network. Each neuron receives input from the neurons in the previous layer and performs weighted summation through the corresponding activation function to generate output. Accordingly, the neural network learns the features and patterns of the input data to achieve tasks such as classification, regression, and clustering.

[0060] Generative Adversarial Network (GAN): It is a deep learning model consisting of a generator and a discriminator. The generator is responsible for generating fake samples, while the discriminator is responsible for distinguishing between real samples and fake samples generated by the generator. Accordingly, the training process of the generative adversarial network is a game process, which continuously improves its performance through the competition between the generator and the discriminator.

[0061] With the accelerated development of industrialization and urbanization, water pollution has become increasingly serious, which has had a significant impact on human health and ecosystems. Therefore, the development of efficient and accurate water quality monitoring methods is crucial for the protection and sustainable management of water resources. Among them, traditional water quality monitoring methods mainly rely on ground sampling and laboratory analysis. Although these methods can provide accurate water quality parameters, they have disadvantages such as high cost, low efficiency, and limited temporal and spatial coverage. In order to overcome these limitations, remote sensing technology, as a non-contact monitoring method, has been widely used in the field of water quality monitoring due to its advantages such as wide coverage and short update cycle. Among them, the application of remote sensing technology in water environment quality monitoring includes the use of sensors carried by satellites, drones and other platforms to collect spectral data of water bodies, and infer water quality parameters such as turbidity, chlorophyll content, and organic matter content of water bodies by analyzing the reflection and absorption characteristics of water bodies in different bands. However, since the spectral characteristics of water bodies are complex and easily affected by environmental factors, how to accurately extract useful water quality information from a large amount of remote sensing data has become a major challenge facing this field.

[0062] In recent years, with the rapid development of artificial intelligence technology, machine learning and deep learning technologies have been introduced into remote sensing water quality monitoring, improving the degree of automation of data processing and the accuracy of monitoring. However, among related technologies, although machine learning and deep learning technologies have made certain progress in remote sensing water quality monitoring, existing methods still have some shortcomings. For example, most studies rely on remote sensing data from a single source, ignoring the complementarity of data from different sensors and platforms, which limits the comprehensiveness and accuracy of monitoring results. Secondly, when faced with large-scale, high-dimensional remote sensing data, existing data processing methods often have problems with insufficient generalization ability, limited model stability and robustness. In addition, steps such as data preprocessing, feature extraction, and classification usually require a lot of manual intervention and expertise, which limits the efficiency and practicality of water quality monitoring.

[0063] In view of this, a method, system, electronic device and storage medium for remote sensing analysis of water environment quality are provided in the embodiments of the present application. The scheme obtains water environment monitoring data including multi-source remote sensing data, and inputs the multi-source remote sensing data into a preset dimensionality reduction model for dimensionality reduction processing to obtain dimensionality reduction data. Then, the embodiment of the present invention inputs the dimensionality reduction data obtained by dimensionality reduction into a feature extraction model for feature extraction processing to obtain feature data. Finally, the embodiment of the present invention inputs the feature data into a preset classifier for classification to obtain a quality assessment result. Accordingly, the embodiment of the present invention can effectively improve the generalization performance by optimizing the dimensionality reduction of multi-source remote sensing data. In addition, the embodiment of the present invention performs corresponding dimensionality reduction processing by comprehensively utilizing multi-source remote sensing data, and extracts features from the data obtained by dimensionality reduction, and then classifies the extracted feature data, thereby realizing the quality assessment and analysis of the water environment, and can accurately classify and assess the quality level of the corresponding water environment, and effectively improves the generalization performance and robustness of data processing.

[0064] The water environment quality remote sensing analysis method provided in the embodiment of the present application relates to the field of water quality monitoring technology. The water environment quality remote sensing analysis method provided in the embodiment of the present application can be applied to a terminal, can also be applied to a server, and can also 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 water environment quality remote sensing analysis method, etc., but is not limited to the above forms.

[0065] 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.

[0066] Figure 1 is an optional flow chart of the water environment quality remote sensing analysis method provided in the embodiment of the present application, Figure 1 The method may include but is not limited to steps S110 to S140.

[0067] Step S110, acquiring water environment monitoring data, wherein the water environment monitoring data includes multi-source remote sensing data.

[0068] Step S120, inputting the multi-source remote sensing data into a preset dimensionality reduction model for dimensionality reduction processing to obtain dimensionality reduction data. The preset dimensionality reduction model is obtained by training a multi-dimensional adaptive encoder model.

[0069] Step S130, inputting the dimension reduction data into a preset feature extraction model to extract features and obtain feature data. The preset feature extraction model is obtained by training a preset neural network through an ecosystem optimization algorithm.

[0070] Step S140: input the feature data into a preset classifier for classification to obtain a quality assessment result.

[0071] During the working process of this specific embodiment, the embodiment of the present invention first obtains water environment monitoring data. Specifically, the water environment monitoring data in the embodiment of the present invention includes multi-source remote sensing data. Among them, the water environment monitoring data in the embodiment of the present invention comes from multiple remote sensing platforms, including but not limited to satellite images, images taken by drones, and spectral data collected on the ground. Then, the embodiment of the present invention inputs the multi-source remote sensing data into a preset dimensionality reduction model for dimensionality reduction processing to obtain dimensionality reduction data. Specifically, the preset dimensionality reduction model in the embodiment of the present invention includes a multi-dimensional adaptive autoencoder, that is, the preset dimensionality reduction model is obtained by training a multi-dimensional adaptive encoder model. Accordingly, the embodiment of the present invention reduces the dimensionality of multi-source remote sensing data through a preset dimensionality reduction model, and through adaptive adjustment, it can better capture the complex adjustments and internal connections in the multi-source remote sensing data, thereby providing a more accurate and robust feature representation for remote sensing analysis of water environment quality. Further, the embodiment of the present invention inputs the dimensionality reduction data into a preset feature extraction model for feature extraction to obtain feature data. Among them, the embodiment of the present invention uses a neural network as a preset feature extraction model to extract features from the dimensionality reduction data obtained by dimensionality reduction, and trains the preset neural network through an ecosystem optimization algorithm to obtain a preset feature extraction model. It is easy to understand that the embodiment of the present invention introduces an ecosystem optimization algorithm to train a preset neural network to alleviate the tendency to fall into local optimal situations during parameter optimization, prevent premature convergence and local optimality, and thus effectively improve the robustness and stability of the model to input data disturbances. Finally, the embodiment of the present invention inputs the feature data into a preset classifier for classification to obtain a quality assessment result. Specifically, the embodiment of the present invention classifies the data after feature extraction through a preset classifier to perform quality analysis on the water environment and determine the quality assessment result. The embodiment of the present invention improves the generalization performance of data processing by performing dimensionality reduction processing on multi-source remote sensing data, and at the same time extracts features through a preset feature extraction model, and classifies the extracted feature data through a preset classifier, so as to more accurately achieve the quality level assessment of the water environment.

[0072] In some embodiments of the present invention, before executing the step of inputting multi-source remote sensing data into a preset dimensionality reduction model for dimensionality reduction processing to obtain dimensionality reduction data, the water environment quality remote sensing analysis method provided by the embodiment of the present invention further includes but is not limited to the following steps:

[0073] A multi-dimensional adaptive encoder model is constructed, wherein the multi-dimensional adaptive encoder model includes a preset encoder and a preset decoder.

[0074] The preset remote sensing data set is preprocessed to obtain first remote sensing data, wherein the preset remote sensing data set includes remote sensing source data from a plurality of different sources.

[0075] The first remote sensing data is integrated through a weighted fusion strategy to obtain a first fusion feature vector.

[0076] The network structure and parameters of the preset encoder are dynamically adjusted according to the first fused feature vector to compress and encode the first fused feature vector to obtain a low-dimensional feature.

[0077] The feature weights of low-dimensional features are evaluated through the attention mechanism to obtain weighted features.

[0078] The network structure and parameters of the preset decoder are dynamically adjusted according to the low-dimensional features to decode the weighted features and reconstruct the second fused feature vector.

[0079] The network parameters of the multi-dimensional adaptive encoder model are optimized by a preset optimization algorithm according to the first fused feature vector and the second fused feature vector to obtain a preset dimensionality reduction model, wherein the preset optimization algorithm includes a back propagation algorithm and a gradient descent algorithm.

[0080] In this specific embodiment, the embodiment of the present invention first constructs a multi-dimensional adaptive encoder model, including a preset encoder and a preset decoder. Among them, the embodiment of the present invention obtains a preset dimensionality reduction model by training the multi-dimensional adaptive encoder model, and reduces the dimensionality of the data by a multi-dimensional adaptive autoencoder, which can effectively reduce the dimensionality of multivariate data features, and through adaptive adjustment, it can better obtain complex features in multi-source remote sensing data, thereby improving the accuracy and robustness of the feature representation of remote sensing analysis of water environment quality. Next, the embodiment of the present invention preprocesses the preset remote sensing data set to obtain the first remote sensing data. Specifically, the preset remote sensing data set in the embodiment of the present invention refers to a remote sensing data set used to train a preset dimensionality reduction model, including remote sensing source data from several different sources, such as pre-acquired multi-source remote sensing data. Exemplarily, the embodiment of the present invention preprocesses the preset remote sensing data set, and the data sets collected from different remote sensing sources are respectively {X (1) ,X (2) ,...,X (k)}, where k represents the number of remote sensing data sources, X (i) is the feature matrix of the ith data source. Accordingly, for each data source X (i) , the normalization process can be expressed as follows:

[0081]

[0082] Among them, The data are normalized.

[0083] Next, the embodiment of the present invention integrates the first remote sensing data through a weighted fusion strategy to obtain a first fused feature vector. Specifically, the embodiment of the present invention adopts a weighted fusion method to integrate features from different sources together to form a fused feature vector, that is, a first fused feature vector. Exemplarily, the first fused feature vector F is formed in the embodiment of the present invention. fuse The formula is shown in the following formula (2):

[0084]

[0085] Among them, w i is the weight of the ith data source. It should be noted that the weights w of different data sources in the embodiment of the present invention are i is dynamically calculated based on the contribution and relevance of the data source. i is the relevance score of the i-th data source, and the calculation formula is shown in the following formula (3):

[0086]

[0087] Among them, Cov(X (i) ,Y) represents the feature X of the i-th data source (i) The covariance with the target variable Y (such as water quality parameters), and σ Y They are X (i) And the standard deviation of Y. Then, in the embodiment of the present invention, w i According to R i The calculation is as shown in the following formula (4):

[0088]

[0089] It is easy to understand that in the embodiment of the present invention, data sources with high correlation will obtain higher weights. Further, the embodiment of the present invention dynamically adjusts the network structure and parameters of the preset encoder according to the first fused feature vector to compress and encode the first fused feature vector to obtain low-dimensional features. Specifically, the backup embodiment dynamically adjusts the network structure and parameters of the preset encoder according to the dimension and characteristics of the fused feature, compresses and encodes the fused feature, that is, the first fused feature vector, to obtain a corresponding low-dimensional feature representation. Exemplarily, in the embodiment of the present invention, F encode is the encoded feature, E is the encoder parameter, then according to the first fused feature vector F fuse The dimension and complexity of the encoder E are dynamically selected. e and the number of neurons in each layer n e , as shown in the following formula (5):

[0090] l e ,ne =φ(F fuse ) (5)

[0091] Wherein, φ() is a decision function based on rules. Accordingly, in the embodiment of the present invention, the encoder E converts F fuse Encoded as low-dimensional feature F encode , as shown in the following formula (6):

[0092] F encode =σ(E(F fuse )) (6)

[0093] Where σ() includes the ReLU activation function.

[0094] It should be noted that, in some embodiments of the present invention, the number of layers of the preset encoder is l e and the number of neurons in each layer n e The choice is based on the dimension d of the fusion feature fuse and complexity c fuse The embodiment of the present invention adopts a heuristic function to determine l e and n e , as shown in equations (7) and (8):

[0095]

[0096]

[0097] Among them, Indicates the rounding up operation, c fuse It is estimated by the number of clusters obtained by clustering analysis of the fused features.

[0098] Furthermore, the embodiment of the present invention uses the attention mechanism to evaluate the feature weights of low-dimensional features to obtain weighted features. Specifically, the embodiment of the present invention uses the attention mechanism to evaluate the importance of each dimension in the fusion feature to adjust the feature weights and optimize the information content represented by the low-dimensional features. For example, in the embodiment of the present invention, A is the feature importance weight matrix. For the encoded feature F encode , the calculation formula for calculating its feature importance weight is shown in the following formula (9):

[0099] A=softmax(W a ·F encode +b a ) (9)

[0100] Among them, W a and b ais the weight and bias of the attention layer, and softmax() is the softmax function, which ensures that the sum of the weights is 1. Then, the embodiment of the present invention uses the weight matrix A to encode the feature F encode Weighted, get the weighted feature F weighted , as shown in the following formula (10):

[0101] F weighted =A⊙F encode (10)

[0102] It should be noted that, in some embodiments of the present invention, the weight W in the attention mechanism a and bias b a It is learned through a small neural network layer whose input is the encoded feature F encode , as shown in the following formula (11):

[0103] W a ,b a =MLP(F encode θ mlp ) (11)

[0104] Where MLP represents a multi-layer perceptron, θ mlp is its parameter. Accordingly, the multilayer perceptron is designed as a single hidden layer, and the number of neurons in the hidden layer is F encode half the dimension to effectively learn W a and b a .

[0105] Furthermore, the embodiment of the present invention dynamically adjusts the network structure and parameters of the preset decoder according to the low-dimensional features to decode the weighted adjustment, thereby reconstructing the second fused feature vector. Specifically, corresponding to the adaptive coding, the embodiment of the present invention dynamically adjusts the network structure and parameters of the decoder according to the low-dimensional feature representation to reconstruct the fused feature vector. Exemplarily, in the embodiment of the present invention, F decode is the decoded feature, D is the decoder parameter, according to F weighted The dimension and complexity of the decoder D are dynamically selected. d and the number of neurons in each layer n d , as shown in the following formula (12):

[0106] l d ,n d =ψ(F weighted ) (12)

[0107] Wherein, ψ() is a decision function based on rules. Accordingly, in the decoding process of the embodiment of the present invention, the decoder D converts F weighted Decoded into reconstructed feature F decode, as shown in the following formula (13):

[0108] F decode =σ(D(F weighted )) (13)

[0109] Furthermore, the embodiment of the present invention optimizes the network parameters of the multi-dimensional adaptive encoder model through a preset optimization algorithm according to the first fused feature vector and the second fused feature vector, thereby obtaining a preset dimensionality reduction model. Specifically, the preset optimization algorithm in the embodiment of the present invention includes a back propagation and a gradient descent algorithm. Accordingly, the embodiment of the present invention calculates the difference between the reconstructed feature and the original fused feature, that is, the difference between the first fused feature vector and the second fused feature vector, uses the mean square error as the loss function, and optimizes the network parameters of the preset dimensionality reduction model through the back propagation and gradient descent algorithms. Exemplarily, in the embodiment of the present invention, L is the loss function used to calculate F decode With the original fusion feature F fuse The difference between them, the loss function is calculated as follows (14):

[0110]

[0111] Accordingly, the embodiment of the present invention updates the parameters of E and D through back propagation and gradient descent algorithms to minimize the loss L. Among them, the embodiment of the present invention uses dimensionality reduction to preset the output of the dimensionality reduction model as a low-dimensional feature representation after processing by a multi-dimensional adaptive autoencoder, and these features are compact and contain key information of the original multi-source data.

[0112] In some embodiments of the present invention, before executing the step of inputting the dimension reduction data into a preset feature extraction model to extract features and obtain feature data, the water environment quality remote sensing analysis method provided by the embodiment of the present invention further includes but is not limited to the following steps:

[0113] Build a preset neural network.

[0114] Initialize the initial ecosystem, wherein the initial ecosystem includes a number of first populations, and the first population includes a number of first individuals.

[0115] The fitness of each first individual in the initial ecosystem is calculated to obtain first fitness data.

[0116] The probability of the first individual being selected is determined according to the first fitness data.

[0117] Genetic operations are performed according to probability to obtain a number of second populations, wherein the genetic operations include crossover operations and mutation operations.

[0118] The fitness of each second individual in the second population is updated according to the ecological dynamic information to obtain second fitness data, wherein the ecological dynamic information includes environmental change information and ecological interaction information.

[0119] The parameters of the preset neural network are updated according to the second fitness data and the ecological dynamic information to obtain a preset feature extraction model.

[0120] In this specific embodiment, the embodiment of the present invention first constructs a preset neural network. Specifically, the embodiment of the present invention adopts a multi-layer neural network as a feature extraction model. Exemplarily, the structure of the neural network constructed in the embodiment of the present invention includes an input layer, a hidden layer, and an output layer. Among them, the input layer in the embodiment of the present invention sets the number of neurons according to the number of features of the input data. Correspondingly, the embodiment of the present invention includes 2 hidden layers, each of which has 50 and 80 neurons, respectively. In addition, the number of neurons in the output layer in the embodiment of the present invention is set to 100. Further, the embodiment of the present invention optimizes the neural network parameters in a manner based on ecosystem optimization. Among them, in the ecosystem optimization algorithm, each parameter of the neural network (weight w and bias b) is regarded as a biological individual in an ecosystem, and its survival and reproduction ability is determined by its fitness (i.e., the difference from the expected output). These parameters are optimized by simulating the natural selection and genetic variation process in the ecosystem, thereby improving the performance of the neural network. In addition, the ecosystem optimization algorithm in the embodiment of the present invention introduces a diversity maintenance mechanism based on the niche theory to prevent premature convergence and local optimality, and explore a broader solution space by maintaining the diversity of the parameter population. Reference Figure 2 , Figure 2 : is a schematic diagram of the preset feature extraction model training process provided by an embodiment of the present invention. Specifically, the embodiment of the present invention first initializes the initial ecosystem. Among them, the initial ecosystem in the embodiment of the present invention includes a plurality of first populations, and the first population includes a plurality of first individuals. Accordingly, in the embodiment of the present invention, each population represents a parameter set in the neural network, and the fitness of each individual is determined by its performance on a specific task. Exemplarily, the embodiment of the present invention constructs an ecosystem containing N individuals, each of which represents a parameter set P i ={w i ,b i}, where w i and b i Represent the weight and bias of the neural network respectively. Correspondingly, the initial fitness of the individual F i It is determined by its performance on a specific task and evaluated by a nonlinear function f(), which takes into account a variety of ecological and genetic factors, as shown in the following equation (15):

[0121] F i=f(P i ,E,G) (15)

[0122] Among them, E represents environmental variables, G represents genetic factors, and P i is the parameter set of the i-th individual, including the weight w i and bias b i ; F i is the initial fitness of the individual. For example, in some embodiments of the present invention, the environmental complexity coefficient ∈ is set to 0.3 to simulate the impact of the initial environment on the individual fitness.

[0123] It should be noted that, in some embodiments of the present invention, the nonlinear function f() is calculated using factors such as environmental variables (E), genetic factors (G), and individual parameter sets (including weights w and biases b), as shown in the following formula (16):

[0124] f(P i ,E,G)=Sig(κ·L(w i ,b i )+λ·E+μ·G(w i ,b i )+ν·D(w i ,b i ,E)) (16)

[0125] Among them, w i and b i Represent the weight and bias of the i-th individual (neural network parameter set), E is the environmental variable, G(w i ,b i ) is the genetic factor function, L(w i ,b i ) is the inverse of the loss function, D(w i ,b i ,E) is a nonlinear function that describes the dynamic interaction between individual parameters and environmental variables. In addition, Sig() is the Sigmoid activation function, and κ, λ, μ, ν are weight coefficients used to balance the contribution of various factors to fitness.

[0126] Accordingly, the genetic factor function G(w i ,b i ) Considering the diversity and complexity of parameters, the calculation formula is shown in the following formula (17):

[0127]

[0128] Where ρ and θ are weight coefficients used to adjust the contribution of each item; Var(w i ,b i) represents the variance of weights and biases, which is used to measure the diversity of parameters. Accordingly, in the embodiment of the present invention, the environmental interaction function D(w i ,b i ,E) simulate the interaction between individual parameters and the environment, as shown in equation (18):

[0129]

[0130] Among them, η is the weight coefficient, E k Represents the kth element of the environment variable, x kj Represents the environment variable E k The associated input features, tanh() is the hyperbolic tangent function. In addition, the performance evaluation function L(w i ,b i ) Based on the performance of the neural network on a specific task, considering that generally the better the performance, the smaller the loss function value, the embodiment of the present invention uses the negative value of the loss function or its reciprocal to represent the performance, as shown in the following formula (19):

[0131]

[0132] Among them, represents the loss function, and Y is the true label. In addition, It is the model prediction, obtained by the preset Softmax function.

[0133] Furthermore, the embodiment of the present invention calculates the fitness of each first individual in the initial ecosystem to obtain corresponding first fitness data. Specifically, the embodiment of the present invention calculates the fitness of each individual (first individual) in the initial ecosystem to determine its survival ability in the ecosystem. Among them, individuals with higher fitness will have a greater chance of reproducing offspring. Exemplarily, the fitness calculation in the embodiment of the present invention is based on the current state of the individual and its interaction with other individuals in the ecosystem. Among them, S in the embodiment of the present invention ij is the interaction strength between individuals i and j, and the fitness function of the individual is shown in the following formula (20):

[0134] F′ i =F i +α(∑ j≠i S ij F j +∑ j≠i (C′ ij -S′ ij )F j ) (20)

[0135] Among them, F′ iis the fitness after considering ecological interactions; α is the weight adjustment coefficient, which represents the degree of influence of interactions between individuals. In some embodiments of the present invention, α is set to 0.5. In addition, C′ ij and S′ ij is the ecological interaction parameter after adaptive adjustment. ij and S′ ij The adaptive adjustment is based on the ecological network status evaluation Φ, that is, the competition intensity C ij and symbiotic strength S ij Adaptive adjustment is performed, and the adjustment method is shown in the following equations (21) and (22):

[0136] C′ ij =C ij ·θ(Φ,ΔΦ) (21)

[0137] S′ ij =S ij ·ψ(Φ,ΔΦ) (22)

[0138] Wherein, θ and ψ are adaptive functions for adjusting competition and symbiosis intensity, respectively, Φ is the performance feedback function, and ΔΦ represents the change in Φ, that is, the difference between the current evaluation and the previous evaluation. Accordingly, in an embodiment of the present invention, the performance feedback function Φ evaluates the overall state of the current ecological network based on the performance of the neural network on the validation set. In some embodiments of the present invention, the calculation formula of the performance feedback function Φ is shown in the following formula (23):

[0139] Φ=w1·F accuracy (P)-w2·F complexity (C)+w3·F diversity (D)-w4·F time (T) (23)

[0140] Where w1, w2, w3, w4, w5 are weight coefficients used to balance the contribution of different indicators. accuracy (P), F complexity (C) F diversity (D), F time (T) is the component function of the formula. Among them, the accuracy function F accuracy (P) = log(1 + P), complexity function Diversity function F diversity (D) = tanh(D), time function F time(T) = log(1 + T). Accordingly, where P is the accuracy, L is the average loss value; C is the network complexity, which is proportional to the number of parameters in the embodiment of the present invention; D is the data diversity index, which is proportional to the variance of the data features; and T is the training time, which is proportional to the number of iterations.

[0141] In addition, the ecological interaction intensity S in the embodiment of the present invention ij It refers to the degree of mutual influence between two individuals i and j in the ecosystem, including the competition intensity C ij and symbiotic strength S ij , as shown in the following formula (24):

[0142] S ij =λC ij +(1-λ)S ij (twenty four)

[0143] Wherein, λ is an intensity adjustment balance factor, which is used to adjust the relative importance of competition and symbiosis. For example, in some embodiments of the present invention, λ is set to 0.6.

[0144] In addition, in the embodiment of the present invention, the competition intensity C ij It is calculated based on resource requirements and overlap, as shown in the following formula (25):

[0145]

[0146] Among them, C ij is the competition intensity between individuals i and j, and Represents the demand of individuals i and j for the rth resource, R is the total number of resource categories, and the exponential function The role of is to smooth the impact of demand differences between individuals, ensuring that even small demand differences are reflected in the intensity of competition.

[0147] It should be noted that the symbiotic strength S in the embodiment of the present invention is ij It reflects the degree of mutual cooperation between two individuals and is measured by the efficiency of their shared resources, as shown in the following formula (26):

[0148]

[0149] Among them, S ij is the symbiosis strength between individuals i and j.

[0150] Furthermore, the embodiment of the present invention determines the probability of the first individual being selected based on the first fitness data. Specifically, the embodiment of the present invention simulates the process of natural selection to select individuals with higher fitness for reproduction and eliminate individuals with lower fitness. Among them, the probability P of individual i being selected in the embodiment of the present invention is i The calculation formula is shown in formula (27):

[0151]

[0152] Among them, P i is the probability that individual i is selected.

[0153] Furthermore, the embodiment of the present invention performs genetic operations according to corresponding probabilities to obtain a plurality of second populations. Specifically, the genetic operations in the embodiment of the present invention include crossover operations and mutation operations. After calculating the probability of each individual being selected, the embodiment of the present invention performs regeneration and reproduction to simulate the genetic mutation process of organisms. Individuals with higher fitness produce offspring through crossover and mutation operations in the genetic algorithm, thereby generating new individuals and populations. For example, in the embodiment of the present invention, P i ′ and P j ′ is the parent individual, and its offspring P k It is generated by the crossover formula, as shown in the following formula (28):

[0154] P k =γP i ′+(1-γ)P j ′ (28)

[0155] Among them, γ is the crossover coefficient, which determines the fusion ratio of the parental genetic information; P i ′ and P j ' is the parameter set of the parent individual. For example, in some embodiments of the present invention, γ is set to 0.7.

[0156] In addition, the mutation operation in the embodiment of the present invention is implemented by adding a random disturbance ΔP, the size of which is controlled by the mutation rate μ, as shown in the following formula (29):

[0157] P′ k =P k +μΔP (29)

[0158] Among them, P k is the parameter set of the offspring individuals; μ is the mutation rate, which controls the amplitude of the mutation operation. For example, in some embodiments of the present invention, the mutation rate μ is set to 0.1. In addition, the role of the genetic variation disturbance ΔP in the embodiments of the present invention is to increase the diversity of the population. The calculation formula of the disturbance term is shown in the following formula (30):

[0159] ΔP=ζ·(U-0.5) (30)

[0160] Wherein, ΔP is the random disturbance in the mutation operation, ζ is the disturbance amplitude control coefficient, and U is a random vector with the same dimension as P, whose elements are drawn from the uniform distribution U(0,1). In some embodiments of the present invention, ζ is set to 0.05 to control the amplitude of the mutation.

[0161] Furthermore, the embodiment of the present invention updates the fitness of each second individual in the second population according to the ecological dynamic information to obtain second fitness data. Specifically, the ecological dynamic information in the embodiment of the present invention includes environmental change information and ecological interaction information. The embodiment of the present invention simulates ecological interactions between individuals, such as competition and symbiotic relationships, which affect the fitness and survival ability of individuals. Exemplarily, the embodiment of the present invention sets an ecological interaction matrix I, where element I ij represents the type of relationship between individuals i and j, which will affect the fitness update of the individuals, as shown in the following formula (31):

[0162] F″ i =F′ i +∑ j≠i I ij F′ j (31)

[0163] Among them, F″ i is the fitness update function after considering ecological interactions.

[0164] At the same time, the embodiment of the present invention simulates the impact of environmental changes on the ecosystem, such as changes in resources, new adaptive challenges, etc., so that individuals adapt to the new environment and further optimize the parameters within the ecosystem. Exemplarily, in the embodiment of the present invention, environmental changes are simulated by modifying the environmental complexity coefficient ∈, thereby affecting the fitness calculation of individuals. The fitness adjustment caused by environmental pressure is shown in the following formula (32):

[0165] F″′ i =F″ i ·(1+vΔE) (32)

[0166] Among them, F″′ i is the final adjusted fitness, ∈ is the environmental complexity coefficient preset by humans; ΔE is the environmental change factor, which indicates the degree of environmental change from the previous generation to the current generation. The calculation formula is shown in the following formula (33):

[0167]

[0168] Among them, F′ i is the fitness of individual i in the current environment, Fi is the fitness in the previous generation environment, and ΔE is the environmental change factor.

[0169] Finally, the embodiment of the present invention updates the parameters of the preset neural network according to the second fitness data and the ecological dynamic information to obtain a preset feature extraction model. Specifically, the embodiment of the present invention updates the parameters of the neural network according to the fitness of individuals in the ecosystem and the ecological dynamic information to reflect the purification of biological individuals in the ecosystem. Exemplarily, the embodiment of the present invention updates the fitness F″′ after ecological interaction and environmental pressure adjustment i , update the individual parameters as shown in the following formula (34):

[0170] P i new =P i +η(F″′ i -F i )G i (34)

[0171] Among them, P i new is the updated individual parameter set, η is the learning rate, which indicates the step size of parameter update; G i It is the gradient pointing to the direction of maximum fitness increase. For example, in some embodiments of the present invention, η is set to 0.01.

[0172] In some embodiments of the present invention, before executing the step of inputting the feature data into a preset classifier for classification to obtain a quality assessment result, the water environment quality remote sensing analysis method provided by the embodiment of the present invention further includes but is not limited to the following steps:

[0173] Build an ELM model.

[0174] The extreme learning machine model is trained by the fractional gradient descent algorithm to optimize the weights from the hidden layer to the output layer in the extreme learning machine model to obtain a preset classifier.

[0175] In this specific embodiment, the embodiment of the present invention first constructs an extreme learning machine model, and trains the extreme learning machine model through a fractional gradient descent algorithm, thereby optimizing the weights from its hidden layer to the output layer, and then constructing a preset classifier. Specifically, the extreme learning machine (ELM) in the embodiment of the present invention is a machine learning algorithm with a fast learning speed, high generalization accuracy, and will not fall into a local minimum. The embodiment of the present invention adopts an extreme learning machine algorithm based on fractional gradient descent, introduces a fractional gradient descent method in the training process of the extreme learning machine, and adjusts the fractional parameters to more flexibly handle unbalanced data and outliers. In addition, by analyzing the intrinsic characteristics of the data, the features used for classification are adaptively selected and adjusted, thereby improving the accuracy and robustness of the classifier. The embodiment of the present invention first constructs an extreme learning machine structure and initializes the weights and biases from the input layer to the hidden layer. Exemplarily, the weight from the input layer to the hidden layer in the embodiment of the present invention is W px and the bias is b px , the number of neurons in the hidden layer is N px For example, in some embodiments of the present invention, N px is set to twice the number of data features. Accordingly, the dataset after dimensionality reduction is where x i is the eigenvector, y i is the corresponding label.

[0176] Next, the embodiment of the present invention trains the ELM model by using the fractional gradient descent method to optimize the weights from the hidden layer to the output layer. For example, during the training process, the weights β from the hidden layer to the output layer are px The optimization is performed by fractional gradient descent. The fractional order τ is a hyperparameter, which is usually set around 0.5 (median). Accordingly, the optimization process in the embodiment of the present invention is shown in the following formula (35):

[0177]

[0178] Among them, Q τ is the fractional order loss function and η is the learning rate.

[0179] It should be noted that, in the embodiment of the present invention, the weight β px The update depends on the gradient of the fractional loss function. In the calculation of , we first define the fractional loss function Q τ , as shown in the following formula (36):

[0180]

[0181] Next, the embodiment of the present invention calculates Qτ About Beta px The gradient of is shown in equation (37):

[0182]

[0183] Accordingly, the embodiment of the present invention adjusts the parameters of the ELM model according to the feedback of the training data, thereby optimizing the classification effect. When the model reaches the preset classifier verification accuracy, the final classifier model is determined to obtain the preset classifier. For example, the loss function L used in the model optimization of the embodiment of the present invention is px is the cross entropy loss, and its calculation formula is shown in the following formula (38):

[0184]

[0185] Among them, is the model's predicted value for the i-th sample, y i is the actual label.

[0186] In addition, in the process of training the ELM model, the embodiment of the present invention evaluates the performance of the preset classifier obtained by the validation data set to ensure its accuracy and generalization ability. Specifically, the embodiment of the present invention evaluates the performance of the preset classifier obtained by the accuracy rate and the loss function, and the loss function is L px , and its calculation formula is shown in the following formula (39):

[0187]

[0188] Among them, is the cross entropy loss function, and f is the output function of the ELM.

[0189] In some embodiments of the present invention, before executing the step of preprocessing the preset remote sensing data set to obtain the first remote sensing data, the water environment quality remote sensing analysis method provided by the embodiment of the present invention further includes but is not limited to the following steps:

[0190] Spectral data is collected through several remote sensing platforms, and the spectral data is annotated to obtain a spectral feature data set.

[0191] The spectral feature dataset is expanded through a preset generative adversarial network model to obtain a preset remote sensing dataset.

[0192] In this specific embodiment, the embodiment of the present invention first collects spectral data of the water environment through several remote sensing platforms, and at the same time, annotates the collected spectral data, thereby constructing a spectral feature data set. Specifically, the spectral data collected by several remote sensing platforms in the embodiment of the present invention include but are not limited to satellite images, images taken by drones, and spectral data collected on the ground. Among them, the collected images and data have high spatial resolution and rich spectral information. Correspondingly, the embodiment of the present invention annotates the collected spectral data by manual annotation. Exemplarily, in an embodiment of the present invention, fixed-point sampling is performed at a specific location by a handheld or fixed spectrometer as the annotation of the point data. Among them, the data annotation categories include classifications of different water quality levels. For example, in some embodiments of the present invention, the annotation categories include: Class I to II water (excellent), Class III water (good), Class IV water (mildly polluted), Class V water (moderately polluted) and Class V water (severely polluted). Among them, taking spectral vector data as an example, in an embodiment of the present invention, the spectral feature vector of each pixel point in an image is shown in the following formula (40):

[0193] S(ax,ay)=[s a1 ,s a2 ,…,s an ] (40)

[0194] Where S represents the spectral feature vector, s ai represents the i-th spectral feature value, n represents the total number of spectral features, ax and ay represent the horizontal and vertical coordinates of the pixel point in the image, respectively.

[0195] Next, the embodiment of the present invention uses a preset generative adversarial network model to perform data expansion on the spectral feature data set, thereby constructing a preset remote sensing data set. It is easy to understand that the collection, acquisition, annotation and preprocessing of training data in the embodiment of the present invention are time-consuming and labor-intensive, and insufficient training samples easily lead to poor generalization ability of the model, while affecting the accuracy of the model. Therefore, the embodiment of the present invention uses a preset generative adversarial network model to perform data expansion on the corresponding spectral feature data to increase the number of training samples, so as to improve the generalization ability and accuracy of the model.

[0196] In some embodiments of the present invention, before performing data expansion on the spectral feature data set by using a preset generative adversarial network model to obtain a preset remote sensing data set, the water environment quality remote sensing analysis method provided by the embodiment of the present invention further includes but is not limited to the following steps:

[0197] Construct a first generative adversarial network model, wherein the first generative adversarial network model includes a generator network and a discriminator network.

[0198] According to the loss function, the generator network and the discriminator network are optimized through adversarial training, and the network parameters of the first generative adversarial network model are updated through the gradient descent method to obtain a preset generative adversarial network model. Among them, the input layer input of the generator network includes control noise, and the loss function of the generator network includes a manifold constraint term.

[0199] In this specific embodiment, the embodiment of the present invention is based on the water environment quality remote sensing analysis task of multi-source remote sensing data, and the generative adversarial network framework of manifold learning can effectively enhance the generalization ability of the model and improve the accuracy of water environment quality remote sensing analysis. Among them, the generative adversarial network framework in the embodiment of the present invention not only learns the distribution of data, but also considers the structure of data on low-dimensional manifolds, so that the generated data is more diverse and has practical significance, and can better simulate the complexity and diversity of multi-source remote sensing data. In addition, the embodiment of the present invention introduces an anti-noise injection mechanism in the generative network, and injects control noise into the network input layer to enhance the robustness of the model to the disturbance of the input data, thereby improving the quality of data expansion and the stability of the model. Accordingly, the embodiment of the present invention first constructs a first generative adversarial network model. Specifically, the first generative adversarial network model in the embodiment of the present invention includes a generator network and a discriminator network, and the network parameters of the generator and the discriminator are first initialized. Among them, the generator is used to capture the distribution of multi-source remote sensing data, and the discriminator is used to distinguish between real data and generated data generated by the generator. Exemplarily, the parameters of the generator G and the discriminator D in the embodiment of the present invention are initialized as θ g and θ d . The generator G attempts to pass the mapping function G(bz; θ g ) Generate data from latent space vector bz The discriminator D evaluates the probability D(bx; θ) that the corresponding sample comes from the real data bx d ).

[0200] Next, the embodiment of the present invention performs adversarial training optimization on the generator network and the discriminator network according to the loss function, and optimizes the network parameters of the first generative adversarial network model by the gradient descent method, thereby obtaining a preset generative adversarial network model. Specifically, the input of the generator network in the embodiment of the present invention includes controlled noise. The embodiment of the present invention increases the diversity and complexity of generated data by injecting controlled noise into the input layer of the generator. It is easy to understand that the embodiment of the present invention can not only increase the variability of the data by injecting controlled noise, but also simulate the random noise existing in the actual remote sensing data, so that the generated data is closer to the real scene. Exemplarily, for each generated sample, the latent space vector bz is obtained by injecting noise bn, and I is the unit matrix. Therefore, the latent space vector after noise injection is as follows (41):

[0201] bz′=bz+αbn (41)

[0202] Wherein, α is a noise intensity control parameter. For example, in the embodiment of the present invention, the parameter α is set to 0.1. In addition, in the embodiment of the present invention, the noise intensity control parameter α is dynamically adjusted according to the performance of the model on the validation set. When the performance evaluation index P is the similarity between the generated data and the real data, the update of α is shown in the following formula (42):

[0203]

[0204] Among them, η α is the learning rate of α, It is the partial derivative of the performance evaluation index P with respect to α, which can reflect the impact of small changes in α on the model performance.

[0205] Accordingly, during the adversarial training process, the generator attempts to generate more and more realistic remote sensing data, while the discriminator distinguishes between real data and generated data. Therefore, through this adversarial process, the generator can learn how to generate data similar to real remote sensing data. Exemplarily, the generator G and the discriminator D are optimized through adversarial training, and the loss function of the generator G is shown in the following formula (43):

[0206]

[0207] In addition, the loss function of the discriminator D is shown in the following formula (44):

[0208]

[0209] At the same time, in the embodiment of the present invention, a manifold constraint term is introduced into the loss function of the generative network. By introducing the manifold learning constraint, the generated data is not only realistic at the pixel level, but also matches the real data in structure and distribution. The embodiment of the present invention adds an additional manifold constraint term to the loss function to force the generated data to maintain the same manifold structure as the real data in the high-dimensional space. For example, in order to ensure that the generated data maintains the manifold structure of the real data, the embodiment of the present invention adds the manifold constraint L manifold Added to the loss function of the generator, where f() is the mapping function from the data space to the low-dimensional manifold space, and the manifold constraint is as follows (45):

[0210]

[0211] Correspondingly, the generator loss function after adding the manifold constraint term is shown in the following formula (46):

[0212] L′ G=L G +λL manifold (46)

[0213] Wherein, λ is a weight parameter for balancing the generator adversarial loss and the manifold constraint. For example, in an embodiment of the present invention, the manifold constraint weight λ is set to 10.

[0214] In addition, in some embodiments of the present invention, the mapping function f() uses an autoencoder structure, where f() corresponds to the encoding part of the autoencoder, as shown in the following formula (47):

[0215] f(bx)=σ(W f ·bx+b f ) (47)

[0216] Among them, W f and b f are the weight matrix and bias vector of the encoding part of the autoencoder, and σ() is the ReLU nonlinear activation function.

[0217] Furthermore, the embodiment of the present invention alternately optimizes the generator and the discriminator, and continuously updates the network parameters by the gradient descent method. In each iteration, the embodiment of the present invention first fixes the relevant parameters of the discriminator to update the relevant parameters of the generator, and then fixes the relevant parameters of the generator to update the relevant parameters of the discriminator, so as to achieve the optimal balance point. Exemplarily, the embodiment of the present invention updates θ by the gradient descent method. g and θ d , where the formula for updating the parameters of the generator G is shown in equation (48):

[0218]

[0219] Accordingly, the formula for updating the parameters of the discriminator D is shown in equation (49):

[0220]

[0221] Among them, η g and η d are the learning rates of the generator and the discriminator, L C is the auxiliary classifier loss function. In one embodiment, it is usually set to n g =0.0002 and η d =0.0001.

[0222] In addition, the auxiliary classifier loss function in the embodiment of the present invention is further refined by introducing the auxiliary classifier C to enhance the classification ability of the discriminator for the generated data, as shown in the following formula (50):

[0223]

[0224] Among them, C(G(bz′ i )) represents the auxiliary classifier C for generating data G(bz′ i )’s classification output, y i is the true label of the generated data, and in the embodiment of the present invention, the label of the generated data is 0. Based on this, the discriminator not only needs to judge the authenticity of the data, but also needs to correctly classify the generated data, thereby improving the discriminator's discrimination ability.

[0225] It is easy to understand that the embodiment of the present invention repeats the above steps until the number of iterations reaches a preset maximum number of iterations. For example, in the embodiment of the present invention, the maximum number of iterations of the model is 500. When the model training is completed, the output of the model is an enhanced remote sensing data set generated by the data expansion algorithm.

[0226] In some embodiments of the present invention, the feature data is input into a preset classifier for classification to obtain a quality assessment result, including but not limited to the following steps:

[0227] The characteristic data is analyzed by a preset classifier to obtain classification information, wherein the classification information includes water environment quality classification level information.

[0228] The quality of the corresponding water environment is assessed according to the classification information to obtain the quality assessment result.

[0229] In this specific embodiment, the embodiment of the present invention first analyzes the feature data through a preset classifier to obtain corresponding classification data, and then performs a quality assessment on the corresponding water environment according to the classification information to obtain a quality assessment result. Specifically, the classification information in the embodiment of the present invention includes water environment quality classification level information. For example, the classification information output by the preset classifier in the embodiment of the present invention includes Class I to Class II water (excellent), Class III water (good), Class IV water (mildly polluted), Class V water (moderately polluted) and Class V water (severely polluted). Accordingly, the embodiment of the present invention inputs the feature data extracted by the feature extraction model into a preset separator for classification to determine the corresponding water environment quality classification level. Then, the embodiment of the present invention performs a quality assessment on the water environment through the corresponding classification results, that is, the classification information, thereby realizing the analysis of the corresponding water environment and obtaining a quality assessment result.

[0230] Next, the solution of the embodiment of the present invention is described in detail and explained in conjunction with a specific water environment quality remote sensing analysis scenario:

[0231] The embodiment of the present invention first collects spectral data through several remote sensing platforms, and annotates the spectral data, thereby obtaining a spectral feature data set. Then, the embodiment of the present invention trains a generative adversarial network model based on manifold learning to construct a preset generative adversarial network model, and then expands the spectral feature data set through the preset generative adversarial network model, thereby constructing a preset remote sensing data set. Then, the embodiment of the present invention trains a multi-dimensional adaptive autoencoder through a preset remote sensing data set to obtain a preset dimensionality reduction model, so that the acquired data can be reduced in dimension through the preset dimensionality reduction model, thereby better extracting the complex features and internal connections in the multi-source remote sensing data, and providing a more accurate and robust feature representation for remote sensing analysis of water environment quality. Then, the embodiment of the present invention uses a neural network as a feature extraction model to extract features from the data. Accordingly, the embodiment of the present invention trains and optimizes the constructed preset neural network based on the neural network parameter optimization method of ecosystem optimization, thereby obtaining a preset feature extraction model. Further, the embodiment of the present invention inputs the feature-extracted data into the constructed extreme learning machine model for model training, and trains the extreme learning machine by introducing the fractional gradient descent method, thereby constructing a preset classifier. Finally, the embodiment of the present invention processes water environment samples through the trained model to complete the remote sensing analysis task of water environment quality. test , which contains multi-source remote sensing data of unknown waters. The embodiment of the present invention first inputs these data into a preset dimensionality reduction model for dimensionality reduction processing to obtain dimensionality reduction data. Further, the embodiment of the present invention inputs the dimensionality reduced data into a preset feature extraction model for feature extraction to obtain corresponding feature data. Further, the feature extracted data is input into a preset classifier for classification to obtain a quality assessment result, thereby realizing water environment analysis.

[0232] It is easy to understand that the embodiments of the present invention can accurately classify and evaluate the quality level of the water environment by comprehensively utilizing multi-source remote sensing data and advanced data processing technology. At the same time, through the optimization methods of data expansion and feature dimension reduction, the model can effectively process new and unseen remote sensing data, thereby improving the generalization performance. In addition, the embodiments of the present invention effectively improve the robustness of the model to input data disturbances and the stability in complex environments by introducing anti-noise mechanisms and ecosystem optimization algorithms. Accordingly, the embodiments of the present invention adopt adaptive and dynamic adjustment strategies, so that the model can more efficiently process a large amount of multi-source remote sensing data, thereby shortening the data processing time.

[0233] See also Figure 3 The present application also provides a water environment quality remote sensing analysis system, which can implement the above-mentioned water environment quality remote sensing analysis method, and the system includes:

[0234] The first module 210 is used to obtain water environment monitoring data, wherein the water environment monitoring data includes multi-source remote sensing data.

[0235] The second module 220 is used to input the multi-source remote sensing data into a preset dimension reduction model for dimension reduction processing to obtain dimension reduction data. The preset dimension reduction model is obtained by training a multi-dimensional adaptive encoder model.

[0236] The third module 230 is used to input the dimension reduction data into a preset feature extraction model for feature extraction to obtain feature data. The preset feature extraction model is obtained by training a preset neural network through an ecosystem optimization algorithm.

[0237] The fourth module 240 is used to input the feature data into a preset classifier for classification to obtain a quality assessment result.

[0238] 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.

[0239] 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-mentioned water environment quality remote sensing analysis method when executing the computer program. The electronic device can be any intelligent terminal including a tablet computer, a car computer, etc.

[0240] 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.

[0241] See also Figure 4 , Figure 4 The hardware structure of an electronic device of another embodiment is illustrated, and the electronic device includes:

[0242] 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;

[0243] 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 water environment quality remote sensing analysis method of the embodiment of this application;

[0244] Input / output interface 330, used to implement information input and output;

[0245] Communication interface 340, used to realize communication interaction between the device and other devices, which can be realized by wired mode (such as USB, network cable, etc.) or wireless mode (such as mobile network, WIFI, Bluetooth, etc.);

[0246] 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 );

[0247] 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 .

[0248] 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 above-mentioned water environment quality remote sensing analysis method is implemented.

[0249] 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.

[0250] 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.

[0251] 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.

[0252] 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.

[0253] 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.

[0254] 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 of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0255] 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.

[0256] 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.

[0257] 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.

[0258] 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.

[0259] 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, which is stored in a storage medium and includes multiple instructions for a computer device (which can be a personal computer, a server, or a 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 (ROM), random access memory (RAM), disk or optical disk, and other media that can store programs.

[0260] 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 water environment quality remote sensing analysis method, characterized in that: The method comprises: Acquiring water environment monitoring data; wherein the water environment monitoring data includes multi-source remote sensing data; Inputting the multi-source remote sensing data into a preset dimensionality reduction model for dimensionality reduction processing to obtain dimensionality reduction data; wherein the preset dimensionality reduction model is obtained by training a multi-dimensional adaptive encoder model; Inputting the dimension reduction data into a preset feature extraction model to extract features and obtain feature data; wherein the preset feature extraction model is obtained by training a preset neural network through an ecosystem optimization algorithm; Inputting the feature data into a preset classifier for classification to obtain a quality assessment result; Before executing the step of inputting the multi-source remote sensing data into a preset dimensionality reduction model for dimensionality reduction processing to obtain dimensionality reduction data, the method further includes: Constructing the multi-dimensional adaptive encoder model; wherein the multi-dimensional adaptive encoder model includes a preset encoder and a preset decoder; Preprocessing a preset remote sensing data set to obtain first remote sensing data; wherein the preset remote sensing data set includes remote sensing source data from a plurality of different sources; Integrating the first remote sensing data by a weighted fusion strategy to obtain a first fusion feature vector; Dynamically adjusting the network structure and parameters of the preset encoder according to the first fused feature vector to compress and encode the first fused feature vector to obtain a low-dimensional feature; The feature weights of the low-dimensional features are evaluated through an attention mechanism to obtain weighted features; Dynamically adjusting the network structure and parameters of the preset decoder according to the low-dimensional features to decode the weighted features and reconstruct a second fused feature vector; Optimizing network parameters of the multi-dimensional adaptive encoder model by a preset optimization algorithm according to the first fused feature vector and the second fused feature vector to obtain the preset dimensionality reduction model; wherein the preset optimization algorithm includes a back propagation algorithm and a gradient descent algorithm; Before executing the step of preprocessing the preset remote sensing data set to obtain the first remote sensing data, the method further includes: Collecting spectral data through several remote sensing platforms, and annotating the spectral data to obtain a spectral feature data set; The spectral feature data set is expanded by using a preset generative adversarial network model to obtain the preset remote sensing data set.

2. The method according to claim 1, characterized in that Before executing the step of inputting the dimension reduction data into a preset feature extraction model to extract features to obtain feature data, the method further includes: Constructing the preset neural network; Initializing an initial ecosystem; wherein the initial ecosystem includes a plurality of first populations, and the first populations include a plurality of first individuals; Calculating the fitness of each of the first individuals in the initial ecosystem to obtain first fitness data; Determining the probability of the first individual being selected according to the first fitness data; Performing genetic operations according to the probabilities to obtain a plurality of second populations; wherein the genetic operations include crossover operations and mutation operations; updating the fitness of each second individual in the second population according to the ecological dynamic information to obtain second fitness data; wherein the ecological dynamic information includes environmental change information and ecological interaction information; The parameters of the preset neural network are updated according to the second fitness data and the ecological dynamic information to obtain the preset feature extraction model.

3. The method according to claim 1, characterized in that: Before executing the step of inputting the feature data into a preset classifier for classification to obtain a quality assessment result, the method further includes: Construct an ELM model; The extreme learning machine model is trained by a fractional gradient descent algorithm to optimize the weights from the hidden layer to the output layer in the extreme learning machine model to obtain the preset classifier.

4. The method according to claim 1, characterized in that: Before executing the step of performing data expansion on the spectral feature data set by using a preset generative adversarial network model to obtain the preset remote sensing data set, the method further includes: Constructing a first generative adversarial network model; wherein the first generative adversarial network model includes a generator network and a discriminator network; The generator network and the discriminator network are optimized through adversarial training according to the loss function, and the network parameters of the first generative adversarial network model are updated by the gradient descent method to obtain the preset generative adversarial network model; wherein the input layer input of the generator network includes control noise, and the loss function of the generator network includes a manifold constraint term.

5. The method according to claim 1, characterized in that The step of inputting the feature data into a preset classifier for classification to obtain a quality assessment result includes: Analyzing the characteristic data by the preset classifier to obtain classification information; wherein the classification information includes water environment quality classification level information; A quality assessment is performed on the corresponding water environment according to the classification information to obtain the quality assessment result.

6. A water environment quality remote sensing analysis system, characterized in that: The system comprises: The first module is used to obtain water environment monitoring data; wherein the water environment monitoring data includes multi-source remote sensing data; The second module is used to input the multi-source remote sensing data into a preset dimensionality reduction model for dimensionality reduction processing to obtain dimensionality reduction data; wherein the preset dimensionality reduction model is obtained by training a multi-dimensional adaptive encoder model; The third module is used to input the dimension reduction data into a preset feature extraction model to extract features and obtain feature data; wherein the preset feature extraction model is obtained by training a preset neural network through an ecosystem optimization algorithm; The fourth module is used to input the feature data into a preset classifier for classification to obtain a quality assessment result; Before executing the step of inputting the multi-source remote sensing data into a preset dimensionality reduction model for dimensionality reduction processing to obtain dimensionality reduction data, the method further includes: Constructing the multi-dimensional adaptive encoder model; wherein the multi-dimensional adaptive encoder model includes a preset encoder and a preset decoder; Preprocessing a preset remote sensing data set to obtain first remote sensing data; wherein the preset remote sensing data set includes remote sensing source data from a plurality of different sources; Integrating the first remote sensing data by a weighted fusion strategy to obtain a first fusion feature vector; Dynamically adjusting the network structure and parameters of the preset encoder according to the first fused feature vector to compress and encode the first fused feature vector to obtain a low-dimensional feature; The feature weights of the low-dimensional features are evaluated through an attention mechanism to obtain weighted features; Dynamically adjusting the network structure and parameters of the preset decoder according to the low-dimensional features to decode the weighted features and reconstruct a second fused feature vector; Optimizing network parameters of the multi-dimensional adaptive encoder model by a preset optimization algorithm according to the first fused feature vector and the second fused feature vector to obtain the preset dimensionality reduction model; wherein the preset optimization algorithm includes a back propagation algorithm and a gradient descent algorithm; Before executing the step of preprocessing the preset remote sensing data set to obtain the first remote sensing data, the method further includes: Collecting spectral data through several remote sensing platforms, and annotating the spectral data to obtain a spectral feature data set; The spectral feature data set is expanded by using a preset generative adversarial network model to obtain the preset remote sensing data set.

7. 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 5.

8. 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 5 is implemented.

Images