Water pollution source identification method and system based on UAV multispectral imaging

Abstract

The present invention discloses a method and system for identifying water pollution sources based on multispectral images of unmanned aerial vehicles, including: controlling a unmanned aerial vehicle equipped with a multispectral imaging device to inspect a target water area to obtain a multispectral image, and using the positioning system of the unmanned aerial vehicle to record the image geographical location information corresponding to the multispectral image; preprocessing the multispectral image to obtain a preprocessed multispectral image; extracting spectral features from the preprocessed multispectral image to obtain multispectral features; inputting the multispectral features into a trained water pollution source type prediction model, outputting a water pollution source type probability vector corresponding to each pixel of the preprocessed multispectral image, and determining the pollution source type corresponding to each pixel; determining the location distribution of the water pollution source according to the geographical location information of each pixel of the preprocessed multispectral image and the determined pollution source type. The present invention can improve the accuracy and timeliness of identifying water pollution sources.

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned aerial vehicles, and in particular to a method and system for identifying water pollution sources based on unmanned aerial vehicle multispectral images. Background Art

[0002] In the field of water environment monitoring, traditional methods for identifying water pollution sources mainly rely on on-site sampling and laboratory analysis. This method requires staff to collect water samples at multiple sampling points in the target waters, and then bring the water samples back to the laboratory to use chemical analysis methods to detect the composition and concentration of various pollutants. For example, for industrial wastewater discharge monitoring, it is necessary to set up sampling points near the factory outlet and at different distances downstream, collect water samples regularly, and determine indicators such as heavy metal content and organic pollutant concentration through complex chemical analysis processes, and then judge the type of pollution source and the degree of pollution.

[0003] However, this traditional method has many limitations. First, the spatial coverage of on-site sampling is limited, and only information from discrete sampling points can be obtained, which makes it difficult to fully reflect the pollution status of the entire target water area. For large areas of water, it is easy to miss some potential pollution source areas. Secondly, the sampling and analysis process is time-consuming. It often takes hours or even days from sampling to obtaining results. It is impossible to provide pollution information in a timely manner, resulting in delayed response when dealing with sudden water pollution incidents. Furthermore, traditional methods are costly, including labor costs, sampling equipment costs, and chemical analysis reagent costs, and require professional technicians to operate and analyze, which to a certain extent limits large-scale and high-frequency monitoring applications.

[0004] In addition, with the development of remote sensing technology, some water pollution source identification methods based on satellite remote sensing images have been proposed. These methods use multispectral sensors carried by satellites to obtain remote sensing images of target waters, and then identify water pollution sources by analyzing the spectral characteristics of the images. For example, by comparing the spectral changes of the same waters in different periods, combined with the known spectral feature library of pollution sources, the type of pollution source that may exist can be determined. However, there are problems of inaccurate identification and poor timeliness in the use of satellite remote sensing images for water pollution source identification. On the one hand, the spectral signals received by satellite sensors will be disturbed by complex factors such as atmospheric scattering and absorption when passing through the atmosphere, which causes the acquired water body spectral information to be distorted, resulting in large deviations in the pollution source identification results based on this analysis. On the other hand, the satellite revisit cycle is long, and it is impossible to conduct high-frequency monitoring of specific waters, and it is difficult to obtain pollution information with strong timeliness, and it is impossible to capture the dynamic changes of pollution in a timely manner, which leads to large deviations and uncertainties in the identification of pollution sources and the assessment of pollution status.

[0005] Therefore, existing water pollution source identification technologies, whether traditional on-site sampling and analysis or satellite remote sensing imaging methods, have problems with poor timeliness and inaccurate identification, making it difficult to efficiently and accurately achieve the task of identifying and monitoring water pollution sources in target waters. Summary of the invention

[0006] The embodiments of the present invention provide a method and system for identifying water pollution sources based on multispectral images of unmanned aerial vehicles, which can improve the accuracy and timeliness of identifying water pollution sources.

[0007] An embodiment of the present invention provides a method for identifying water pollution sources based on multispectral images of unmanned aerial vehicles, comprising the following steps:

[0008] Controlling a drone equipped with a multispectral imaging device to inspect the target waters to obtain a multispectral image, and using a positioning system of the drone to record image geographic location information corresponding to the multispectral image;

[0009] The multispectral image is subjected to radiation correction by using a preset radiation transfer correction model, and the multispectral image is subjected to geometric correction and pixel geographical location information registration by using the image geographical location information and ground control points to obtain a preprocessed multispectral image;

[0010] Extracting spectral features from the preprocessed multispectral image to obtain multispectral features;

[0011] Input the multispectral features into a trained water pollution source type prediction model, and output a water pollution source type probability vector corresponding to each pixel of the preprocessed multispectral image;

[0012] Determine the type of pollution source corresponding to each pixel of the preprocessed multispectral image according to the water pollution source type probability vector corresponding to each pixel;

[0013] The location distribution of water pollution sources is determined based on the geographical location information of each pixel of the preprocessed multispectral image and the determined pollution source type.

[0014] As an improvement of the above scheme, the multispectral image is subjected to radiation correction by using a preset radiation transfer correction model, and the multispectral image is subjected to geometric correction and pixel geographical location information registration by using the image geographical location information and ground control points to obtain a preprocessed multispectral image, including:

[0015] The radiation transfer correction model is used to perform radiation correction on each pixel in the multispectral image to obtain the corrected true reflectivity ρ that can reflect the water body. w (λ) multispectral image; wherein the radiation transfer correction model is:

[0016] L org (λ) is the original radiance of a pixel in the multispectral image at wavelength λ, which is calculated in advance; L path (λ) is the radiant brightness of the sensor of the multi-spectral imaging device carried by the UAV through the atmospheric path radiation, which is obtained in advance; L refl (λ) is the radiant brightness reaching the sensor after reflection from the water body, which is obtained in advance; E0(λ) is the solar irradiance at the top of the atmosphere obtained in advance; ρ a (λ) is the atmospheric reflectivity obtained in advance; t atm (λ) is the atmospheric transmittance obtained in advance; t water (λ) is the water permeability obtained in advance;

[0017] Determine the real geographic coordinates (X, Y, Z) of the pre-selected ground control point corresponding to the shooting position of the multispectral image, and input the real geographic coordinates of the ground control point and the pixel coordinates (x, y) of each pixel of the corresponding multispectral image after radiation correction into a preset geometric correction model, and use the least squares method to iteratively solve the polynomial coefficient a of the geometric correction model ijk , using the polynomial coefficient a obtained by solving ijk , coordinate transformation is performed on each pixel of the entire multispectral image to achieve geometric correction and pixel geographic location information registration, and the preprocessed multispectral image is obtained; the geometric correction model is: Among them, P i (x, y) (i = 1, 2, ..., 8) is a polynomial function about x and y, in the form of: m, n are the polynomial degrees, which are determined in advance according to the image resolution of the multispectral imaging equipment and the complexity of the terrain at the shooting location; a ijk are the polynomial coefficients.

[0018] As an improvement of the above scheme, the multispectral feature is input into a trained water pollution source type prediction model, and the water pollution source type probability vector corresponding to each pixel of the preprocessed multispectral image is output, including:

[0019] The feature vector X of the multi-spectral feature is n) input the trained input layer of the water pollution source type prediction model of the feedforward neural network, and then perform hidden layer calculation and output layer calculation of the water pollution source type prediction model in sequence, and finally output the water pollution source type probability vector P corresponding to each pixel of the preprocessed multispectral image; wherein the water pollution source type prediction model comprises an input layer, a hidden layer and an output layer; the number of nodes in the input layer is the same as the dimension of the multispectral feature, which is set to n, and each node receives the corresponding multispectral feature value; one hidden layer is set, and the number of nodes is m; the number of nodes in the output layer is equal to the number of water pollution source types, which is set to k, and each node outputs the probability value of the corresponding water pollution source type.

[0020] As an improvement of the above solution, the method of determining the pollution source type corresponding to each pixel according to the water pollution source type probability vector corresponding to each pixel of the preprocessed multispectral image includes:

[0021] For each pixel of the preprocessed multispectral image, according to the output water pollution source type probability vector, the pollution source type with the highest probability is selected as the pollution source type corresponding to the pixel.

[0022] As an improvement of the above solution, the location distribution of water pollution sources is determined according to the geographical location information of each pixel of the preprocessed multispectral image and the determined pollution source type, including:

[0023] Get the longitude lon of each pixel of the preprocessed multispectral image i and latitude lat i The geographical location information and the pollution source type corresponding to each pixel are determined i , where i represents the i-th pixel;

[0024] The geographic distance d between two pixels i and j is calculated by the following formula ij :

[0025] r is the radius of the earth, which is a constant parameter used to reflect the geometric scale of the earth in the formula, so that the calculated geographical distance conforms to the actual spherical characteristics of the earth and ensures the accuracy of the distance calculation; lat i ,lat j are the latitude values ​​of the i-th pixel and the j-th pixel, respectively, in degrees. They participate in trigonometric calculations in the formula and together with the longitude information determine the relative positions of the two pixels on the earth's surface, thereby calculating the distance between them; lon i ,lon j The longitude values ​​corresponding to the positions of the i-th pixel and the j-th pixel, respectively, are also in degrees, and together with the latitude values ​​constitute geographic coordinates, which are used to accurately represent the position of the pixel on the earth's surface;

[0026] Core point determination: For each pixel i, calculate the number of pixels n in its ∈-neighborhood i , where ∈ is the pre-set distance threshold; let N ∈ (i) represents the ∈-neighborhood pixel set of pixel i, and the calculation formula is: If n i ≥MinPts, the pixel is marked as a core point, MinPts is the minimum point threshold; n i It refers to the number of pixels that meet the above conditions in the ∈-neighborhood of the i-th pixel, that is, the number of pixels whose distance to pixel i is less than or equal to the preset distance threshold ∈ and have the same pollution source type; t i ,t j Respectively represent the pollution source types corresponding to the i-th pixel and the j-th pixel, which are used to further screen the pixels in the ∈-neighborhood. Only when the pollution source types of the two pixels are the same will they be calculated in n i is taken into account to ensure that clustering is based on pixels of the same pollution source type, so that the final clustering result can accurately reflect the distribution of different types of water pollution sources;

[0027] Cluster expansion: Starting from any unvisited core point p, create a new cluster C and add p to C, then for each pixel q in the ∈-neighborhood of p; if q is an unvisited core point, add q to C and recursively process the ∈-neighborhood; if q is not a core point but belongs to the same pollution source type as p and is not assigned to any cluster, add q to C; repeat this process until all core points have been visited and all reachable pixels have been assigned to the corresponding clusters;

[0028] Cluster center determination: For each cluster C m , its central location The calculation adopts the weighted average method, m represents the mth cluster, C m represents the mth cluster, which is a set of pixels with the same pollution source type and are spatially related, reflecting a distribution grouping of water pollution sources in geographic space; the calculation formula is:

[0029] Represents the mth cluster C m The longitude value of the center position, in degrees, is obtained by weighted average calculation of the longitude information of all pixels in the cluster. It represents the center position of the cluster in the longitude direction in the geographic space and is used to accurately mark the approximate position of the water pollution source represented by the cluster in the east-west direction on the map of the geographic information GIS system; iis the pixel weight, which is a value determined in advance based on the relevant characteristics of the pixel's spectral reflectance in the multispectral image. It is used to reflect the different importance of different pixels to the determination of the center position when calculating the cluster center.

[0030] Distribution range determination: calculation of cluster C m The range in longitude and latitude; Cluster C m The minimum and maximum values ​​of pixel longitude, Cluster C m The minimum and maximum values ​​of the pixel latitude and distribution range can be expressed as: Respectively represent the mth cluster C m The distribution range in the longitude and latitude directions, in degrees, is obtained by calculating the difference between the maximum and minimum values ​​of the longitude and latitude of the pixels in the cluster. It is used to intuitively reflect the coverage of the water pollution source represented by the cluster in the east-west direction, which helps to accurately draw the distribution area shape and size of the pollution source on the geographic information GIS system map;

[0031] The center position of each cluster And distribution range information The data are imported into the geographic information system (GIS), and the drawing function of the geographic information system is used to draw the distribution area of ​​water pollution sources represented by each cluster using different colors or symbols according to different types of pollution sources.

[0032] Another embodiment of the present invention provides a water pollution source identification system based on multispectral images of unmanned aerial vehicles, including:

[0033] A control module is used to control a drone equipped with a multispectral imaging device to inspect a target water area to obtain a multispectral image, and at the same time use a positioning system of the drone to record image geographic location information corresponding to the multispectral image;

[0034] A preprocessing module, used to perform radiation correction on the multispectral image by using a preset radiation transfer correction model, and to perform geometric correction and pixel geographical location information registration on the multispectral image by using the image geographical location information and ground control points, so as to obtain a preprocessed multispectral image;

[0035] A feature extraction module is used to extract spectral features from the preprocessed multispectral image to obtain multispectral features;

[0036] A prediction module, used for inputting the multispectral features into a trained water pollution source type prediction model, and outputting a water pollution source type probability vector corresponding to each pixel of the preprocessed multispectral image;

[0037] A first analysis module, for determining the type of pollution source corresponding to each pixel of the preprocessed multispectral image according to the probability vector of the water pollution source type corresponding to each pixel;

[0038] The second analysis module is used to determine the location distribution of water pollution sources based on the geographical location information of each pixel of the preprocessed multispectral image and the determined pollution source type.

[0039] As an improvement of the above solution, the preprocessing module is specifically used for:

[0040] The radiation transfer correction model is used to perform radiation correction on each pixel in the multispectral image to obtain the corrected true reflectivity ρ that can reflect the water body. w (λ) multispectral image; wherein the radiation transfer correction model is:

[0041] L org (λ) is the original radiance of a pixel in the multispectral image at wavelength λ, which is calculated in advance; L path (λ) is the radiant brightness of the sensor of the multi-spectral imaging device carried by the UAV through the atmospheric path radiation, which is obtained in advance; L refl (λ) is the radiant brightness reaching the sensor after reflection from the water body, which is obtained in advance; E0(λ) is the solar irradiance at the top of the atmosphere obtained in advance; ρ a (λ) is the atmospheric reflectivity obtained in advance; t atm (λ) is the atmospheric transmittance obtained in advance; t water (λ) is the water permeability obtained in advance;

[0042] Determine the real geographic coordinates (X, Y, Z) of the pre-selected ground control point corresponding to the shooting position of the multispectral image, and input the real geographic coordinates of the ground control point and the pixel coordinates (x, y) of each pixel of the corresponding multispectral image after radiation correction into a preset geometric correction model, and use the least squares method to iteratively solve the polynomial coefficient a of the geometric correction model ijk , using the polynomial coefficient a obtained by solving ijk , coordinate transformation is performed on each pixel of the entire multispectral image to achieve geometric correction and pixel geographic location information registration, and the preprocessed multispectral image is obtained; the geometric correction model is:

[0043] Among them, P i (x, y) (i = 1, 2, ..., 8) is a polynomial function about x and y, in the form of: m, n are the polynomial degrees, which are determined in advance according to the image resolution of the multispectral imaging equipment and the complexity of the terrain at the shooting location; a ijk are the polynomial coefficients.

[0044] As an improvement of the above solution, the prediction module is specifically used for:

[0045] The feature vector X of the multi-spectral feature is n ) input the trained input layer of the water pollution source type prediction model of the feedforward neural network, and then perform hidden layer calculation and output layer calculation of the water pollution source type prediction model in sequence, and finally output the water pollution source type probability vector P corresponding to each pixel of the preprocessed multispectral image; wherein the water pollution source type prediction model comprises an input layer, a hidden layer and an output layer; the number of nodes in the input layer is the same as the dimension of the multispectral feature, which is set to n, and each node receives the corresponding multispectral feature value; one hidden layer is set, and the number of nodes is m; the number of nodes in the output layer is equal to the number of water pollution source types, which is set to k, and each node outputs the probability value of the corresponding water pollution source type.

[0046] As an improvement of the above solution, the first analysis module is specifically used for:

[0047] For each pixel of the preprocessed multispectral image, according to the output water pollution source type probability vector, the pollution source type with the highest probability is selected as the pollution source type corresponding to the pixel.

[0048] As an improvement of the above solution, the second analysis module is specifically used for:

[0049] Get the longitude lon of each pixel of the preprocessed multispectral image i and latitude lat i The geographical location information and the pollution source type corresponding to each pixel are determined i , where i represents the i-th pixel;

[0050] The geographic distance d between two pixels i and j is calculated by the following formula ij :

[0051] r is the radius of the earth, which is a constant parameter used to reflect the geometric scale of the earth in the formula, so that the calculated geographical distance conforms to the actual spherical characteristics of the earth and ensures the accuracy of the distance calculation; lat i ,lat j are the latitude values ​​of the i-th pixel and the j-th pixel, respectively, in degrees. They participate in trigonometric calculations in the formula and together with the longitude information determine the relative positions of the two pixels on the earth's surface, thereby calculating the distance between them; loni ,lon j The longitude values ​​corresponding to the positions of the i-th pixel and the j-th pixel, respectively, are also in degrees, and together with the latitude values ​​constitute geographic coordinates, which are used to accurately represent the position of the pixel on the earth's surface;

[0052] Core point determination: For each pixel i, calculate the number of pixels n in its ∈-neighborhood i , where ∈ is the pre-set distance threshold; let N ∈ (i) represents the ∈-neighborhood pixel set of pixel i, and the calculation formula is: If n i ≥MinPts, the pixel is marked as a core point, MinPts is the minimum point threshold; n i It refers to the number of pixels that meet the above conditions in the ∈-neighborhood of the i-th pixel, that is, the number of pixels whose distance to pixel i is less than or equal to the preset distance threshold ∈ and have the same pollution source type; t i ,t j Respectively represent the pollution source types corresponding to the i-th pixel and the j-th pixel, which are used to further screen the pixels in the ∈-neighborhood. Only when the pollution source types of the two pixels are the same will they be calculated in n i is taken into account to ensure that clustering is based on pixels of the same pollution source type, so that the final clustering result can accurately reflect the distribution of different types of water pollution sources;

[0053] Cluster expansion: Starting from any unvisited core point p, create a new cluster C and add p to C, then for each pixel q in the ∈-neighborhood of p; if q is an unvisited core point, add q to C and recursively process the ∈-neighborhood; if q is not a core point but belongs to the same pollution source type as p and is not assigned to any cluster, add q to C; repeat this process until all core points have been visited and all reachable pixels have been assigned to the corresponding clusters;

[0054] Cluster center determination: For each cluster C m , its central location The calculation adopts the weighted average method, m represents the mth cluster, C m represents the mth cluster, which is a set of pixels with the same pollution source type and are spatially related, reflecting a distribution grouping of water pollution sources in geographic space; the calculation formula is:

[0055] Represents the mth cluster C mThe longitude value of the center position, in degrees, is obtained by weighted average calculation of the longitude information of all pixels in the cluster. It represents the center position of the cluster in the longitude direction in the geographic space and is used to accurately mark the approximate position of the water pollution source represented by the cluster in the east-west direction on the map of the geographic information GIS system; i is the pixel weight, which is a value determined in advance based on the relevant characteristics of the pixel's spectral reflectance in the multispectral image. It is used to reflect the different importance of different pixels to the determination of the center position when calculating the cluster center.

[0056] Distribution range determination: calculation of cluster C m The range in longitude and latitude; Cluster C m The minimum and maximum values ​​of pixel longitude, Cluster C m The minimum and maximum values ​​of the pixel latitude and distribution range can be expressed as: Respectively represent the mth cluster C m The distribution range in the longitude and latitude directions, in degrees, is obtained by calculating the difference between the maximum and minimum values ​​of the longitude and latitude of the pixels in the cluster. It is used to intuitively reflect the coverage of the water pollution source represented by the cluster in the east-west direction, which helps to accurately draw the distribution area shape and size of the pollution source on the geographic information GIS system map;

[0057] The center position of each cluster And distribution range information The data are imported into the geographic information system (GIS), and the drawing function of the geographic information system is used to draw the distribution area of ​​water pollution sources represented by each cluster using different colors or symbols according to different types of pollution sources.

[0058] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0059] By using drones equipped with multispectral imaging equipment to inspect the target waters, multispectral images and corresponding geographic location information are obtained. Then, the preset radiation transmission correction model is used to perform radiation correction, geometric correction and pixel geographic location information registration in sequence with the help of image geographic location information and ground control points to pre-process the image. Then, its spectral characteristics are extracted and input into the trained water pollution source type prediction model to obtain the water pollution source type probability vector corresponding to each pixel, and then the pollution source type of each pixel is determined. Finally, the location distribution of the water pollution source is clarified based on the pixel geographic location information. Compared with the technical problems of poor timeliness and poor accuracy of traditional on-site sampling and analysis methods, this patented technical solution can achieve more comprehensive, accurate, timely and low-cost identification and location distribution determination of water pollution sources in target waters, effectively improving the efficiency and accuracy of water pollution source monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 It is a flow chart of a method for identifying water pollution sources based on multispectral images of unmanned aerial vehicles provided by an embodiment of the present invention;

[0061] Figure 2 It is a structural schematic diagram of a water pollution source identification system based on multispectral imaging of unmanned aerial vehicles provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0062] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0063] See also Figure 1 , is a flow chart of a method for identifying water pollution sources based on multispectral images of unmanned aerial vehicles provided by an embodiment of the present invention. The method for identifying water pollution sources based on multispectral images of unmanned aerial vehicles comprises the following steps S10 to S15:

[0064] S10, controlling a drone equipped with a multispectral imaging device to inspect the target waters to obtain a multispectral image, and using a positioning system of the drone to record image geographic location information corresponding to the multispectral image;

[0065] S11, performing radiation correction on the multispectral image by using a preset radiation transfer correction model, and performing geometric correction and pixel geographical location information registration on the multispectral image by using the image geographical location information and ground control points to obtain a preprocessed multispectral image;

[0066] S12, extracting spectral features from the preprocessed multispectral image to obtain multispectral features;

[0067] S13, inputting the multispectral feature into a trained water pollution source type prediction model, and outputting a water pollution source type probability vector corresponding to each pixel of the preprocessed multispectral image;

[0068] S14, determining the type of pollution source corresponding to each pixel according to the water pollution source type probability vector corresponding to each pixel of the preprocessed multispectral image;

[0069] S15, determining the location distribution of water pollution sources according to the geographical location information of each pixel of the preprocessed multispectral image and the determined pollution source type.

[0070] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0071] By using drones equipped with multispectral imaging equipment to inspect the target waters, multispectral images and corresponding geographic location information are obtained. Then, the preset radiation transmission correction model is used to perform radiation correction, geometric correction and pixel geographic location information registration in sequence with the help of image geographic location information and ground control points to pre-process the image. Then, its spectral characteristics are extracted and input into the trained water pollution source type prediction model to obtain the water pollution source type probability vector corresponding to each pixel, and then the pollution source type of each pixel is determined. Finally, the location distribution of the water pollution source is clarified based on the pixel geographic location information. Compared with the technical problems of poor timeliness and poor accuracy of traditional on-site sampling and analysis methods, this patented technical solution can achieve more comprehensive, accurate, timely and low-cost identification and location distribution determination of water pollution sources in target waters, effectively improving the efficiency and accuracy of water pollution source monitoring.

[0072] As an example, the process of drone inspection and data collection: Select a suitable drone platform, which should have stable flight performance, sufficient endurance and high-precision positioning system (such as GPS / Beidou navigation system). Equipped with multispectral imaging equipment, the multispectral imaging equipment can image the target waters in multiple specific spectral bands, such as selecting bands covering the visible light to near-infrared spectrum range, such as 4-8 different bands in the range of 400-900nm, to obtain rich spectral information of the target waters. Before the flight inspection, plan a reasonable flight route based on factors such as the scope, shape and surrounding environment of the target waters. Set the parameters such as the flight altitude, speed and shooting interval of the drone to ensure the spatial resolution and coverage integrity of the multispectral image. For example, the flight altitude can be set between 50-200 meters, the flight speed can be controlled at 5-15 meters / second, and the shooting interval is determined according to the flight speed and the required image overlap rate (generally the image overlap rate is 60%-80%) to ensure the accuracy and reliability of subsequent image processing and analysis. The drone flies along a preset route. During the flight, the multispectral imaging device shoots the target waters at set time intervals or trigger conditions to obtain multispectral image data. At the same time, the drone's positioning system records the image geographic location information (longitude, latitude, altitude) corresponding to each multispectral image in real time. This information will be stored together with the multispectral image data to provide basic data for subsequent image correction and pollution source location.

[0073] It should be noted that the feature extraction method for extracting spectral features from the preprocessed multispectral image can refer to the existing feature extraction technology, which will not be described in detail here. Among them, the multispectral features can be the shape features of the spectral reflectance curve (such as peak value, valley value, slope, etc.), spectral absorption features (such as absorption depth and absorption width of a specific band, etc.) and spectral derivative features (first-order derivative, second-order derivative, etc.).

[0074] As an improvement of the above scheme, the multispectral image is subjected to radiation correction by using a preset radiation transfer correction model, and the multispectral image is subjected to geometric correction and pixel geographical location information registration by using the image geographical location information and ground control points to obtain a preprocessed multispectral image, including:

[0075] The radiation transfer correction model is used to perform radiation correction on each pixel in the multispectral image to obtain the corrected true reflectivity ρ that can reflect the water body. w (λ) multispectral image; wherein the radiation transfer correction model is:

[0076] L org (λ) is the original radiance of a pixel in the multispectral image at wavelength λ, which is calculated in advance; Lpath (λ) is the radiant brightness of the sensor of the multi-spectral imaging device carried by the UAV through the atmospheric path radiation, which is obtained in advance; L refl (λ) is the radiant brightness reaching the sensor after reflection from the water body, which is obtained in advance; E0(λ) is the solar irradiance at the top of the atmosphere obtained in advance; ρ a (λ) is the atmospheric reflectivity obtained in advance; t atm (λ) is the atmospheric transmittance obtained in advance; t water (λ) is the water permeability obtained in advance;

[0077] Determine the real geographic coordinates (X, Y, Z) of the pre-selected ground control point corresponding to the shooting position of the multispectral image, and input the real geographic coordinates of the ground control point and the pixel coordinates (x, y) of each pixel of the corresponding multispectral image after radiation correction into a preset geometric correction model, and use the least squares method to iteratively solve the polynomial coefficient a of the geometric correction model ijk , using the polynomial coefficient a obtained by solving ijk , coordinate transformation is performed on each pixel of the entire multispectral image to achieve geometric correction and pixel geographic location information registration, and the preprocessed multispectral image is obtained; the geometric correction model is:

[0078] Among them, P i (x, y) (i = 1, 2, ..., 8) is a polynomial function about x and y, in the form of: m, n are the polynomial degrees, which are determined in advance according to the image resolution of the multispectral imaging equipment and the complexity of the terrain at the shooting location; a ijk are the polynomial coefficients.

[0079] In this embodiment, the multispectral image acquired by the drone is accurately processed to obtain a pre-processed multispectral image that can be used for subsequent precise water pollution source identification and analysis. First, in view of the radiation distortion problem in the multispectral image, a preset specific radiation transmission correction model is used. The model comprehensively considers the original radiation brightness, atmospheric path radiation, water body reflected radiation, and solar irradiance at the top of the atmosphere, atmospheric reflectivity, atmospheric transmittance, and water body transmittance. Through rigorous formula calculation, each pixel is corrected for radiation, and the true reflectivity of the water body is restored, thereby eliminating the radiation error caused by the optical properties of the atmosphere and the water body itself. Then, using the pre-selected ground control points corresponding to the shooting location of the multispectral image, combined with their precise real geographic coordinates and the pixel coordinates of each pixel in the radiation-corrected image, the preset geometric correction model is input. This geometric correction model uses a polynomial form to construct a mapping relationship between pixel coordinates and geographic coordinates, and uses the least squares method to iteratively solve the polynomial coefficients, and then transforms the pixel coordinates of the entire multispectral image according to the coefficients obtained, to achieve geometric correction and complete the pixel geographic location information registration, and finally obtain the pre-processed multispectral image, which lays an accurate data foundation for subsequent spectral feature extraction and pollution source identification. From the above analysis, it can be seen that this embodiment constructs a comprehensive and sophisticated radiation transmission correction model, comprehensively considers a variety of radiation-related factors, and can more accurately restore the true reflectivity of the water body, which greatly improves the accuracy of radiation correction compared to traditional simple correction methods. At the same time, in the geometric correction, a polynomial geometric correction model based on ground control points and iterative least squares method is used, and the degree of the polynomial is determined by fully combining the image resolution and terrain complexity, which effectively improves the accuracy of geometric correction and pixel geographic location information registration. The technical effect is that the quality of multispectral images is significantly improved, making the subsequent image-based spectral feature extraction more reliable, thereby improving the accuracy of water pollution source identification. The precise pixel geographic location information registration provides a strong guarantee for accurately determining the location distribution of pollution sources, which overall enhances the effectiveness and practicality of water pollution source identification technology based on drone multispectral images.

[0080] For ease of understanding, the following example is given here: For example, in a water pollution source monitoring scenario of a large lake, the drone flies according to a preset route and shoots multispectral images. In the radiation correction stage, for a certain pixel in the image, its original radiation brightness at a specific wavelength is known. Combined with the radiation brightness of the atmospheric path radiation reaching the sensor calculated in advance by the atmospheric optical model, the radiation brightness reaching the sensor after reflection from the water body, and the solar irradiance at the top of the atmosphere determined according to the position of the sun and the atmospheric conditions, atmospheric reflectivity, atmospheric transmittance, and water body transmittance, the radiation transmission correction model is used to accurately calculate the corrected reflectivity to achieve radiation correction of the entire image. Afterwards, ground control points such as bridge endpoints and corner points of landmark buildings on the shore are selected around the lake, and their real geographic coordinates are measured by high-precision GPS. The geographic coordinates of these ground control points and the coordinates of each pixel in the radiation-corrected image are input into the geometric correction model, and the polynomial degree is determined according to the image resolution (such as higher resolution corresponds to higher polynomial degree settings) and the complexity of the terrain around the lake. The polynomial coefficients are iteratively solved by the least squares method to transform the coordinates of the image pixels. After this series of processing, the obtained pre-processed multispectral image can clearly and accurately reflect the spectral characteristics and geographical location information of the lake water body. Whether it is identifying local spectral anomaly areas caused by industrial wastewater discharge or locating the distribution range of agricultural non-point source pollution around the lake shore, it can provide high-precision data support, thus providing a scientific basis for lake water environment protection and pollution control.

[0081] As an improvement of the above scheme, the multispectral feature is input into a trained water pollution source type prediction model, and the water pollution source type probability vector corresponding to each pixel of the preprocessed multispectral image is output, including:

[0082] The feature vector X of the multi-spectral feature is n ) input the trained input layer of the water pollution source type prediction model of the feedforward neural network, and then perform hidden layer calculation and output layer calculation of the water pollution source type prediction model in sequence, and finally output the water pollution source type probability vector P corresponding to each pixel of the preprocessed multispectral image; wherein the water pollution source type prediction model comprises an input layer, a hidden layer and an output layer; the number of nodes in the input layer is the same as the dimension of the multispectral feature, which is set to n, and each node receives the corresponding multispectral feature value; one hidden layer is set, and the number of nodes is m; the number of nodes in the output layer is equal to the number of water pollution source types, which is set to k, and each node outputs the probability value of the corresponding water pollution source type.

[0083] In this embodiment, the water pollution source type prediction model constructed by the feedforward neural network is used to analyze the multispectral features extracted by the preliminary processing, and then predict the water pollution source type probability vector corresponding to each pixel. Specifically, the multispectral features are first organized into the form of feature vectors, which are input into the input layer of the water pollution source type prediction model, where the number of nodes in the input layer is set to n according to the dimension of the multispectral features, and each node is responsible for receiving the corresponding multispectral feature value to ensure the complete input of the data. Subsequently, the data will be passed to the hidden layer in turn, and the features will be further abstracted and integrated through the nodes (the number is m) set inside the hidden layer and the corresponding calculation rules, and the deep-level association between the multispectral features and the water pollution source types will be excavated. Finally, the data processed by the hidden layer flows to the output layer again, and the number of nodes in the output layer is set to k according to the number of water pollution source types, and each node outputs the probability value of the corresponding water pollution source type, so as to obtain the water pollution source type probability vector corresponding to each pixel, and provide a quantitative basis for the subsequent accurate judgment of the pollution source type to which the pixel belongs. Among them, the specific training process of the feedforward neural network can refer to the prior art, which will not be repeated here.

[0084] As an improvement of the above solution, the method of determining the pollution source type corresponding to each pixel according to the water pollution source type probability vector corresponding to each pixel of the preprocessed multispectral image includes:

[0085] For each pixel of the preprocessed multispectral image, according to the output water pollution source type probability vector, the pollution source type with the highest probability is selected as the pollution source type corresponding to the pixel.

[0086] In this embodiment, a clear determination of the pollution source type is made based on the probability vector of the water pollution source type corresponding to each pixel of the preprocessed multispectral image output in the previous step. Since the previous model outputs the probability vector of different water pollution source types corresponding to each pixel, it reflects the possibility that the pixel belongs to each type of pollution source, and this step uses the probability size as a quantitative indicator to determine the unique pollution source type. For each pixel, traverse the various probability values ​​in the corresponding water pollution source type probability vector, and select the pollution source type corresponding to the value with the largest probability, and identify it as the pollution source type actually corresponding to the pixel. In such a concise and direct way, the conversion from probability description to specific pollution source type determination is realized, providing a basic classification basis for the subsequent accurate presentation of the distribution of water pollution sources in the target waters.

[0087] For ease of understanding, the following example is given here: For example, when identifying water pollution sources in a lake affected by various human activities in the surrounding area, after a series of processing in the early stage, for a certain pixel in the lake image, after the operation of the water pollution source type prediction model, the corresponding water pollution source type probability vector is output as (0.1, 0.3, 0.6), where it is assumed that the probabilities of the three pollution source types of industrial sewage, agricultural non-point source sewage, and domestic sewage are respectively corresponding. According to the method of the technical steps, by comparing the sizes of these three probability values, it can be clearly seen that 0.6 is the largest probability value, and the pollution source type corresponding to it is domestic sewage, so it can be determined that the pollution source type corresponding to this pixel is domestic sewage. This operation is performed on each pixel in the lake image, and finally it can be determined which type of pollution source causes pollution in each area of ​​the entire lake. For example, in a certain area of ​​the lake near the residential area, most pixels are determined to correspond to the pollution source type of domestic sewage; while the pixels near the farmland area mostly correspond to agricultural non-point source sewage, and some pixels near the factory correspond to industrial sewage, which intuitively presents the distribution of lake water pollution sources.

[0088] As an improvement of the above solution, the location distribution of water pollution sources is determined according to the geographical location information of each pixel of the preprocessed multispectral image and the determined pollution source type, including:

[0089] Get the longitude lon of each pixel of the preprocessed multispectral image i and latitude lat i The geographical location information and the pollution source type corresponding to each pixel are determined i , where i represents the i-th pixel;

[0090] The geographic distance d between two pixels i and j is calculated by the following formula ij :

[0091] r is the radius of the earth, which is a constant parameter used to reflect the geometric scale of the earth in the formula, so that the calculated geographical distance conforms to the actual spherical characteristics of the earth and ensures the accuracy of the distance calculation; lat i ,lat j are the latitude values ​​of the i-th pixel and the j-th pixel respectively, in degrees, and are used in the formula to participate in trigonometric calculations. Together with the longitude information, they determine the relative positions of the two pixels on the earth's surface, thereby calculating the distance between them; lon i ,lon j The longitude values ​​corresponding to the positions of the i-th pixel and the j-th pixel, respectively, are also in degrees, and together with the latitude values ​​constitute geographic coordinates, which are used to accurately represent the position of the pixel on the earth's surface;

[0092] Core point determination: For each pixel i, calculate the number of pixels n in its ∈-neighborhood i , where ∈ is the pre-set distance threshold; let N ∈ (i) represents the ∈-neighborhood pixel set of pixel i, and the calculation formula is: If n i ≥MinPts, the pixel is marked as a core point, MinPts is the minimum point threshold; n i It refers to the number of pixels that meet the above conditions in the ∈-neighborhood of the i-th pixel, that is, the number of pixels whose distance to pixel i is less than or equal to the preset distance threshold ∈ and have the same pollution source type; t i ,t j Respectively represent the pollution source types corresponding to the i-th pixel and the j-th pixel, which are used to further screen the pixels in the ∈-neighborhood. Only when the pollution source types of the two pixels are the same will they be calculated in n i is taken into account to ensure that clustering is based on pixels of the same pollution source type, so that the final clustering result can accurately reflect the distribution of different types of water pollution sources;

[0093] Cluster expansion: Starting from any unvisited core point p, create a new cluster C and add p to C, then for each pixel q in the ∈-neighborhood of p; if q is an unvisited core point, add q to C and recursively process the ∈-neighborhood; if q is not a core point but belongs to the same pollution source type as p and is not assigned to any cluster, add q to C; repeat this process until all core points have been visited and all reachable pixels have been assigned to the corresponding clusters;

[0094] Cluster center determination: For each cluster C m , its central location The calculation adopts the weighted average method, m represents the mth cluster, C m represents the mth cluster, which is a set of pixels with the same pollution source type and are spatially related, reflecting a distribution grouping of water pollution sources in geographic space; the calculation formula is:

[0095] Represents the mth cluster C m The longitude value of the center position, in degrees, is obtained by weighted average calculation of the longitude information of all pixels in the cluster. It represents the center position of the cluster in the longitude direction in the geographic space and is used to accurately mark the approximate position of the water pollution source represented by the cluster in the east-west direction on the map of the geographic information GIS system; iis the pixel weight, which is a value determined in advance based on the relevant characteristics of the pixel's spectral reflectance in the multispectral image. It is used to reflect the different importance of different pixels to the determination of the center position when calculating the cluster center.

[0096] Distribution range determination: calculation of cluster C m The range in longitude and latitude; Cluster C m The minimum and maximum values ​​of pixel longitude, Cluster C m The minimum and maximum values ​​of the pixel latitude and distribution range can be expressed as: Respectively represent the mth cluster C m The distribution range in the longitude and latitude directions, in degrees, is obtained by calculating the difference between the maximum and minimum values ​​of the longitude and latitude of the pixels in the cluster. It is used to intuitively reflect the coverage of the water pollution source represented by the cluster in the east-west direction, which helps to accurately draw the distribution area shape and size of the pollution source on the geographic information GIS system map;

[0097] The center position of each cluster And distribution range information The data are imported into the geographic information system (GIS), and the drawing function of the geographic information system is used to draw the distribution area of ​​water pollution sources represented by each cluster using different colors or symbols according to different types of pollution sources.

[0098] In this embodiment, based on the geographical location information of each pixel in the pre-processed multispectral image and the determined pollution source type, the location distribution of water pollution sources in the target waters is accurately presented. First, the geographical location information of each pixel including longitude and latitude and the corresponding pollution source type data are obtained, and then a special geographical distance calculation formula is used to calculate the geographical distance between any two pixels based on the radius of the earth and the longitude and latitude information of the pixel, so as to measure the spatial relationship between pixels. Then, through the core point determination step, according to the preset distance threshold and the minimum point number threshold, the pixels that meet the conditions (the distance is within the threshold and the pollution source type is the same) are screened out as core points. After that, a cluster extension operation is carried out, and clusters are created starting from the unvisited core points, and the surrounding pixels that meet the requirements are included in the cluster according to specific rules, until all core points and reachable pixels are reasonably assigned to the corresponding clusters. Then, the center position of each cluster is calculated by weighted average, and the pixel weight (related to the spectral reflectance-related characteristics) is comprehensively considered to more scientifically determine its center in the longitude direction, and at the same time determine the distribution range of the cluster in the longitude and latitude directions. Finally, the center position and distribution range information of each cluster are imported into the geographic information GIS system, and its drawing function is used to draw the corresponding distribution area with different colors or symbols according to different pollution source types, so as to intuitively show the location distribution of water pollution sources. Specifically, compared with the prior art, this embodiment has the following innovations:

[0099] Distance calculation takes into account the spherical characteristics of the earth: When calculating the geographic distance between pixels, a specific formula that combines the earth's radius and longitude and latitude information is used to fully consider the spherical shape of the earth. Compared with simple plane distance calculation methods, it is more in line with actual geographical conditions, making distance measurement more accurate, and laying an accurate foundation for subsequent distance-based clustering analysis.

[0100] Clustering strategy based on pollution source type: In the process of core point determination and cluster expansion, the pollution source type of pixels is strictly screened and clustered to ensure that each cluster represents the same type of water pollution source. This targeted clustering method can clearly distinguish the spatial distribution of different pollution sources, avoid different types of pollution sources from being mixed in the same cluster, and improve the accuracy of location distribution determination.

[0101] Weighted average to determine cluster center: The cluster center position is calculated using weights determined by features related to pixel spectral reflectance. It no longer relies solely on the geographical location of the pixel, but instead comprehensively considers the impact of image features on the determination of the cluster center, making the cluster center more representative and more accurately reflecting the core location of the pollution source represented by the cluster.

[0102] Through the above series of sophisticated calculation and processing steps, the specific location distribution range and center position of different types of water pollution sources in the target waters can be accurately outlined. Whether it is a point pollution source (such as a sewage outlet) or a surface pollution source (such as an agricultural non-point source pollution area), it can be clearly presented in the geographic information system, providing strong support for the targeted positioning of pollution control. At the same time, with the help of the mapping function of the GIS system, the distribution areas of different types of pollution sources are displayed in different colors or symbols, making the complex distribution of water pollution sources clear at a glance, making it convenient for environmental protection personnel, relevant decision-making departments, etc. to quickly understand and grasp the pollution status of waters, and help to efficiently formulate governance strategies and monitoring plans.

[0103] For ease of understanding, the following example is given here: For example, when monitoring water pollution sources in a small lake located in the urban-rural junction and surrounded by factories, farmlands and residential areas, the geographical location information of each pixel of the pre-processed multispectral image and the determined pollution source type (assuming that it is divided into industrial sewage, agricultural non-point source sewage, and domestic sewage) have been obtained through the previous process. First, calculate the geographical distance between pixels. For example, two pixels in the lake, one is located on the north bank of the lake, with longitude and latitude (116°E, 30°N), and the other is located in the center of the lake, with longitude and latitude (116.1°E, 30°N). Use the given distance calculation formula (substituting parameters such as the earth radius) to calculate the actual geographical distance between them. Then determine the core point, set the distance threshold to 500 meters, and the minimum point threshold to 5. For a certain pixel, if the number of pixels within 500 meters and the same industrial sewage pollution source type reaches 5 or more, it is marked as a core point. Then, cluster expansion starts from a core point of an unvisited industrial sewage type. If there are other unvisited industrial sewage core points in its neighborhood or pixels that are not core points but are also industrial sewage types and have not been assigned clusters, they will be included in the same cluster. When clustering is completed, for example, a cluster formed for industrial sewage contains multiple pixels. According to the spectral reflectance of the pixel in the multispectral image, each pixel is assigned a corresponding weight (such as a slightly higher weight for pixels with high reflectance), and the center position of the cluster in the longitude direction is calculated by weighted average, and then its distribution range in the longitude and latitude directions is determined. Finally, the center locations and distribution range information of these industrial wastewater clusters, agricultural non-point source wastewater clusters, domestic wastewater clusters, etc. are imported into the GIS system. On the map, red is used to represent the industrial wastewater distribution area, green is used to represent the agricultural non-point source wastewater area, and blue is used to represent the domestic wastewater area. In this way, we can very intuitively see the distribution of different pollution sources around the lake. For example, it is found that industrial wastewater is mainly concentrated on the west side of the lake near the factory, and agricultural non-point source wastewater is widely distributed around the farmland on the south bank of the lake. This provides accurate spatial pointing information for the subsequent adoption of corresponding governance measures, such as factory pollution supervision and prevention and control of farmland non-point source pollution.

[0104] See also Figure 2 , is a schematic diagram of a water pollution source identification system based on multispectral imaging of unmanned aerial vehicles provided by an embodiment of the present invention. The water pollution source identification system based on multispectral imaging of unmanned aerial vehicles includes:

[0105] The control module 10 is used to control the drone equipped with a multispectral imaging device to inspect the target waters to obtain a multispectral image, and at the same time use the positioning system of the drone to record the image geographical location information corresponding to the multispectral image;

[0106] The preprocessing module 11 is used to perform radiation correction on the multispectral image by using a preset radiation transfer correction model, and to perform geometric correction and pixel geographical location information registration on the multispectral image by using the image geographical location information and ground control points to obtain a preprocessed multispectral image;

[0107] A feature extraction module 12 is used to extract spectral features from the preprocessed multispectral image to obtain multispectral features;

[0108] A prediction module 13, used for inputting the multispectral features into a trained water pollution source type prediction model, and outputting a water pollution source type probability vector corresponding to each pixel of the preprocessed multispectral image;

[0109] A first analysis module 14 is used to determine the type of pollution source corresponding to each pixel of the preprocessed multispectral image according to the probability vector of the water pollution source type corresponding to each pixel;

[0110] The second analysis module 15 is used to determine the location distribution of water pollution sources according to the geographical location information of each pixel of the pre-processed multispectral image and the determined pollution source type.

[0111] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0112] By using drones equipped with multispectral imaging equipment to inspect the target waters, multispectral images and corresponding geographic location information are obtained. Then, the preset radiation transmission correction model is used to perform radiation correction, geometric correction and pixel geographic location information registration in sequence with the help of image geographic location information and ground control points to pre-process the image. Then, its spectral characteristics are extracted and input into the trained water pollution source type prediction model to obtain the water pollution source type probability vector corresponding to each pixel, and then the pollution source type of each pixel is determined. Finally, the location distribution of the water pollution source is clarified based on the pixel geographic location information. Compared with the technical problems of poor timeliness and poor accuracy of traditional on-site sampling and analysis methods, this patented technical solution can achieve more comprehensive, accurate, timely and low-cost identification and location distribution determination of water pollution sources in target waters, effectively improving the efficiency and accuracy of water pollution source monitoring.

[0113] As an improvement of the above solution, the preprocessing module is specifically used for:

[0114] The radiation transfer correction model is used to perform radiation correction on each pixel in the multispectral image to obtain the corrected true reflectivity ρ that can reflect the water body. w (λ) multispectral image; wherein the radiation transfer correction model is:

[0115] L org (λ) is the original radiance of a pixel in the multispectral image at wavelength λ, which is calculated in advance; L path (λ) is the radiant brightness of the sensor of the multi-spectral imaging device carried by the UAV through the atmospheric path radiation, which is obtained in advance; L refl (λ) is the radiant brightness reaching the sensor after reflection from the water body, which is obtained in advance; E0(λ) is the solar irradiance at the top of the atmosphere obtained in advance; ρ a (λ) is the atmospheric reflectivity obtained in advance; t atm (λ) is the atmospheric transmittance obtained in advance; t water (λ) is the water permeability obtained in advance;

[0116] Determine the real geographic coordinates (X, Y, Z) of the pre-selected ground control point corresponding to the shooting position of the multispectral image, and input the real geographic coordinates of the ground control point and the pixel coordinates (x, y) of each pixel of the corresponding multispectral image after radiation correction into a preset geometric correction model, and use the least squares method to iteratively solve the polynomial coefficient a of the geometric correction model ijk , using the polynomial coefficient a obtained by solving ijk, coordinate transformation is performed on each pixel of the entire multispectral image to achieve geometric correction and pixel geographic location information registration, and the preprocessed multispectral image is obtained; the geometric correction model is:

[0117] Among them, P i (x, y) (i = 1, 2, ..., 8) is a polynomial function about x and y, in the form of: m, n are the polynomial degrees, which are determined in advance according to the image resolution of the multispectral imaging equipment and the complexity of the terrain at the shooting location; a ijk are the polynomial coefficients.

[0118] As an improvement of the above solution, the prediction module is specifically used for:

[0119] The feature vector X of the multi-spectral feature is n ) input the trained input layer of the water pollution source type prediction model of the feedforward neural network, and then perform hidden layer calculation and output layer calculation of the water pollution source type prediction model in sequence, and finally output the water pollution source type probability vector P corresponding to each pixel of the preprocessed multispectral image; wherein the water pollution source type prediction model comprises an input layer, a hidden layer and an output layer; the number of nodes in the input layer is the same as the dimension of the multispectral feature, which is set to n, and each node receives the corresponding multispectral feature value; one hidden layer is set, and the number of nodes is m; the number of nodes in the output layer is equal to the number of water pollution source types, which is set to k, and each node outputs the probability value of the corresponding water pollution source type.

[0120] As an improvement of the above solution, the first analysis module is specifically used for:

[0121] For each pixel of the preprocessed multispectral image, according to the output water pollution source type probability vector, the pollution source type with the highest probability is selected as the pollution source type corresponding to the pixel.

[0122] As an improvement of the above solution, the second analysis module is specifically used for:

[0123] Get the longitude lon of each pixel of the preprocessed multispectral image i and latitude lat i The geographical location information and the pollution source type corresponding to each pixel are determined i , where i represents the i-th pixel;

[0124] The geographic distance d between two pixels i and j is calculated by the following formula ij :

[0125] r is the radius of the earth, which is a constant parameter used to reflect the geometric scale of the earth in the formula, so that the calculated geographical distance conforms to the actual spherical characteristics of the earth and ensures the accuracy of the distance calculation; lat i ,lat j are the latitude values ​​of the i-th pixel and the j-th pixel, respectively, in degrees. They participate in trigonometric calculations in the formula and together with the longitude information determine the relative positions of the two pixels on the earth's surface, thereby calculating the distance between them; lon i ,lon j The longitude values ​​corresponding to the positions of the i-th pixel and the j-th pixel, respectively, are also in degrees, and together with the latitude values ​​constitute geographic coordinates, which are used to accurately represent the position of the pixel on the earth's surface;

[0126] Core point determination: For each pixel i, calculate the number of pixels n in its ∈-neighborhood i , where ∈ is the pre-set distance threshold; let N ∈ (i) represents the ∈-neighborhood pixel set of pixel i, and the calculation formula is: If n i ≥MinPts, the pixel is marked as a core point, MinPts is the minimum point threshold; n i It refers to the number of pixels that meet the above conditions in the ∈-neighborhood of the i-th pixel, that is, the number of pixels whose distance to pixel i is less than or equal to the preset distance threshold ∈ and have the same pollution source type; t i ,t j Respectively represent the pollution source types corresponding to the i-th pixel and the j-th pixel, which are used to further screen the pixels in the ∈-neighborhood. Only when the pollution source types of the two pixels are the same will they be calculated in n i is taken into account to ensure that clustering is based on pixels of the same pollution source type, so that the final clustering result can accurately reflect the distribution of different types of water pollution sources;

[0127] Cluster expansion: Starting from any unvisited core point p, create a new cluster C and add p to C, then for each pixel q in the ∈-neighborhood of p; if q is an unvisited core point, add q to C and recursively process the ∈-neighborhood; if q is not a core point but belongs to the same pollution source type as p and is not assigned to any cluster, add q to C; repeat this process until all core points have been visited and all reachable pixels have been assigned to the corresponding clusters;

[0128] Cluster center determination: For each cluster C m , its central location The calculation adopts the weighted average method, m represents the mth cluster, C mrepresents the mth cluster, which is a set of pixels with the same pollution source type and are spatially related, reflecting a distribution grouping of water pollution sources in geographic space; the calculation formula is:

[0129] Represents the mth cluster C m The longitude value of the center position, in degrees, is obtained by weighted average calculation of the longitude information of all pixels in the cluster. It represents the center position of the cluster in the longitude direction in the geographic space and is used to accurately mark the approximate position of the water pollution source represented by the cluster in the east-west direction on the map of the geographic information GIS system; i is the pixel weight, which is a value determined in advance based on the relevant characteristics of the pixel's spectral reflectance in the multispectral image. It is used to reflect the different importance of different pixels to the determination of the center position when calculating the cluster center.

[0130] Distribution range determination: calculation of cluster C m The range in longitude and latitude; Cluster C m The minimum and maximum values ​​of pixel longitude, Cluster C m The minimum and maximum values ​​of the pixel latitude and distribution range can be expressed as: Respectively represent the mth cluster C m The distribution range in the longitude and latitude directions, in degrees, is obtained by calculating the difference between the maximum and minimum values ​​of the longitude and latitude of the pixels in the cluster. It is used to intuitively reflect the coverage of the water pollution source represented by the cluster in the east-west direction, which helps to accurately draw the distribution area shape and size of the pollution source on the geographic information GIS system map;

[0131] The center position of each cluster And distribution range information The data are imported into the geographic information system (GIS), and the drawing function of the geographic information system is used to draw the distribution area of ​​water pollution sources represented by each cluster using different colors or symbols according to different types of pollution sources.

[0132] It should be noted that the system embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. In addition, in the drawings of the device embodiments provided by the present invention, the connection relationship between the modules indicates that there is a communication connection between them, which may be specifically implemented as one or more communication buses or signal lines. A person of ordinary skill in the art may understand and implement it without creative work.

[0133] The above is a preferred embodiment of the present invention. It should be pointed out that a person skilled in the art can make several improvements and modifications without departing from the principle of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for identifying water pollution sources based on multispectral images of unmanned aerial vehicles, characterized in that: The following steps are involved: Controlling a drone equipped with a multispectral imaging device to inspect the target waters to obtain a multispectral image, and using a positioning system of the drone to record image geographic location information corresponding to the multispectral image; The multispectral image is subjected to radiation correction by using a preset radiation transfer correction model, and the multispectral image is subjected to geometric correction and pixel geographical location information registration by using the image geographical location information and ground control points to obtain a preprocessed multispectral image; Extracting spectral features from the preprocessed multispectral image to obtain multispectral features; Input the multispectral features into a trained water pollution source type prediction model, and output a water pollution source type probability vector corresponding to each pixel of the preprocessed multispectral image; Determine the type of pollution source corresponding to each pixel of the preprocessed multispectral image according to the water pollution source type probability vector corresponding to each pixel; The location distribution of water pollution sources is determined based on the geographical location information of each pixel of the preprocessed multispectral image and the determined pollution source type.

2. The method for identifying water pollution sources based on multispectral images of unmanned aerial vehicles according to claim 1, characterized in that: The multispectral image is subjected to radiation correction by using a preset radiation transfer correction model, and the multispectral image is subjected to geometric correction and pixel geographical location information registration by using the image geographical location information and ground control points to obtain a preprocessed multispectral image, including: The radiation transfer correction model is used to perform radiation correction on each pixel in the multispectral image to obtain the corrected true reflectivity ρ that can reflect the water body. w (λ) multispectral image; wherein the radiation transfer correction model is: L org (λ) is the original radiance of a pixel in the multispectral image at wavelength λ, which is calculated in advance; L path (λ) is the radiant brightness of the sensor of the multi-spectral imaging device carried by the UAV through the atmospheric path radiation, which is obtained in advance; L refl (λ) is the radiant brightness reaching the sensor after reflection from the water body, which is obtained in advance; E0(λ) is the solar irradiance at the top of the atmosphere obtained in advance; ρ a (λ) is the atmospheric reflectivity obtained in advance; t atm (λ) is the atmospheric transmittance obtained in advance; t water (λ) is the water permeability obtained in advance; Determine the real geographic coordinates (X, Y, Z) of the pre-selected ground control point corresponding to the shooting position of the multispectral image, and input the real geographic coordinates of the ground control point and the pixel coordinates (x, y) of each pixel of the corresponding multispectral image after radiation correction into a preset geometric correction model, and use the least squares method to iteratively solve the polynomial coefficient a of the geometric correction model ijk , using the polynomial coefficient a obtained by solving ijk , coordinate transformation is performed on each pixel of the entire multispectral image to achieve geometric correction and pixel geographic location information registration, and the preprocessed multispectral image is obtained; the geometric correction model is: Among them, P i (x, y) (i = 1, 2, ..., 8) is a polynomial function about x and y, in the form of: m, n are the polynomial degrees, which are determined in advance according to the image resolution of the multispectral imaging equipment and the complexity of the terrain at the shooting location; a ijk are the polynomial coefficients.

3. The water pollution source identification method based on multispectral imaging of unmanned aerial vehicles as claimed in claim 2 is characterized in that: The step of inputting the multispectral feature into a trained water pollution source type prediction model and outputting a water pollution source type probability vector corresponding to each pixel of the preprocessed multispectral image comprises: The feature vector X of the multi-spectral feature is n ) input the trained input layer of the water pollution source type prediction model of the feedforward neural network, and then perform hidden layer calculation and output layer calculation of the water pollution source type prediction model in sequence, and finally output the water pollution source type probability vector P corresponding to each pixel of the preprocessed multispectral image; wherein the water pollution source type prediction model comprises an input layer, a hidden layer and an output layer; the number of nodes in the input layer is the same as the dimension of the multispectral feature, which is set to n, and each node receives the corresponding multispectral feature value; one hidden layer is set, and the number of nodes is m; the number of nodes in the output layer is equal to the number of water pollution source types, which is set to k, and each node outputs the probability value of the corresponding water pollution source type.

4. The method for identifying water pollution sources based on multispectral images of unmanned aerial vehicles as claimed in claim 3, characterized in that: The determining the pollution source type corresponding to each pixel according to the water pollution source type probability vector corresponding to each pixel of the preprocessed multispectral image includes: For each pixel of the preprocessed multispectral image, according to the output water pollution source type probability vector, the pollution source type with the highest probability is selected as the pollution source type corresponding to the pixel.

5. The method for identifying water pollution sources based on multispectral images of unmanned aerial vehicles as claimed in claim 4, characterized in that: Determining the location distribution of water pollution sources according to the geographical location information of each pixel of the preprocessed multispectral image and the determined pollution source type includes: Get the longitude lon of each pixel of the preprocessed multispectral image i and latitude lat i The geographical location information and the pollution source type corresponding to each pixel are determined i , where i represents the i-th pixel; The geographic distance d between two pixels i and j is calculated by the following formula ij : r is the radius of the earth, which is a constant parameter used to reflect the geometric scale of the earth in the formula, so that the calculated geographical distance conforms to the actual spherical characteristics of the earth and ensures the accuracy of the distance calculation; lat i ,lat j are the latitude values ​​of the i-th pixel and the j-th pixel, respectively, in degrees. They participate in trigonometric calculations in the formula and together with the longitude information determine the relative positions of the two pixels on the earth's surface, thereby calculating the distance between them; lon i ,lon j The longitude values ​​corresponding to the positions of the i-th pixel and the j-th pixel, respectively, are also in degrees, and together with the latitude values ​​constitute geographic coordinates, which are used to accurately represent the position of the pixel on the earth's surface; Core point determination: For each pixel i, calculate the number of pixels n in its ∈-neighborhood i , where ∈ is the pre-set distance threshold; let N ∈ (i) represents the ∈-neighborhood pixel set of pixel i, and the calculation formula is: If n i ≥MinPts, the pixel is marked as a core point, MinPts is the minimum point threshold; n i It refers to the number of pixels that meet the above conditions in the ∈-neighborhood of the i-th pixel, that is, the number of pixels whose distance to pixel i is less than or equal to the preset distance threshold ∈ and have the same pollution source type; t i ,t j Respectively represent the pollution source types corresponding to the i-th pixel and the j-th pixel, which are used to further screen the pixels in the ∈-neighborhood. Only when the pollution source types of the two pixels are the same will they be calculated in n i is taken into account to ensure that clustering is based on pixels of the same pollution source type, so that the final clustering result can accurately reflect the distribution of different types of water pollution sources; Cluster expansion: Starting from any unvisited core point p, create a new cluster C and add p to C, then for each pixel q in the ∈-neighborhood of p; if q is an unvisited core point, add q to C and recursively process the ∈-neighborhood; if q is not a core point but belongs to the same pollution source type as p and is not assigned to any cluster, add q to C; repeat this process until all core points have been visited and all reachable pixels have been assigned to the corresponding clusters; Cluster center determination: For each cluster C m , its central location The calculation adopts the weighted average method, m represents the mth cluster, C m represents the mth cluster, which is a set of pixels with the same pollution source type and are spatially related, reflecting a distribution grouping of water pollution sources in geographic space; the calculation formula is: Represents the mth cluster C m The longitude value of the center position, in degrees, is obtained by weighted average calculation of the longitude information of all pixels in the cluster. It represents the center position of the cluster in the longitude direction in the geographic space and is used to accurately mark the approximate position of the water pollution source represented by the cluster in the east-west direction on the map of the geographic information GIS system; i is the pixel weight, which is a value determined in advance based on the relevant characteristics of the pixel's spectral reflectance in the multispectral image. It is used to reflect the different importance of different pixels to the determination of the center position when calculating the cluster center. Distribution range determination: calculation of cluster C m The range in longitude and latitude; Cluster C m The minimum and maximum values ​​of pixel longitude, Cluster C m The minimum and maximum values ​​of the pixel latitude and distribution range can be expressed as: Respectively represent the mth cluster C m The distribution range in the longitude and latitude directions, in degrees, is obtained by calculating the difference between the maximum and minimum values ​​of the longitude and latitude of the pixels in the cluster. It is used to intuitively reflect the coverage of the water pollution source represented by the cluster in the east-west direction, which helps to accurately draw the distribution area shape and size of the pollution source on the geographic information GIS system map; The center position of each cluster And distribution range information The data are imported into the geographic information system (GIS), and the drawing function of the geographic information system is used to draw the distribution area of ​​water pollution sources represented by each cluster using different colors or symbols according to different types of pollution sources.

6. A water pollution source identification system based on multispectral imaging of unmanned aerial vehicles, characterized in that: include: A control module is used to control a drone equipped with a multispectral imaging device to inspect a target water area to obtain a multispectral image, and at the same time use a positioning system of the drone to record image geographic location information corresponding to the multispectral image; A preprocessing module, used to perform radiation correction on the multispectral image by using a preset radiation transfer correction model, and to perform geometric correction and pixel geographical location information registration on the multispectral image by using the image geographical location information and ground control points, so as to obtain a preprocessed multispectral image; A feature extraction module is used to extract spectral features from the preprocessed multispectral image to obtain multispectral features; A prediction module, used for inputting the multispectral features into a trained water pollution source type prediction model, and outputting a water pollution source type probability vector corresponding to each pixel of the preprocessed multispectral image; A first analysis module, for determining the type of pollution source corresponding to each pixel of the preprocessed multispectral image according to the probability vector of the water pollution source type corresponding to each pixel; The second analysis module is used to determine the location distribution of water pollution sources based on the geographical location information of each pixel of the preprocessed multispectral image and the determined pollution source type.

7. The water pollution source identification system based on multispectral imaging of unmanned aerial vehicles as claimed in claim 6 is characterized in that: The preprocessing module is specifically used for: The radiation transfer correction model is used to perform radiation correction on each pixel in the multispectral image to obtain the corrected true reflectivity ρ that can reflect the water body. w (λ) multispectral image; wherein the radiation transfer correction model is: L org (λ) is the original radiance of a pixel in the multispectral image at wavelength λ, which is calculated in advance; L path (λ) is the radiant brightness of the sensor of the multi-spectral imaging device carried by the UAV through the atmospheric path radiation, which is obtained in advance; L refl (λ) is the radiant brightness reaching the sensor after reflection from the water body, which is obtained in advance; E0(λ) is the solar irradiance at the top of the atmosphere obtained in advance; ρ a (λ) is the atmospheric reflectivity obtained in advance; t atm (λ) is the atmospheric transmittance obtained in advance; t water (λ) is the water permeability obtained in advance; Determine the real geographic coordinates (X, Y, Z) of the pre-selected ground control point corresponding to the shooting position of the multispectral image, and input the real geographic coordinates of the ground control point and the pixel coordinates (x, y) of each pixel of the corresponding multispectral image after radiation correction into a preset geometric correction model, and use the least squares method to iteratively solve the polynomial coefficient a of the geometric correction model ijk , using the polynomial coefficient a obtained by solving ijk , coordinate transformation is performed on each pixel of the entire multispectral image to achieve geometric correction and pixel geographic location information registration, and the preprocessed multispectral image is obtained; the geometric correction model is: Among them, P i (x, y) (i = 1, 2, ..., 8) is a polynomial function about x and y, in the form of: m, n are the polynomial degrees, which are determined in advance according to the image resolution of the multispectral imaging equipment and the complexity of the terrain at the shooting location; a ijk are the polynomial coefficients.

8. The water pollution source identification system based on multispectral imaging of unmanned aerial vehicles as claimed in claim 7 is characterized in that: The prediction module is specifically used for: The feature vector X of the multi-spectral feature is n ) input the trained input layer of the water pollution source type prediction model of the feedforward neural network, and then perform hidden layer calculation and output layer calculation of the water pollution source type prediction model in sequence, and finally output the water pollution source type probability vector P corresponding to each pixel of the preprocessed multispectral image; wherein the water pollution source type prediction model comprises an input layer, a hidden layer and an output layer; the number of nodes in the input layer is the same as the dimension of the multispectral feature, which is set to n, and each node receives the corresponding multispectral feature value; one hidden layer is set, and the number of nodes is m; the number of nodes in the output layer is equal to the number of water pollution source types, which is set to k, and each node outputs the probability value of the corresponding water pollution source type.

9. The water pollution source identification system based on multispectral imaging of unmanned aerial vehicles as claimed in claim 8, characterized in that: The first analysis module is specifically used for: For each pixel of the preprocessed multispectral image, according to the output water pollution source type probability vector, the pollution source type with the highest probability is selected as the pollution source type corresponding to the pixel.

10. The water pollution source identification system based on multispectral imaging of unmanned aerial vehicles according to claim 9, characterized in that: The second analysis module is specifically used for: Get the longitude lon of each pixel of the preprocessed multispectral image i and latitude lat i The geographical location information and the pollution source type corresponding to each pixel are determined i , where i represents the i-th pixel; The geographic distance d between two pixels i and j is calculated by the following formula ij : r is the radius of the earth, which is a constant parameter used to reflect the geometric scale of the earth in the formula, so that the calculated geographical distance conforms to the actual spherical characteristics of the earth and ensures the accuracy of the distance calculation; lat i ,lat j are the latitude values ​​of the i-th pixel and the j-th pixel, respectively, in degrees. They participate in trigonometric calculations in the formula and together with the longitude information determine the relative positions of the two pixels on the earth's surface, thereby calculating the distance between them; lon i ,lon j The longitude values ​​corresponding to the positions of the i-th pixel and the j-th pixel, respectively, are also in degrees, and together with the latitude values ​​constitute geographic coordinates, which are used to accurately represent the position of the pixel on the earth's surface; Core point determination: For each pixel i, calculate the number of pixels n in its ∈-neighborhood i , where ∈ is the pre-set distance threshold; let N ∈ (i) represents the ∈-neighborhood pixel set of pixel i, and the calculation formula is: If n i ≥MinPts, the pixel is marked as a core point, MinPts is the minimum point threshold; n i It refers to the number of pixels that meet the above conditions in the ∈-neighborhood of the i-th pixel, that is, the number of pixels whose distance to pixel i is less than or equal to the preset distance threshold ∈ and have the same pollution source type; t i ,t j Respectively represent the pollution source types corresponding to the i-th pixel and the j-th pixel, which are used to further screen the pixels in the ∈-neighborhood. Only when the pollution source types of the two pixels are the same will they be calculated in n i is taken into account to ensure that clustering is based on pixels of the same pollution source type, so that the final clustering result can accurately reflect the distribution of different types of water pollution sources; Cluster expansion: Starting from any unvisited core point p, create a new cluster C and add p to C, then for each pixel q in the ∈-neighborhood of p; If q is an unvisited core point, add q to C and recursively process the ∈-neighborhood; If q is not a core point but belongs to the same pollution source type as p and is not assigned to any cluster, then q is added to C; This process is repeated until all core points have been visited and all reachable pixels have been assigned to the corresponding clusters; Cluster center determination: For each cluster C m , its central location The calculation adopts the weighted average method, m represents the mth cluster, C m represents the mth cluster, which is a set of pixels with the same pollution source type and are spatially related, reflecting a distribution grouping of water pollution sources in geographic space; the calculation formula is: Represents the mth cluster C m The longitude value of the center position, in degrees, is obtained by weighted average calculation of the longitude information of all pixels in the cluster. It represents the center position of the cluster in the longitude direction in the geographic space and is used to accurately mark the approximate position of the water pollution source represented by the cluster in the east-west direction on the map of the geographic information GIS system; i is the pixel weight, which is a value determined in advance based on the relevant characteristics of the pixel's spectral reflectance in the multispectral image. It is used to reflect the different importance of different pixels to the determination of the center position when calculating the cluster center. Distribution range determination: calculation of cluster C m The range in longitude and latitude; Cluster C m The minimum and maximum values ​​of pixel longitude, Cluster C m The minimum and maximum values ​​of the pixel latitude and distribution range can be expressed as: Respectively represent the mth cluster C m The distribution range in the longitude and latitude directions, in degrees, is obtained by calculating the difference between the maximum and minimum values ​​of the longitude and latitude of the pixels in the cluster. It is used to intuitively reflect the coverage of the water pollution source represented by the cluster in the east-west direction, which helps to accurately draw the distribution area shape and size of the pollution source on the geographic information GIS system map; The center position of each cluster And distribution range information The data are imported into the geographic information system (GIS), and the drawing function of the geographic information system is used to draw the distribution area of ​​water pollution sources represented by each cluster using different colors or symbols according to different types of pollution sources.

Images