Algae content information monitoring method based on multi-source remote sensing data

Abstract

The present invention relates to the technical field of algae content monitoring. Disclosed is an algae content information monitoring method based on multi-source remote sensing data. The present invention comprises: acquiring multi-source remote sensing data comprising satellite remote sensing data, unmanned aerial vehicle remote sensing data and ground monitoring data, and preprocessing the data, involving radiation correction, geometric correction and noise elimination; and fusing the multi-source data into a trained algae monitoring model for prediction, and using Kalman filtering to assimilate observation data and model prediction data. The present invention fuses observation data and model prediction data by means of Kalman filtering technology, so as to dynamically adjust the model state, such that the model can more accurately reflect the actual observation situation. The Kalman filtering optimizes the real-time prediction capability of models by balancing the uncertainties between observation data and prediction data, thereby ensuring the accuracy and real-time performance of monitoring results, and also ensuring the reliability and adaptability of models under various environmental conditions.

Description

Algae content information monitoring method based on multi-source remote sensing data TECHNICAL FIELD

[0001] The present application belongs to the field of algae content monitoring, and particularly relates to an algae content information monitoring method based on multi-source remote sensing data. BACKGROUND

[0002] Algae are an important component of aquatic ecosystems, serving as both primary producers and playing a crucial role in water quality indicators. Despite their key role in maintaining ecological balance, abnormal proliferation of algae can lead to water bloom phenomena, which in turn triggers a series of environmental and public health problems such as water eutrophication, fish mortality, and water quality deterioration. Therefore, accurate monitoring of algae content and distribution in water bodies and timely understanding of water quality are of great significance for water resource management and environmental protection. Conventional techniques for algae content information monitoring mainly utilize multi-spectral or hyperspectral sensors mounted on satellites to obtain spectral data of large-scale water bodies. By analyzing these data, the distribution and content of algae in water bodies can be inferred. Alternatively, multi-spectral or hyperspectral sensors can be mounted on unmanned aerial vehicles (UAVs) to obtain high-resolution spectral data of local areas. While satellite remote sensing and UAV remote sensing have their respective advantages in algae monitoring, current data monitoring processes rely solely on image and spectral information analysis, which has some obvious shortcomings. First, while image and spectral information can provide intuitive information on the surface distribution of algae in water bodies, it is difficult to gain in-depth understanding of the dynamics and distribution of algae within the water body. This surface monitoring limitation can lead to misjudgment of the overall health of the water body. Furthermore, relying solely on image and spectral data for monitoring makes it difficult to effectively cope with complex water environments, especially when faced with different environmental conditions and variable climate factors, which limits the reliability and adaptability of the monitoring system.

[0003] Therefore, the present application is proposed.

[0004] SUMMARY

[0005] The technical problem to be solved by the present application is to overcome the shortcomings of the prior art and provide an algae content information monitoring method based on multi-source remote sensing data, which solves the problems raised in the background.

[0006] To solve the above technical problems, the basic idea of the technical solution of the present application is as follows:

[0007] An algae content information monitoring method based on multi-source remote sensing data, comprising the following steps:

[0008] Multi-source remote sensing data including satellite remote sensing data, UAV remote sensing data, and ground monitoring data are obtained, and the data are pre-processed by radiation correction, geometric correction, and noise removal.

[0009] The pre-processed real remote sensing data is collected, a training data set is established, and a generative adversarial network model is used to train the training data to generate virtual data with the same statistical characteristics as the real data. The generated virtual data is combined with the real data to expand the training data set, which is used for training of the algae monitoring model;

[0010] The expanded training data set is used to train a random forest algorithm to build an algae monitoring model;

[0011] The multi-source data is fused into the trained algae monitoring model for prediction, the observation data and the model prediction data are assimilated using Kalman filtering, and the model state is adjusted to match the observation data according to the observation data and the model prediction data to obtain accurate algae content monitoring results;

[0012] An independent ground monitoring data set is used to verify the entire monitoring system to evaluate its accuracy and reliability.

[0013] Optionally, the multi-source remote sensing data includes but is not limited to Landsat, Sentinel-2, MODIS satellite data, unmanned aerial vehicle multi-spectral and hyperspectral data, and data collected by ground sensors.

[0014] Optionally, the generative adversarial network model includes a generator and a discriminator, the generator is used to generate virtual data, and the generator loss function is defined by the formula LG=-E z-pz(z) [logD(D(z))],the discriminator is used to evaluate the authenticity of the virtual data, and the discriminator loss function is defined by the formula LD=-E x～pdata(x) [logD(x)]-Ez~pz(z)[log(1-D(G(z)))],wherein z is the input noise variable of the generator, which is usually sampled from a standard normal distribution, x is a real data sample, p z (z) is the probability distribution of the noise variable, p data (x) is the probability distribution of the real data, and E is the expected value.

[0015] The training process alternately optimizes the generator and the discriminator to make the generated data have the same statistical characteristics as the real data. The trained generator is used to generate virtual remote sensing data. The generated virtual data is combined with the real data in proportion to form an expanded training data set, which is used for subsequent training of the algae monitoring model.

[0016] Optionally, the steps of training a random forest algorithm to build an algae monitoring model using the expanded training data set are

[0017] The expanded training data set is divided into multiple subsets, each of which is used to train a decision tree. In the training process of each tree, a part of features is randomly selected for splitting, the structure of the tree is constructed, and the above process is repeated to construct multiple decision trees, and finally an algae monitoring model is formed.

[0018] Optionally, before obtaining the monitoring result, the following steps are further included:

[0019] Spectral data of different algae are collected using a hyperspectral sensor, a spectral database of algae is established, spectral features of different algae are analyzed and extracted from the spectral data, and key spectral fingerprint information is identified.

[0020] An algae identification model based on spectral fingerprints is constructed using a support vector machine algorithm, the extracted spectral features are used as input to train the SVM model, and the model is used to classify and identify algae from multi-source remote sensing data to obtain algae distribution information.

[0021] Optionally, the trained algae monitoring model is used to predict the multi-source remote sensing data to obtain a preliminary expression of the algae content distribution as follows: The input data X is predicted using the model parameters Θ t to obtain the predicted algae content

[0022] Optionally, the observed data and the model prediction data are assimilated using Kalman filtering, and the model state is adjusted to be closer to the actual observed data value through the formula The posterior state estimate at time t is represented by the prior state estimate through the Kalman gain K t combined with the observation data y t ;

[0023] The calculation formula of the Kalman gain K t is K t = P t|t-1 H T (HP t|t-1 H T +R) -1 , which reflects the trade-off between observation uncertainty and prior state uncertainty.

[0024] The calculation formula of the posterior estimation covariance matrix P t|t is P t|t =(I-K t H)P t|t-1 , which is used to measure the confidence of the posterior state estimate, wherein, is the posterior state estimate at time t, is the prior state estimate at time t, and y tis the observation data at time t, P t|t-1 is the prior estimation covariance matrix at time t, H is an observation matrix, R is an observation noise covariance matrix, and I is an identity matrix.

[0025] Optionally, after obtaining the monitoring result, the virtual model needs to be calibrated according to the state estimation after the assimilation processing, to ensure that the model state accurately reflects the actual situation, and the expression is: wherein, is the calibrated model parameter, Θ is the model parameter, T is the time step, y t is the observation data at time t, f(X t ; Θ) is the prediction output of the model.

[0026] After the above technical solution is adopted, the present application has the following beneficial effects compared with the prior art. Of course, any product implementing the present application does not necessarily need to achieve all the advantages described below:

[0027] 1. The present application fuses observation data and model prediction data through Kalman filtering technology, dynamically adjusts the model state, and makes it more accurately reflect the actual observation situation. Kalman filtering balances the uncertainty of observation data and prediction data, optimizes the real-time prediction ability of the model, ensures the accuracy and real-time performance of the monitoring result, and ensures the reliability and adaptability of the model under different environmental conditions.

[0028] 2. The present application can generate virtual data with the same statistical characteristics as real data through the generation of a generative adversarial network. The introduction of virtual data also improves the generalization ability of the model, so that the model not only performs well on training data, but also maintains high-precision predictability in unseen data and actual applications, enhancing the robustness and generalization ability of the model.

[0029] The specific embodiments of the present application will be described in further detail below with reference to the accompanying drawings. BRIEF DESCRIPTION OF DRAWINGS

[0030] The accompanying drawings in the following description are only some embodiments, and other drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings. In the drawings

[0031] In the drawings:

[0032] Fig. 1 is a block diagram of an information monitoring method.

[0033] It should be noted that these drawings and written descriptions are not intended to limit the scope of the concept of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0034] The application will be described in further detail in connection with the drawings.

[0035] Referring to FIG. 1, in the embodiment, a method for monitoring algal content information based on multi-source remote sensing data is provided, including the following steps:

[0036] Multi-source remote sensing data including satellite remote sensing data, unmanned aerial vehicle remote sensing data and ground monitoring data are acquired, and the data are preprocessed by radiation correction, geometric correction and noise removal; the multi-source remote sensing data include but are not limited to Landsat, Sentinel-2, MODIS satellite data, unmanned aerial vehicle multi-spectral and hyperspectral data, and data collected by ground sensors.

[0037] Radiation correction eliminates the influence of atmosphere and sensor, geometric correction corrects spatial deformation, and noise removal improves signal-to-noise ratio, thereby providing high-quality input data for subsequent model training and data assimilation, reducing errors and biases, ensuring the accuracy and reliability of the monitoring results, and at the same time, the standardized processing procedure also enhances the usability and application value of the data.

[0038] Real remote sensing data are collected, a training data set is established, and the training data are trained using a generative adversarial network model to generate virtual data with the same statistical characteristics as the real data, the generated virtual data are combined with the real data to expand the training data set, which is used for training of the algal monitoring model; the setting of the virtual data increases the diversity and scale of the data set, enhances the robustness and generalization ability of the model, and also ensures the statistical consistency of the virtual data with the real data through the generative adversarial network, thereby improving the effectiveness and accuracy of the model training, reducing the risk of overfitting, and improving the reliability and precision of the algal monitoring results.

[0039] The random forest algorithm is trained using the expanded training data set to construct an algal monitoring model.

[0040] The multi-source data is fused into the trained algae monitoring model for prediction, the Kalman filter is used to assimilate the observation data and the model prediction data, and the model state is adjusted according to the observation data and the model prediction data to match the observation data, so as to obtain accurate algae content monitoring results; through the fusion of multi-source data, the advantages of different data sources are fully utilized, the comprehensiveness and representativeness of the data are improved, and the model prediction is more accurate and reliable. Secondly, the Kalman filter method effectively combines model prediction and real-time observation data, dynamically adjusts the model state through iterative optimization, eliminates errors and biases that may be caused by a single data source, and enhances the adaptability and robustness of the model. Through assimilation, the comprehensive utilization of observation data and model prediction data makes the final monitoring results more close to the actual situation, with higher accuracy and reliability. Finally, the process of dynamically adjusting the model state ensures that the monitoring system can respond to environmental changes in a timely manner, providing real-time and accurate algae content monitoring information, providing scientific basis for ecological environment protection and water quality management, and greatly improving the practicality and effectiveness of the monitoring system.

[0041] The entire monitoring system is verified using independent ground monitoring data sets to evaluate its accuracy and reliability.

[0042] In this embodiment, the generative adversarial network model includes two parts, a generator and a discriminator. The generator is used to generate virtual data, and the generator loss function is defined as L G =-E z～pz(z) [logD(G(z))] The discriminator is used to evaluate the authenticity of the virtual data, and the discriminator loss function is defined as LD=-E x～pdata(x) [logD(x)]-Ez~pz(z)[log(1-D(G(z)))] where z is the input noise variable of the generator, usually sampled from a standard normal distribution, x is a real data sample, p z (z) is the probability distribution of the noise variable, p data (x) is the probability distribution of the real data, and E is the expected value; by maximizing the correct classification of the discriminator for real data and virtual data, the generator is continuously improved. Through this antagonistic mechanism, the generative adversarial network can effectively improve the quality of virtual data, so that the generated data is highly consistent with the real data in statistical characteristics, and finally improves the accuracy and reliability of the algae monitoring model.

[0043] The training process generates virtual remote sensing data using the trained generator by alternately optimizing the generator and the discriminator to make the generated data have the same statistical characteristics as the real data; the generated virtual data is combined with the real data in proportion to form an expanded training data set, which is used for subsequent training of the algae monitoring model. Through the training of the generative adversarial network (GAN) model, the alternately optimized generator and discriminator ensure that the generated virtual data can realistically simulate the statistical characteristics of the real data, thereby effectively expanding the size and diversity of the training data set in the case of insufficient data or difficult data collection. Secondly, the generated virtual data is combined with the real data in proportion to create a more rich and comprehensive training data set, greatly improving the model training effect and enhancing the robustness and generalization ability of the model. In this way, the algae monitoring model can better capture the complex algae distribution characteristics and reduce the risk of overfitting, improving the accuracy and reliability of the model in practical applications. In addition, the introduction of virtual data can supplement and make up for the deficiencies in real data.

[0044] In this embodiment, the random forest algorithm is trained using the expanded training data set, and the steps for constructing the algae monitoring model are

[0045] The expanded training data set is divided into multiple subsets, and each subset is used to train a decision tree. In the training process of each tree, a part of the features is randomly selected for splitting to construct the structure of the tree, and the above process is repeated to construct multiple decision trees, and finally the algae monitoring model is formed. By dividing the expanded training data set into multiple subsets and training independent decision trees for each subset, the diversity of the data can be fully utilized to prevent overfitting problems that may be caused by a single data set. Secondly, in the training process of each tree, a part of the features is randomly selected for splitting, which effectively introduces randomness and increases the robustness of the model, ensuring that each tree learns independently on different feature subsets, thereby improving the generalization ability of the overall model. By constructing multiple decision trees and integrating the prediction results of these trees, the final random forest model can more accurately capture the complex characteristics of algae distribution, reduce the bias and variance of a single model, and improve the stability and accuracy of the prediction.

[0046] The embodiment further includes the following steps before obtaining the monitoring result:

[0047] Spectrum data of different algae is collected using a hyperspectral sensor, a spectrum database of algae is established, and spectrum features of different algae are analyzed and extracted from the spectrum data to identify key spectrum fingerprint information;

[0048] The support vector machine algorithm is used to construct an algae identification model based on spectral fingerprints. The extracted spectral features are used as input to train the SVM model, which is used for algae classification and identification of multi-source remote sensing data to obtain algae distribution information. The hyperspectral sensor is used to collect spectral data of different algae, and a comprehensive algae spectral database is established to provide a rich data basis for subsequent analysis. Second, the key spectral features and fingerprint information are extracted from the spectral data, which helps to accurately distinguish different algae and improve the accuracy of the classification model. The support vector machine (SVM) algorithm is used to construct an algae identification model, and the SVM model is trained to fully utilize the hyperspectral features for efficient algae classification and identification. This method not only enhances the generalization ability and robustness of the model, but also effectively handles high-dimensional data and improves the classification effect. Finally, the algae identification model based on spectral fingerprints can accurately identify and classify algae in multi-source remote sensing data, generate detailed algae distribution information, and provide scientific basis for ecological environment monitoring and water quality management, further improving the accuracy and reliability of algae monitoring.

[0049] In this embodiment, the trained algae monitoring model is used to predict multi-source remote sensing data, and the initial expression of algae content distribution is obtained as follows: Using the model parameters Θ to predict the input data X t , the predicted algae content

[0050] In this embodiment, the Kalman filter is used to assimilate the observation data and the model prediction data, and the model state is adjusted to be closer to the actual observation data value through the formula , where represents the posterior state estimate at time t, and the prior state estimate is adjusted by the Kalman gain K t combined with the observation data y t .

[0051] The calculation formula of the Kalman gain K t is K t = P t|t-1 H T (HP t|t-1 H T +R) -1 , which reflects the trade-off between observation uncertainty and prior state uncertainty.

[0052] The calculation formula of the posterior estimation covariance matrix P t|t is P t|t =(I-K t H)P t|t-1 , which is used to measure the confidence of the posterior state estimate, where is the posterior state estimate at time t, is the observation data at time t, P t is the observation data at time t, P t|t-1 is the prior estimation covariance matrix at time t, H is the observation matrix, R is the observation noise covariance matrix, and I is the identity matrix. Through assimilation processing, not only can model prediction and real-time observation data be comprehensively utilized to dynamically adjust the model state and improve the accuracy and real-time performance of the prediction, but also the confidence of the posterior estimation can be quantified to ensure the reliability and scientificity of the monitoring results. Kalman filtering effectively balances the uncertainty of the observation data and the prediction data to achieve accurate state estimation, greatly improving the precision of the algae content monitoring. Secondly, assimilation processing has the ability to process multi-source heterogeneous data, can update and correct the monitoring model in real time, ensure that it adapts to changing environmental conditions, and enhances the robustness and applicability of the system, improving the intelligentization and automation level of the overall monitoring system.

[0053] In this embodiment, after obtaining the monitoring results, the virtual model needs to be calibrated according to the state estimation after assimilation processing to ensure that the model state accurately reflects the actual situation, and the expression is: wherein, is the calibrated model parameter, Θ is the model parameter, T is the time step, y t is the observation data at time t, f(X t ; Θ) is the prediction output of the model, and the above formula represents that the sum of squares of errors between the observation data y t and the prediction output f(X t ; Θ) is minimized to optimize the model parameter Θ to obtain the calibrated model parameter By accurately calibrating the model parameter, the ability of the virtual model to reflect the actual situation can be significantly improved. First, minimizing the sum of squares of errors between the observation data and the prediction output ensures that the direction of model parameter adjustment is optimal, thereby improving the accuracy of model prediction. Secondly, through this calibration method, the cumulative error of the model at different time steps can be effectively reduced, and the stability and reliability of the model in long-term prediction can be improved. In addition, this method can comprehensively consider the dynamic changes in time series data, so that the model parameter calibration is not only suitable for the current state, but also adapts to future changes, enhancing the generalization ability of the model. The use of this formula can also quantify the effect of model parameter adjustment, providing a scientific basis for subsequent model optimization and improvement. Through accurate calibration of the virtual model, the high credibility of the monitoring results is ensured.

[0054] The present application is not limited to the above embodiments, and anyone should know that structural changes made under the inspiration of the present application fall within the scope of the present application. Any technical solution with the same or similar technical solutions as the present application falls within the scope of the present application. The technical, shape, and structure parts not described in detail in the present application are well-known technologies.

Claims

1. A method for monitoring algal content information based on multi-source remote sensing data, characterized in that, The method comprises the following steps: Obtain multi-source remote sensing data including satellite remote sensing data, unmanned aerial vehicle remote sensing data and ground monitoring data, and perform preprocessing such as radiation correction, geometric correction and noise removal on the data; Collect the preprocessed real remote sensing data, establish a training data set, and use a generative adversarial network model to train the training data to generate virtual data with the same statistical characteristics as the real data, combine the generated virtual data with the real data, expand the training data set, and use it for training of an algae monitoring model; Train a random forest algorithm using the expanded training data set to build an algae monitoring model; Fuse the multi-source data into the trained algae monitoring model for prediction, use Kalman filtering to assimilate the observation data and model prediction data, and adjust the model state according to the observation data and model prediction data to match the observation data to obtain algae content monitoring results; Verify the entire monitoring system using an independent ground monitoring data set to evaluate its accuracy and reliability. 2.The algal content information monitoring method based on multi-source remote sensing data according to claim 1, characterized in that, The multi-source remote sensing data includes but is not limited to Landsat, Sentinel-2, MODIS satellite data, unmanned aerial vehicle multi-spectral and hyperspectral data, and data collected by ground sensors. 3.The algal content information monitoring method based on multi-source remote sensing data according to claim 1, characterized in that, The generative adversarial network model comprises two parts of a generator and a discriminator, the generator is used for generating virtual data, and a formula L G = -E z～pz(z) [logD(G(z))] is used to define a generator loss function, the discriminator is used for evaluating the authenticity of the virtual data, and a formula LD=-E x～pdata(x) [logD(x)]-Ez~pz(z)[log(1-D(G(z)))] is used to define a discriminator loss function, wherein z is an input noise variable of the generator, is usually sampled from a standard normal distribution, x is a real data sample, p z (z) is a probability distribution of the noise variable, p data (x) is a probability distribution of the real data, and E is an expected value; The training process alternately optimizes the generator and the discriminator to make the generated data have the same statistical characteristics as the real data, and uses the trained generator to generate virtual remote sensing data; combine the generated virtual data with the real data in proportion to form an expanded training data set for subsequent training of an algae monitoring model. 4.The method of claim 1, wherein, The steps of training a random forest algorithm using the expanded training data set to build an algae monitoring model are Divide the expanded training data set into multiple subsets, and each subset is used to train a decision tree. In the training process of each tree, a part of features is randomly selected for splitting to build the structure of the tree, and the above process is repeated to build multiple decision trees to finally form an algae monitoring model. 5.The algal content information monitoring method based on multi-source remote sensing data according to claim 1, characterized in that, The following steps are included before obtaining the monitoring results: Use a hyperspectral sensor to collect spectral data of different algae, establish an algae spectral database, and analyze and extract spectral features of different algae from the spectral data to identify key spectral fingerprint information; Use a support vector machine algorithm to build an algae recognition model based on spectral fingerprints, use the extracted spectral features as input to train the SVM model, and perform algae classification and recognition on the multi-source remote sensing data to obtain algae distribution information.

6. The method according to claim 5, wherein, The trained algae monitoring model is used for prediction on multi-source remote sensing data, and an initial expression of algae content distribution is obtained. The input data X is predicted using the model parameters Θ t to obtain the predicted algae content 7. The method according to claim 6, wherein, The observation data and the model prediction data are assimilated using Kalman filtering, and the model state is adjusted to be closer to the value of the actual observation data by the formula The posterior state estimation at time t is represented by the prior state estimation The Kalman gain K t The observation data y t is combined to adjust the model state; Kalman gain K t The formula for calculating K t = P t|t-1 H T (HP t|t-1 H T +R) -1 reflects the trade-off between observation uncertainty and prior state uncertainty; posterior estimation covariance matrix P t|t The calculation formula of P t|t = (I - K t H)P t|t-1 , used to measure the confidence of the posterior state estimation, wherein, posterior state estimate for time t, For the prior state estimate at time t, y t For the observation data at time t, P t|t-1 Let H be the prior estimate covariance matrix at time t, H be the observation matrix, R be the observation noise covariance matrix, and I be the identity matrix. 8.The method of claim 1, wherein, After obtaining the monitoring results, the virtual model needs to be calibrated according to the state estimation after assimilation to ensure that the model state accurately reflects the actual situation, and the expression is: wherein, is the calibrated model parameter, Θ is the model parameter, T is the time step, y t is the observation data at time t, f(X t ; Θ) is the prediction output of the model.

Images