A lake pollutant detection method, system and storage medium based on remote sensing data

Abstract

The present invention provides a lake pollutant detection method, system, and storage medium based on remote sensing data, comprising: obtaining a remote sensing image of a lake to be tested; performing calculations on the remote sensing image of the lake to be tested using a corresponding inversion model to obtain inversion maps of several pollutant indicators for the lake; using the obtained remote sensing image of the lake to be tested as input to a trained neural network model, with each neural network model outputting a residual map of one of the pollutant indicators for the lake to be tested; calculating distribution maps of each pollutant indicator for the lake to be tested based on the inversion map of the pollutant indicator for the lake and the residual map of the pollutant indicator; generating a pollutant distribution information map for the lake to be tested based on the distribution maps of each pollutant indicator for the lake to be tested and an electronic map; the pollutant distribution information map is used to represent pollutant indicator data at each coordinate position in the lake to be tested. The present invention improves the accuracy of lake pollution inversion based on remote sensing images.

Description

Technical Field

[0001] The present invention relates to the field of water quality inversion, and in particular to a lake pollutant detection method and system based on remote sensing data and a storage medium thereof. Background Art

[0002] Traditional river and lake water quality monitoring mainly uses methods such as field sampling and laboratory analysis. This monitoring method needs to be carried out at fixed points and profiles in rivers and lakes. Although it can achieve a certain level of data accuracy through years of monitoring, recording and laboratory analysis, it cannot reflect the overall temporal and spatial conditions of river and lake water quality. It is also time-consuming and labor-intensive, with a limited monitoring area. It only has local and typical representative significance and cannot meet the requirements of real-time, rapid and large-scale monitoring and evaluation.

[0003] The development and advancement of remote sensing technology has opened up new avenues for monitoring and studying river and lake water bodies. Remote sensing water quality monitoring technology, with its remarkable characteristics of high dynamics, low cost, and macroscopic scope, offers irreplaceable advantages over conventional testing for studying water pollutants in rivers and lakes. It not only meets the needs of large-scale water quality monitoring but also reveals the spatial and temporal distribution and variability of water quality, thus addressing the shortcomings of surface sampling alone. It can also reveal the distribution of pollutants, their migration characteristics, and their impact ranges, which are difficult to reveal using conventional methods, providing a basis for the scientific placement of surface sampling points. Simultaneously, the number of water quality parameters that can be monitored has increased, and the accuracy of inversion models has continued to improve, playing a significant role in water resource conservation, planning, and sustainable development. However, limitations in the spatial and temporal resolution of satellite remote sensing imagery, as well as interference from factors such as landscape heterogeneity, atmospheric noise, and the position of the sun, can hinder the practical application of hyperspectral remote sensing technology, thereby affecting the final calculation results of the inversion model. Summary of the Invention

[0004] In response to the above problems, the present invention provides a lake pollutant detection method, system and storage medium based on remote sensing data, which effectively improves the accuracy of lake pollutant inversion based on remote sensing images.

[0005] To solve the above problems, the present invention provides a technical solution: a lake pollutant detection method based on remote sensing data, characterized by comprising the following steps:

[0006] S1, obtain remote sensing images of the lake to be measured;

[0007] S2, calculating the remote sensing image of the lake to be measured using the corresponding inversion model to obtain inversion maps of several pollutant indicators of the lake; the pollutant indicators include chlorophyll a, total nitrogen, total phosphorus, suspended matter concentration, and temperature;

[0008] S3, using the remote sensing image of the lake to be tested obtained in step S1 as input to multiple trained neural network models, each of which outputs a residual map of one of the pollutant indicators of the lake to be tested;

[0009] The residual map of each pollutant index is used to reflect the difference between the inversion map of the pollutant index calculated by the inversion model and the actual distribution data of the pollutant index in the lake to be tested;

[0010] S4, calculating the distribution map of each pollutant index of the lake to be tested based on the inversion map of each pollutant index of the lake obtained in step S2 and the residual map of each pollutant index obtained in step S3;

[0011] S5, generating a pollutant distribution information map of the lake to be tested based on the distribution map of each pollutant indicator and the electronic map of the lake to be tested; the pollutant distribution information map is used to characterize the chlorophyll a, total nitrogen, total phosphorus, suspended matter concentration and temperature data at each coordinate position in the lake to be tested.

[0012] In the above technical solution, the single sample information of the training set of the neural network model for outputting the residual map of any pollutant indicator includes: a remote sensing image of a lake as a model input and a residual map of the pollutant indicator of the lake as a training label;

[0013] Among them, the remote sensing image of a certain lake is obtained through the public remote sensing database;

[0014] Substituting the remote sensing image of the lake into the inversion model of the pollutant index to obtain an inversion map of the pollutant index of the lake;

[0015] The measured data and coordinate information of the pollutant index of each measuring point in the lake at the time corresponding to the remote sensing image are obtained, and the measured distribution map of the pollutant index of the lake at the time corresponding to the remote sensing image is generated in combination with the electronic map of the lake; the residual map of the pollutant index of the lake is generated according to the inversion map of the pollutant index of the lake and the measured distribution map of the pollutant index as the training label of the neural network model.

[0016] In the above technical solution, the method for generating any of the measured distribution maps of pollutant indicators comprises the following steps:

[0017] The acquisition time of the remote sensing image of a certain lake and the position of each measuring point in the electronic map of the certain lake are obtained, the measured data of the pollutant index at each measuring point in the lake at the acquisition time are obtained, and the corresponding measured data of the pollutant index collected at the position of each measuring point in the electronic map are calibrated as the measured distribution map of the pollutant index in the lake.

[0018] In the above technical solution, if the lake to be measured has several measuring points, a new training set and test set are generated based on the measured data and coordinate information of chlorophyll a, total nitrogen, total phosphorus, suspended matter concentration and temperature of each measuring point and the remote sensing image of the lake to be measured, and the new training set and test set are used to continue training the neural network model that has been trained; the retrained neural network model is used as the new trained neural network model to perform subsequent calculations.

[0019] In the above technical solution, the step of generating the pollutant distribution information map includes: obtaining the data of each pollutant indicator at each coordinate position in the lake to be tested according to the distribution map of each pollutant indicator of the lake to be tested, and calibrating the corresponding chlorophyll a, total nitrogen, total phosphorus, suspended matter concentration and temperature data at each coordinate position in the electronic map of the lake to be tested to form a pollutant distribution information map.

[0020] In the above technical solutions, the training process of each neural network model is the same, including the following steps:

[0021] Construct a training set based on the corresponding pollutant index data;

[0022] Set up the corresponding generator: the generator consists of several layers of neural networks, the input is the remote sensing image of the lake, and the corresponding output is the residual map of the corresponding pollutant index;

[0023] Set up the discriminator: The discriminator consists of several layers of neural networks and is used to determine whether the residual map of the corresponding pollutant indicator of a single sample is true;

[0024] The discriminator is used to conduct adversarial training on the generator using the training set.

[0025] In the above technical solution, the residual map of the corresponding pollutant index of a single sample generated by the generator is masked and then used for the discriminator calculation;

[0026] The masking process uses the following formula:

[0027]

[0028] G(x)=G′(x)×Mask(y)

[0029] Among them, y represents the residual map of the corresponding pollutant index of the input single sample; y(i, j) represents the true residual value of the residual map of the corresponding pollutant index of the input single sample at the grid (i, j), i∈[0, N), j∈[0, M), N*M is the coverage of the electronic map of the lake after rasterization processing; G'(x) is the original generator function, and G(x) is the generator function after mask processing.

[0030] In the above technical solution, the distribution map X of any pollutant index of the lake to be tested m The following formula is used for calculation;

[0031] X m =G m (x m )+Y m ;m=1, 2, 3, 4, 5, 6;

[0032] Among them, G m Represents the trained neural network model function corresponding to a certain pollutant indicator, x m Represents the remote sensing image of the lake to be measured; Y m The inversion diagram of the pollutant index generated in step S2 is shown.

[0033] The present invention also provides a lake pollutant detection system based on remote sensing data, comprising: a remote sensing image acquisition module, an inversion map generation module, a residual map calculation module, an index distribution map calculation module, and a pollutant distribution map calculation module; wherein the remote sensing image acquisition module is used to obtain remote sensing images of the lake to be tested; the inversion map generation module is used to calculate the remote sensing images of the lake to be tested through a corresponding inversion model, and obtain inversion maps of several pollutant indicators of the lake respectively; the pollutant indicators include chlorophyll a, total nitrogen, total phosphorus, suspended matter concentration, and temperature; the residual map calculation module is used to use the remote sensing images of the lake to be tested as inputs of multiple trained neural network models, and the multiple neural network models respectively output residual maps of one of the pollutant indicators of the lake to be tested; the index distribution map calculation module is used to calculate the distribution maps of various pollutant indicators of the lake to be tested based on the inversion maps of various pollutant indicators of the lake and the residual maps of various pollutant indicators; the pollutant distribution map calculation module is used to generate a pollutant distribution information map of the lake to be tested based on the distribution maps of various pollutant indicators of the lake to be tested and an electronic map.

[0034] The present invention also provides a computer-readable storage medium, which stores a program for a lake pollutant detection method based on remote sensing data. When the program for a lake pollutant detection method based on remote sensing data is executed by a processor, the steps of the lake pollutant detection method based on remote sensing data described in the above technical solution are implemented.

[0035] Compared with the existing technology, the present invention uses an inversion model to generate water quality inversion results, and uses a neural network model to correct the water quality inversion results, so as to avoid the problem that the satellite images of water quality remote sensing monitoring do not meet the water quality monitoring requirements, thereby causing deviations in the water quality calculation results.

[0036] The measured data of lake pollutant indicators of the present invention effectively guide and constrain the output of the neural network model, thereby accelerating the model training speed and improving the accuracy of lake pollutant detection.

[0037] This method leverages the powerful fitting and generalization capabilities of deep learning methods to assess lake pollutant concentrations using an inversion model. This method uses a pre-trained neural network model and outputs a residual graph of the inversion results, which is used to modify the inversion model output, thereby improving the accuracy of the inversion results.

[0038] At the same time, due to the powerful generalization ability of the neural network model, the model after the adversarial training is completed uses a small amount of partial measured point data to continue training the model, continuously improving the output accuracy of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a schematic diagram of the process of the present invention;

[0040] Figure 2 This is the lake chlorophyll-a inversion map of this specific embodiment;

[0041] Figure 3 This is the lake total nitrogen inversion map of this specific embodiment;

[0042] Figure 4 This is the lake total phosphorus inversion map of this specific embodiment;

[0043] Figure 5 This is the inversion map of lake suspended matter concentration in this specific embodiment. DETAILED DESCRIPTION

[0044] The present invention is further illustrated below with reference to specific examples. It should be understood that these examples are only used to illustrate the present invention and are not used to limit the scope of the present invention. After reading the present invention, modifications of various equivalent forms of the present invention made by those skilled in the art all fall within the scope defined by the claims attached to this application.

[0045] The present invention provides a lake pollutant detection method based on remote sensing data, which is characterized by comprising the following steps:

[0046] S1. Obtain remote sensing images of the lake to be measured. The characteristics of various entities can be reflected in remote sensing images. Polluted water has different spectral characteristics than clean water. These spectral characteristics are reflected in the absorption or reflection of specific wavelengths. These spectral characteristics can be captured by remote sensors and reflected in remote sensing images.

[0047] S2, calculate the remote sensing image of the lake to be measured through the corresponding inversion model to obtain the chlorophyll a inversion map of the lake (such as Figure 2 As shown), total nitrogen inversion diagram (as shown Figure 3 As shown), total phosphorus inversion map (as shown Figure 4 As shown), suspended matter concentration inversion diagram (as shown Figure 5 Temperature inversion diagram.

[0048] The commonly used method for inversion model of chlorophyll a is to establish empirical or semi-empirical models of the optimal band or various band combinations based on the sensitive band of chlorophyll a. There are three commonly used algorithms: (1) Determine the chlorophyll a inversion model based on the reflectance peak of chlorophyll a at 700nm; (2) Establish a chlorophyll a inversion model based on the proportional relationship between the reflectance peak of chlorophyll a at 700nm and the absorption peak at 675nm or 560nm; (3) Use a semi-empirical aerial monitoring method to utilize the effect of chlorophyll a on the reflectance of water bodies in the blue-green light band and establish a chlorophyll a concentration inversion model based on the difference or ratio of the spectral reflectance of these two bands.

[0049] The inversion model for total nitrogen and total phosphorus is primarily based on the close correlation between total nitrogen, total phosphorus, and chlorophyll content, establishing remote sensing information models for these parameters. By establishing an artificial neural network inversion model for six parameters—total nitrogen, total phosphorus, chlorophyll, suspended solids, chemical oxygen demand, and dissolved oxygen—for a specific region, it is possible to effectively invert lake water quality parameters from remote sensing imagery, with inversion errors generally kept below 25%.

[0050] When the suspended solids concentration is greater than 350 mg / L, the spectral reflectance is linearly related to the suspended solids concentration. The estimation accuracy of the quadratic regression model using the first-order differential of the reflectance is higher than that using the peak reflectance. The inversion model for suspended solids uses a classic model that monitors the relatively stable suspended solids concentration.

[0051] Remote sensing monitoring of water temperatures can be achieved through a combination of thermal infrared and microwave remote sensing, using optical techniques or computers to perform density segmentation on images. Lake temperature can be inverted using the radiation propagation equation using the 10th band of Landsat 8, serving as a temperature inversion model.

[0052] S3, using the remote sensing images of the lake to be tested obtained in step S1 as inputs to multiple trained neural network models, and the multiple neural network models respectively output chlorophyll a residual map, total nitrogen residual map, total phosphorus residual map, suspended matter concentration residual map, and temperature residual map of the lake to be tested;

[0053] The chlorophyll a residual map is used to reflect the difference between the chlorophyll a inversion map of the lake to be tested calculated by the inversion model and the actual distribution data of chlorophyll a in the lake to be tested;

[0054] The total nitrogen residual map is used to reflect the difference between the total nitrogen inversion map of the lake to be measured calculated by the inversion model and the actual distribution data of the total nitrogen in the lake to be measured;

[0055] The total phosphorus residual map is used to reflect the difference between the total phosphorus residual map of the lake to be tested calculated by the inversion model and the actual distribution data of the total phosphorus in the lake to be tested;

[0056] The suspended matter concentration residual map is used to reflect the difference between the suspended matter concentration inversion map of the lake to be measured calculated by the inversion model and the actual distribution data of the suspended matter concentration of the lake to be measured;

[0057] The temperature residual map is used to reflect the difference between the temperature inversion map of the lake to be measured calculated by the inversion model and the real distribution data of the temperature of the lake to be measured;

[0058] S4, calculating the chlorophyll a distribution map, total nitrogen distribution map, total phosphorus distribution map, suspended matter concentration distribution map, and temperature distribution map of the lake to be tested based on the chlorophyll a inversion map, total nitrogen inversion map, total phosphorus inversion map, suspended matter concentration inversion map, and temperature inversion map of the lake obtained in step S2 and the chlorophyll a residual map, total nitrogen residual map, total phosphorus residual map, suspended matter concentration residual map, and temperature residual map obtained in step S3;

[0059] S5. Generate a pollutant distribution information map of the lake to be tested based on the chlorophyll a distribution map, total nitrogen distribution map, total phosphorus distribution map, suspended matter concentration distribution map, temperature distribution map and electronic map of the lake to be tested; the pollutant distribution information map is used to characterize the chlorophyll a, total nitrogen, total phosphorus, suspended matter concentration and temperature data at each coordinate position in the lake to be tested.

[0060] In the above technical solution, the single sample information of the training set of the neural network model for outputting the chlorophyll a residual map includes a remote sensing image of a lake as a model input and the chlorophyll a residual map of the lake as a training label;

[0061] Among them, the remote sensing image of a certain lake is obtained through the public remote sensing database;

[0062] Substituting the real-time remote sensing image of the lake into the chlorophyll a inversion model, the chlorophyll a inversion map of the lake is obtained, as shown in Figure 2 As shown;

[0063] The chlorophyll a measured data and coordinate information of each current measurement point in the lake are obtained, and the measured chlorophyll a distribution map of the lake is generated in combination with the electronic map of the lake; the chlorophyll a residual map of the lake is generated based on the chlorophyll a inversion map and the measured chlorophyll a distribution map of the lake, which serves as the training label of the neural network model.

[0064] The single sample information of the training set of the neural network model for outputting the total nitrogen residual map includes a remote sensing image of a lake as a model input and a total nitrogen residual map of the lake as a training label;

[0065] Among them, the remote sensing image of a certain lake is obtained through the public remote sensing database;

[0066] Substituting the real-time remote sensing image of the lake into the total nitrogen inversion model, the total nitrogen inversion map of the lake is obtained;

[0067] The chlorophyll measured data and coordinate information of each current measurement point in the lake are obtained, and the total nitrogen measured distribution map of the lake is generated in combination with the electronic map of the lake; the total nitrogen residual map of the lake is generated according to the total nitrogen inversion map and the total nitrogen measured distribution map of the lake as the training label of the neural network model.

[0068] The single sample information of the training set of the neural network model for outputting the total phosphorus residual map includes a remote sensing image of a lake as the model input and the total phosphorus residual map of the lake as the training label;

[0069] Among them, the remote sensing image of a certain lake is obtained through the public remote sensing database;

[0070] Substituting the real-time remote sensing image of the lake into the total phosphorus inversion model, the total phosphorus inversion map of the lake was obtained;

[0071] The chlorophyll measured data and coordinate information of each current measurement point in the lake are obtained, and the total phosphorus measured distribution map of the lake is generated in combination with the electronic map of the lake; the total phosphorus residual map of the lake is generated based on the total phosphorus inversion map and the total phosphorus measured distribution map of the lake as the training label of the neural network model.

[0072] The single sample information of the training set of the neural network model for outputting the suspended matter concentration residual map includes a remote sensing image of a lake as the model input and the suspended matter concentration residual map of the lake as the training label;

[0073] Among them, the remote sensing image of a certain lake is obtained through the public remote sensing database;

[0074] Substituting the real-time remote sensing image of the lake into the inversion model of suspended matter concentration, the inversion map of suspended matter concentration of the lake is obtained;

[0075] The chlorophyll measured data and coordinate information of each current measurement point in the lake are obtained, and the electronic map of the lake is combined to generate the measured distribution map of the suspended matter concentration of the lake; based on the suspended matter concentration inversion map and the measured distribution map of the suspended matter concentration of the lake, the residual map of the suspended matter concentration of the lake is generated as the training label of the neural network model.

[0076] The single sample information of the training set of the neural network model for outputting the temperature residual map includes a remote sensing image of a lake as the model input and the temperature residual map of the lake as the training label;

[0077] Among them, the remote sensing image of a certain lake is obtained through the public remote sensing database;

[0078] Substituting the real-time remote sensing image of the lake into the temperature inversion model, a temperature inversion map of the lake is obtained;

[0079] The chlorophyll measured data and coordinate information of each current measurement point in the lake are obtained, and the measured temperature distribution map of the lake is generated in combination with the electronic map of the lake; the temperature residual map of the lake is generated according to the temperature inversion map and the measured temperature distribution map of the lake, which serves as the training label of the neural network model.

[0080] In the above technical solution, the method for generating the measured chlorophyll a distribution map comprises the following steps:

[0081] Obtain the coordinate position of each measurement point in the electronic map of a lake, obtain the measured chlorophyll a data of each measurement point in the lake, and calibrate the corresponding collected measured chlorophyll a data at the coordinate position of each measurement point in the electronic map as the measured chlorophyll a distribution map of the lake.

[0082] In a lake with a simple environment, making it easy to obtain chlorophyll a concentration at various locations in real time, the following method is used: Based on the concentration range of 0-0.2 on the chlorophyll a inversion map of the lake, an average of 20 points is taken with 0.01 as the dividing segment. That is, based on the chlorophyll a inversion map, the most typical coordinate positions corresponding to chlorophyll a concentrations of 0, 0.01, 0.02…, and 0.20 are found. The standard for determining the most typical coordinate position is to identify a color block area corresponding to a single concentration value and obtain the coordinates of the center point of the color block area as the most typical coordinate position corresponding to that concentration value. The operator measures the chlorophyll a concentration at the most typical coordinate position corresponding to each concentration value determined in the lake in real time. Based on each most typical coordinate position and its corresponding measured chlorophyll a concentration value, a measured chlorophyll a distribution map is drawn.

[0083] In the case of a complex lake environment where each chlorophyll a concentration measurement point can only cover a local area:

[0084] Based on the measured values ​​and coordinates of the chlorophyll a concentration measurement points, a measured chlorophyll a map for the region is generated. Based on the coordinate range of the region, a corresponding chlorophyll a inversion map and remote sensing image are obtained. Based on the chlorophyll a inversion map and the measured chlorophyll a map for the region, a chlorophyll a residual map is generated for the region. The chlorophyll a residual map and remote sensing image for the region are used to train the corresponding neural network model.

[0085] The method for generating the measured total nitrogen distribution map comprises the following steps:

[0086] The positions of the measurement points in the electronic map of a lake are obtained, the total nitrogen measured data of the measurement points in the lake are obtained, and the corresponding collected total nitrogen measured data are calibrated at the positions of the measurement points in the electronic map to serve as the total nitrogen measured distribution map of the lake.

[0087] In a lake with a simple environment, allowing for easy real-time acquisition of total nitrogen concentrations at various locations, the following is done: Based on the concentration range of 0-4.0 on the total nitrogen inversion map for the lake, an average of 40 points is taken, divided by 0.1. This means finding the most typical coordinate locations corresponding to total nitrogen concentrations of 0, 0.1, 0.2, ..., and 4.0, respectively, based on the total nitrogen inversion map. The standard for determining the most typical coordinate locations is to identify a color block area corresponding to a single concentration value and obtain the coordinates of the center point of this color block area as the most typical coordinate location corresponding to that concentration value. Operators measure the total nitrogen concentration at the most typical coordinate locations corresponding to each concentration value determined for the lake in real time. Based on each most typical coordinate location and its corresponding measured total nitrogen concentration value, a total nitrogen distribution map is drawn.

[0088] In the case of a complex lake environment where each total nitrogen concentration measurement point can only cover a local area:

[0089] Based on the measured values ​​and coordinates of the total nitrogen concentration measurement points, a total nitrogen measured map of the area is generated. Based on the coordinate range of the area, a corresponding total nitrogen inversion map and remote sensing image are obtained. Based on the total nitrogen inversion map and the measured total nitrogen map, a total nitrogen residual map is generated for the area. The corresponding neural network model is trained using the total nitrogen residual map and remote sensing image for the area.

[0090] The method for generating the total phosphorus measured distribution map comprises the following steps:

[0091] Obtain the position of each measurement point in the electronic map of a lake, obtain the total phosphorus measured data of each measurement point in the lake, and calibrate the corresponding collected total phosphorus measured data at the position of each measurement point in the electronic map as the total phosphorus measured distribution map of the lake.

[0092] In the case of a lake with a simple environment, which makes it easy to obtain the total phosphorus concentration at each location in real time: based on the concentration range of 0-0.20 on the total phosphorus inversion map of the lake, an average of 20 points is taken with 0.01 as the dividing segment. That is, based on the total phosphorus inversion map, the most typical coordinate positions corresponding to total phosphorus concentrations of 0, 0.01, 0.02, ..., 0.20 are found. The standard for determining the most typical coordinate position is to determine the color block area corresponding to a single concentration value, obtain the coordinates of the center point of the color block area, and use it as the most typical coordinate position corresponding to that concentration value. The operator measures the total phosphorus concentration at the most typical coordinate position corresponding to each concentration value determined in the lake in real time, and draws a total phosphorus measured distribution map based on each most typical coordinate position and its corresponding total phosphorus concentration measured value.

[0093] In the case of a complex lake environment where each total phosphorus concentration measurement point can only cover a local area:

[0094] Based on the measured values ​​and coordinates of the total phosphorus concentration measurement points, a total phosphorus measurement map for the area is generated. Based on the coordinate range of the area, a corresponding total phosphorus inversion map and remote sensing image are obtained. Based on the total phosphorus inversion map and the measured total phosphorus map for the area, a total phosphorus residual map for the area is generated. The corresponding neural network model is trained using the total phosphorus residual map and remote sensing image for the area.

[0095] The method for generating the measured distribution map of suspended matter concentration comprises the following steps:

[0096] The positions of the various measurement points in the electronic map of a lake are obtained, the measured data of the suspended matter concentration at the various measurement points in the lake are obtained, and the corresponding collected measured data of the suspended matter concentration are calibrated at the positions of the various measurement points in the electronic map as the measured distribution map of the suspended matter concentration of the lake.

[0097] In a lake with a simple environment, which makes it easy to obtain suspended solids concentrations at various locations in real time, the following is done: based on the concentration range of 0-80 on the suspended solids inversion map of the lake, an average of 80 points is taken with 1.0 as the dividing segment. That is, based on the suspended solids inversion map, the most typical coordinate positions corresponding to suspended solids concentrations of 0, 1.0, 2.0, ..., 80 are found. The standard for determining the most typical coordinate positions is to determine the color block area corresponding to a single concentration value, obtain the coordinates of the center point of the color block area, and use this as the most typical coordinate position corresponding to that concentration value. The operator measures the suspended solids concentration at the most typical coordinate positions corresponding to each concentration value determined for the lake in real time, and draws a measured suspended solids distribution map based on each most typical coordinate position and its corresponding measured suspended solids concentration value.

[0098] In the case of a complex lake environment where each suspended matter concentration measurement point can only cover a local area:

[0099] Based on the measured values ​​and coordinates of the suspended matter concentration measurement points, a measured suspended matter map for the area is generated. Based on the coordinate range of the area, a corresponding suspended matter inversion map and remote sensing image are obtained. Based on the suspended matter inversion map and the measured suspended matter map for the area, a suspended matter residual map is generated for the area. The corresponding neural network model is trained using the suspended matter residual map and remote sensing image for the area.

[0100] The method for generating the measured temperature distribution map comprises the following steps:

[0101] The positions of the various measurement points in the electronic map of a lake are obtained, the measured temperature data of the various measurement points in the lake are obtained, and the corresponding collected measured temperature data are calibrated at the positions of the various measurement points in the electronic map as the measured temperature distribution map of the lake.

[0102] In the above technical solution, if the lake to be measured has several measuring points, a new training set and test set are generated based on the measured data and coordinate information of chlorophyll a, total nitrogen, total phosphorus, suspended matter concentration and temperature of each measuring point and the remote sensing image of the lake to be measured, and the new training set and test set are used to continue training the neural network model that has been trained; the retrained neural network model is used as the new trained neural network model to perform subsequent calculations.

[0103] In the above technical solution, the step of generating the pollutant distribution information map includes: obtaining the chlorophyll a, total nitrogen, total phosphorus, suspended matter concentration and temperature data of each coordinate position in the lake to be tested according to the chlorophyll a distribution map, total nitrogen distribution map, total phosphorus distribution map, suspended matter concentration distribution map and temperature distribution map of the lake to be tested, and calibrating the corresponding chlorophyll a, total nitrogen, total phosphorus, suspended matter concentration and temperature data at each coordinate position in the electronic map of the lake to be tested to form a pollutant distribution information map.

[0104] In the above technical solutions, the training process of the neural network model is the same, including the following steps:

[0105] Construct the corresponding training set;

[0106] Set up the corresponding generator: the generator consists of several layers of neural networks. The input is the remote sensing image of the lake, and the corresponding output is the chlorophyll a residual map, the total nitrogen residual map, the total phosphorus residual map, the suspended matter concentration residual map or the temperature residual map;

[0107] Set up the discriminator: The discriminator consists of several layers of neural networks and is used to determine whether the chlorophyll a residual map, total nitrogen residual map, total phosphorus residual map, suspended solids concentration residual map, or temperature residual map of a single sample is true;

[0108] The discriminator is used to conduct adversarial training on the generator using the training set.

[0109] In the above technical solution, the chlorophyll a residual map, total nitrogen residual map, total phosphorus residual map, suspended matter concentration residual map or temperature residual map of a single sample generated by the generator is masked and then used for the discriminator calculation;

[0110] The masking process uses the following formula:

[0111]

[0112] G(x)=G′(x)×Mask(y)

[0113] Where y represents the chlorophyll a residual map, total nitrogen residual map, total phosphorus residual map, suspended matter concentration residual map or temperature residual map of the lake corresponding to the input single sample; y(i, j) represents the true residual value of the chlorophyll a residual map, total nitrogen residual map, total phosphorus residual map, suspended matter concentration residual map or temperature residual map of the input single sample at the grid (i, j), i∈[0, N), j∈[0, M), N*M is the coverage of the electronic map of the lake after rasterization processing; G'(x) is the original generator function, and G(x) is the generator function after mask processing.

[0114] In the above technical solution, the chlorophyll a distribution map X1, total nitrogen distribution map X2, total phosphorus distribution map X3, suspended matter concentration distribution map X4, and temperature distribution map X5 of the lake to be tested are all calculated using the following formula;

[0115] X m =G m (x m )+Y m ; m = 1, 2, 3, 4, 5;

[0116] Among them, G m Represents the trained neural network model function, x m Represents the remote sensing image of the lake to be measured; Y m It represents the chlorophyll a inversion map, total nitrogen inversion map, total phosphorus inversion map, suspended matter concentration inversion map or temperature inversion map of the lake to be tested generated in step S2.

[0117] The present invention also provides a lake pollutant detection system based on remote sensing data, comprising: a remote sensing image acquisition module, an inversion map generation module, a residual map calculation module, an index distribution map calculation module, and a pollutant distribution map calculation module; wherein the remote sensing image acquisition module is used to obtain remote sensing images of the lake to be tested; the inversion map generation module is used to calculate the remote sensing images of the lake to be tested through the corresponding inversion model to obtain the chlorophyll a inversion map, total nitrogen inversion map, total phosphorus inversion map, suspended matter concentration inversion map, and temperature inversion map of the lake; the residual map calculation module is used to use the remote sensing images of the lake to be tested as inputs of multiple trained neural network models, and the multiple neural network models respectively output the chlorophyll a inversion map of the lake to be tested. a residual map, total nitrogen residual map, total phosphorus residual map, suspended matter concentration residual map, and temperature residual map of the lake; the indicator distribution map calculation module is used to calculate the chlorophyll a distribution map, total nitrogen distribution map, total phosphorus distribution map, suspended matter concentration distribution map, and temperature distribution map of the lake based on the chlorophyll a inversion map, total nitrogen inversion map, total phosphorus inversion map, suspended matter concentration inversion map, and temperature inversion map of the lake and the chlorophyll a residual map, total nitrogen residual map, total phosphorus residual map, suspended matter concentration residual map, and temperature residual map; the pollutant distribution map calculation module is used to generate a pollutant distribution information map of the lake to be tested based on the chlorophyll a distribution map, total nitrogen distribution map, total phosphorus distribution map, suspended matter concentration distribution map, temperature distribution map, and electronic map of the lake to be tested.

[0118] The present invention also provides a computer-readable storage medium, which stores a program for a lake pollutant detection method based on remote sensing data. When the program for a lake pollutant detection method based on remote sensing data is executed by a processor, the steps of the lake pollutant detection method based on remote sensing data as described in the above technical solution are implemented.

[0119] The specific embodiment of the present invention collected measured remote sensing images and water sample data of 32 lakes, and after eliminating abnormal data, 24 groups of valid data were obtained. The remote sensing reflectance data of the water body at the sampling point was measured and calculated using the ASD-FieldSpec HandHeld 2 ground feature spectrometer. The turbidity was measured on-site using the HACH 2100Q portable turbidity meter. According to the "Technical Specifications for Surface Water and Sewage Monitoring", water samples were collected and preserved, and brought back to the laboratory for spectrophotometric measurement of chlorophyll a concentration (Chl-a), total nitrogen (TN), and total phosphorus (TP). The total suspended solids concentration (TSM) of the water body was measured by weighing, and the water temperature was measured using a thermometer. The concentration distribution of various water quality parameters at the measured sampling points is shown in Table 1.

[0120] Table 1 Distribution of water quality parameter concentrations at sampling points

[0121]

[0122] This study used a Micro Hyperspec VNIR A-Series hyperspectral imager (HeadWallPhotonics Inc.), which has a spectral range of 400–1000 nm, 325 spectral channels, and a spectral resolution of 1.8 nm, to acquire hyperspectral data for 11 lakes. Based on the quality of hyperspectral remote sensing images, 218 spectral bands in the 400–800 nm range were selected. Remote sensing images were acquired between 9:00–11:00 AM and 1:00–3:00 PM.

[0123] During the hyperspectral imaging data acquisition process, the geographical location and attitude information of the image were obtained simultaneously, and high-precision geometric correction was performed through data processing software. The urban surface water hyperspectral anti-shadow water index method was used to extract river and lake water bodies.

[0124] The present invention adopts fast Fourier transform (FFT) to remove the stripe noise of hyperspectral images and adopts the noise-sensitive minimum noise fraction (MNF) method to remove the spectral noise.

[0125] By performing statistical analysis on the synchronously measured remote sensing reflectance and water quality parameters (Chl-a, TSM, TN, TP, and temperature), the bands or band combinations with the highest correlation among different water quality parameters were selected as independent variables. The corresponding inversion models were constructed using exponential, power function, linear, polynomial, and partial least squares regression methods. The chlorophyll a inversion map, total nitrogen inversion map, total phosphorus inversion map, suspended matter concentration inversion map, and temperature inversion map of the lake were calculated based on the inversion model.

[0126] Remote sensing images of the lake to be tested are used as input to multiple trained neural network models, which then output chlorophyll-a residual maps, total nitrogen residual maps, total phosphorus residual maps, suspended matter concentration residual maps, and temperature residual maps for the lake to be tested. Based on the chlorophyll-a inversion maps, total nitrogen inversion maps, total phosphorus inversion maps, suspended matter concentration inversion maps, and temperature inversion maps of the lake and the chlorophyll-a residual maps, total nitrogen residual maps, total phosphorus residual maps, suspended matter concentration residual maps, and temperature residual maps, the chlorophyll-a distribution map, total nitrogen distribution map, total phosphorus distribution map, suspended matter concentration distribution map, and temperature distribution map of the lake to be tested are calculated.

[0127] The comparison results with the measured data (coefficient of determination R2 and root mean square error) show that the detection accuracy is relatively high (Chl-a: R2 = 0.83, RMSE = 2.1 mg / m3; TSM: R2 = 0.82, RMSE = 5.16 mg / L; TN: R2 = 0.59, RMSE = 0.24 mg / L; TP: R2 = 0.81, RMSE = 0.12 mg / L)

[0128] In the foregoing detailed description, various features are grouped together in a single embodiment to simplify the disclosure. This method of disclosure should not be interpreted as reflecting an intention that embodiments of the claimed subject matter require more features than are expressly recited in each claim. On the contrary, as reflected in the appended claims, the invention comprises less than all the features of any individual disclosed embodiment. The appended claims are hereby expressly incorporated into the detailed description, with each claim standing on its own as a separate preferred embodiment of the invention.

[0129] The above description of the disclosed embodiments is intended to enable any person skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other embodiments without departing from the spirit and scope of the present disclosure. Therefore, the present disclosure is not limited to the embodiments presented herein but is intended to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0130] The foregoing description includes examples of one or more embodiments. Of course, it is not possible to describe all possible combinations of components or methods for the purposes of describing the above embodiments, but one of ordinary skill in the art will recognize that the various embodiments may be further combined and arranged. Therefore, the embodiments described herein are intended to encompass all such changes, modifications and variations that fall within the scope of the appended claims. Furthermore, to the extent the term "comprising" is used in the specification or claims, the term is intended to be encompassed in a manner similar to the term "including," as explained in terms of "including," used as a transitional word in the claims. Furthermore, any use of the term "or" in the specification of the claims is intended to mean a "non-exclusive or."

[0131] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A lake pollutant detection method based on remote sensing data, characterized in that: The steps include: S1: Obtain remote sensing images of the lake to be measured; S2: Calculating the remote sensing image using a corresponding inversion model to obtain inversion maps of several pollutant indicators of the lake, wherein the pollutant indicators include chlorophyll a, total nitrogen, total phosphorus, suspended matter concentration, and temperature; S3: inputting the remote sensing image obtained in step S1 into a plurality of independent neural network models that have been trained and correspond to different pollutant indicators, respectively, and the plurality of neural network models respectively output a residual map of one of the pollutant indicators of the lake, the residual map being used to reflect the difference between the inversion map of the pollutant indicator obtained by the inversion model and the actual distribution data of the pollutant indicator of the lake; wherein the training label of the neural network model is: the pixel-by-pixel difference between the measured distribution map generated by the electronic map of the lake and the corresponding inversion map, the measured distribution map being generated by combining the coordinates of the measurement points of the lake at the time of remote sensing image acquisition with the measured values ​​on the electronic map; and performing mask processing on the residual map according to the water area mask generated by the lake boundary on the electronic map, so as to set the residual values ​​outside the water area of ​​the lake to zero; S4: Based on the inversion map of each pollutant index obtained in step S2 and the residual map of each pollutant index obtained in step S3, the corresponding residual map and the inversion map are added and fused pixel by pixel within the mask range of the lake water area to calculate the concentration distribution map matrix of each pollutant index of the lake to be tested; wherein the distribution map of each pollutant index satisfies the equation: Xm=Ym+Gm(x m ) m∈{chlorophyll a, total nitrogen, total phosphorus, suspended matter concentration, temperature}, Y m is the inversion diagram of the corresponding pollutant index, Gm(x m ) is the residual graph of the neural network output corresponding to the pollutant index; S5: Generate a pollutant distribution information map of the lake based on the pollutant index distribution map and the electronic map of the lake, and mark the corresponding chlorophyll a, total nitrogen, total phosphorus, suspended matter concentration and temperature values ​​point by point at the coordinate positions on the electronic map.

2. The lake pollutant detection method based on remote sensing data according to claim 1, characterized in that: The method for generating a measured distribution map of any pollutant index includes the following steps: The acquisition time of the remote sensing image of a certain lake and the position of each measuring point in the electronic map of the certain lake are obtained, the measured data of the pollutant index at each measuring point in the lake at the acquisition time are obtained, and the corresponding measured data of the pollutant index collected at the position of each measuring point in the electronic map are calibrated as the measured distribution map of the pollutant index in the lake.

3. The lake pollutant detection method based on remote sensing data according to claim 1, characterized in that: If the lake to be measured has several measuring points, new training sets and test sets are generated based on the measured data and coordinate information of chlorophyll a, total nitrogen, total phosphorus, suspended matter concentration and temperature of each measuring point and the remote sensing image of the lake to be measured, and the new training sets and test sets are used to continue training the neural network model that has been trained; the retrained neural network model is used as the new trained neural network model to perform subsequent calculations.

4. The lake pollutant detection method based on remote sensing data according to claim 1, characterized in that: The steps for generating the pollutant distribution information map include: obtaining the data of each pollutant indicator at each coordinate position in the lake to be tested according to the distribution map of each pollutant indicator of the lake to be tested, and calibrating the corresponding chlorophyll a, total nitrogen, total phosphorus, suspended matter concentration and temperature data at each coordinate position in the electronic map of the lake to be tested to form a pollutant distribution information map.

5. The lake pollutant detection method based on remote sensing data according to claim 1, characterized in that: The training process for each neural network model is the same and consists of the following steps: Construct a training set based on the corresponding pollutant index data; Set up the corresponding generator: the generator consists of several layers of neural networks, the input is the remote sensing image of the lake, and the corresponding output is the residual map of the corresponding pollutant index; Set up the discriminator: The discriminator consists of several layers of neural networks and is used to determine whether the residual map of the corresponding pollutant indicator of a single sample is true; The discriminator is used to conduct adversarial training on the generator using the training set.

6. The method for detecting lake pollutants based on remote sensing data according to claim 5, characterized in that: The residual map of the corresponding pollutant index of a single sample generated by the generator is masked and then used for discriminator calculation.

7. A lake pollutant detection system based on remote sensing data, characterized in that Used to implement the method described in claim 1, comprising: a remote sensing image acquisition module, an inversion map generation module, a residual map calculation module, an index distribution map calculation module, and a pollutant distribution map calculation module; wherein the remote sensing image acquisition module is used to obtain remote sensing images of the lake to be tested; the inversion map generation module is used to calculate the remote sensing images of the lake to be tested through the corresponding inversion model, and obtain inversion maps of several pollutant indicators of the lake respectively; the pollutant indicators include chlorophyll a, total nitrogen, total phosphorus, suspended matter concentration, and temperature; the residual map calculation module is used to use the remote sensing images of the lake to be tested as inputs of multiple trained neural network models, and the multiple neural network models respectively output residual maps of one of the pollutant indicators of the lake to be tested; the index distribution map calculation module is used to calculate the distribution maps of various pollutant indicators of the lake to be tested based on the inversion maps of various pollutant indicators of the lake and the residual maps of various pollutant indicators; the pollutant distribution map calculation module is used to generate a pollutant distribution information map of the lake to be tested based on the distribution maps of various pollutant indicators of the lake to be tested and the electronic map.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a program for a lake pollutant detection method based on remote sensing data. When the program for a lake pollutant detection method based on remote sensing data is executed by a processor, the steps of the lake pollutant detection method based on remote sensing data as described in any one of claims 1 to 6 are implemented.

Images