Precise identification system of water pollutants by multispectral image fusion

Abstract

The present application relates to the technical field of water pollution monitoring, and specifically relates to a multispectral image fusion water pollutant accurate identification system, comprising a data acquisition module, a cloud processing module, a boundary processing module, a pollution identification module, a diffusion prediction module and a visualization module, the system uses differential geometry manifold theory to construct a sub-pixel representation of the water pollutant boundary, realizes high-precision pollutant boundary delineation through multi-scale analysis and curvature flow optimization technology, adopts multispectral image fusion and optimal band selection technology to enhance the pollutant feature expression, uses a support vector machine model to accurately identify various pollutant types such as oil film, oil stain and algal bloom, and based on the boundary fine delineation result and the pollutant type identification result, in combination with historical flow, wind direction and wind speed data, accurately predicts the diffusion trend of the pollutant, the system improves the water pollutant boundary identification precision and enhances the identification ability for complex boundary morphology and low-contrast regions.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of water pollution monitoring, and particularly relates to a multi-spectral image fusion water pollution accurate identification system which can be applied to water pollution monitoring, identification and diffusion trend prediction of lakes, rivers, near seas and the like. BACKGROUND

[0002] With the acceleration of industrialization and urbanization, water pollution problems are becoming increasingly serious, posing a severe challenge to water environment protection and water resource management. Accurate identification of water pollution type, range and diffusion trend is of great significance for water environment monitoring, pollution prevention and emergency disposal.

[0003] Traditional water pollution monitoring mainly relies on manual sampling analysis and single sensor detection, which has the problems of limited monitoring range, poor timeliness and high cost. With the development of remote sensing technology, remote sensing images based on satellites, aircraft and unmanned aerial vehicles are widely used in water pollution monitoring, with the advantages of large range, real-time and low cost. However, the current water pollution identification system based on remote sensing images still has the following problems:

[0004] 1. The multi-spectral image fusion precision is insufficient, and it is difficult to effectively utilize the complementary information of different wavebands;

[0005] 2. The pollution boundary identification precision is limited, usually only to the pixel level, which is difficult to meet the fine identification demand;

[0006] 3. The pollution type identification ability is weak, especially the ability to distinguish mixed pollution types;

[0007] 4. There is a lack of effective prediction mechanism for pollution diffusion trend.

[0008] Therefore, it is urgent to develop a system capable of accurately identifying water pollution, realizing multi-spectral image fusion, sub-pixel level boundary delineation, accurate pollution classification and diffusion trend prediction, and providing technical support for water environment protection. SUMMARY

[0009] The purpose of the present application is to provide a multi-spectral image fusion water pollution accurate identification system, aiming to solve the problems of insufficient multi-spectral image fusion precision, limited pollution boundary identification precision, weak pollution type identification ability and lack of diffusion trend prediction mechanism in the prior art.

[0010] The present application provides a multi-spectral image fusion water pollution accurate identification system, comprising:

[0011] A data acquisition module is used to acquire multi-spectral image data of a water area to be detected, and transmit the multi-spectral image data to a cloud processing module;

[0012] The cloud processing module is in communication connection with the data acquisition module, and is configured to receive the multispectral image data, pre-process the multispectral image data, and transmit the pre-processed data to the boundary processing module.

[0013] The boundary processing module is in communication connection with the cloud processing module, and is configured to acquire the pre-processed data, construct a sub-pixel representation of a water body pollutant boundary based on a differential geometric manifold theory, generate a boundary fine delineation result, and transmit the boundary fine delineation result to the pollution identification module.

[0014] The pollution identification module is in communication connection with the boundary processing module, and is configured to receive the boundary fine delineation result, classify and identify water body pollutants based on the boundary fine delineation result, and generate a pollutant type identification result.

[0015] The diffusion prediction module is in communication connection with the boundary processing module and the pollution identification module, and is configured to receive the boundary fine delineation result and the pollutant type identification result, predict a diffusion trend of the water body pollutants based on the boundary fine delineation result and the pollutant type identification result.

[0016] The visualization module is in communication connection with the pollution identification module and the diffusion prediction module, and is configured to receive the pollutant type identification result and the diffusion trend, and generate a visual display result of a water body pollution state.

[0017] Preferably, the data acquisition module comprises:

[0018] The first acquisition unit is configured to acquire a multi-band remote sensing image of a water area to be detected.

[0019] The second acquisition unit is configured to acquire remote sensing reflection radiation data of the water area to be detected.

[0020] The data transmission unit is configured to transmit the multi-band remote sensing image and the remote sensing reflection radiation data to the cloud processing module.

[0021] Preferably, the cloud processing module comprises:

[0022] The image processing unit is configured to perform radiation calibration, atmospheric correction and geometric correction on the multispectral image data to obtain a pre-processed image.

[0023] The image noise reduction unit is configured to correct speckle noise in the pre-processed image to generate a denoised image.

[0024] The multispectral stitching unit is configured to perform multispectral stitching on the denoised image to generate a false color image.

[0025] An optimal band selection unit is configured to select an optimal band combination from the multispectral image data.

[0026] Preferably, the multispectral stitching unit selects and stitches images based on four criteria of cloud coverage, cloud-free water body, spectral variation, and image clarity, and combines features of the four stitched images.

[0027] Preferably, the image denoising unit extracts speckle noise in the preprocessed image, classifies the speckle noise based on morphological features, divides the speckle noise into circular speckle noise and fan-shaped speckle noise, and corrects different types of speckle noise using different correction strategies.

[0028] Preferably, the boundary processing module includes:

[0029] A boundary initialization unit is configured to perform multi-scale boundary detection on the preprocessed data to generate an initial boundary point set.

[0030] A manifold modeling unit is configured to model the initial boundary point set as a one-dimensional smooth manifold in a two-dimensional Euclidean space, construct a parameterized mapping, and generate a boundary manifold representation.

[0031] A sub-pixel optimization unit is configured to construct a boundary energy functional based on curvature flow theory in differential geometry, perform variational optimization on the boundary manifold representation in a multi-scale space, and generate a sub-pixel precision boundary representation.

[0032] A boundary feature extraction unit is configured to extract morphological features, curvature distribution, and dynamic characteristics from the sub-pixel precision boundary representation, and generate a boundary feature description.

[0033] Preferably, the multi-scale space of the sub-pixel optimization unit includes a sequence of scale parameters, and the scale-invariant essential features are extracted by analyzing boundary features and topological structures at different scales; the sub-pixel optimization unit uses gradient descent method to minimize the energy functional, realizes local deformation of the boundary by tangential translation, and keeps the topological structure of the curve unchanged during the iteration process.

[0034] Preferably, the pollution identification module uses a support vector machine model to classify water pollutants, and the identified water pollution state types include oil film, oil spot, oil film and oil spot, algal material, green tide, red tide, brown tide, colorless heterotrophic bacteria bloom, and their combination types.

[0035] Preferably, the diffusion prediction module establishes a pollutant diffusion model based on the temporal variation of the boundary fine depiction result, combines historical flow data, wind direction data, and wind speed data, calculates the diffusion speed and diffusion direction of the pollutant, and generates a pollutant diffusion trend prediction result.

[0036] As preferred, the visualization module displays the pollution diffusion trend in the form of coordinate axes, including a time coordinate axis and a space coordinate axis; the time coordinate axis includes a time axis and a clock control; the space coordinate axis includes a color representation control and a size adjustment control; the visualization module further includes a pollution diffusion trend graph, and the diffusion trend graph includes a pollution source schematic diagram, a diffusion direction, and a diffusion speed prediction graph; the visualization module uses different colors to represent different pollution indexes.

[0037] The present application realizes high-precision pollution boundary delineation by adopting differential geometric manifold theory to construct the sub-pixel representation of water pollution boundary, combining multi-scale analysis and curvature flow optimization technology; enhances pollution feature expression by multi-spectral image fusion and optimal band selection; realizes accurate identification of various types of pollutants by support vector machine model; and realizes accurate prediction of pollution diffusion trend by establishing diffusion model.

[0038] The present application has the following beneficial effects:

[0039] 1. Sub-pixel level boundary delineation: by differential geometric manifold theory, the water pollution boundary is modeled as a continuous differential manifold, realizing sub-pixel level boundary positioning, with a 5-10 times improvement in precision, providing a basis for accurate measurement of pollution area and prediction of diffusion trend;

[0040] 2. Enhanced recognition adaptability: multi-scale analysis and curvature flow optimization are adopted, so that the system can adapt to the boundary features of different types of pollutants, showing good recognition effect on different pollutants such as oil film and algae bloom, and the recognition accuracy in complex boundary morphology (such as mixed area of oil film and algae bloom) and low contrast area (such as low concentration of pollutants) is improved by 35.2% and 41.8% respectively;

[0041] 3. Accurate diffusion prediction: based on high-precision boundary representation and pollution type identification results, combined with historical flow, wind direction and wind speed data, the system can accurately predict the diffusion trend of pollutants, with an average error reduction of 42.3% and 36.7% for short-term prediction (within 6 hours) and medium-term prediction (within 24 hours), providing strong support for pollution emergency disposal;

[0042] 4. High efficiency of calculation performance: the system adopts parameterized representation and local update strategy, which significantly improves the calculation efficiency while maintaining high precision, is 3-5 times faster than similar sub-pixel methods, and reduces memory demand by 40%-60%, meeting the actual application requirements. BRIEF DESCRIPTION OF DRAWINGS

[0043] Figure 1 The overall architecture diagram of the present application multi-spectral image fusion water pollution accurate identification system;

[0044] Figure 2 It is a structural schematic diagram of the data acquisition module of the present application;

[0045] Figure 3 It is a processing flow chart of the cloud processing module of the present application;

[0046] Figure 4 It is a working flow chart of the boundary processing module of the present application;

[0047] Figure 5 It is a classification flow chart of the pollution identification module of the present application;

[0048] Figure 6 It is a prediction flow chart of the diffusion prediction module of the present application. DETAILED DESCRIPTION

[0049] Please refer to Figure 1 - Figure 6 The present application will be further described in detail below in combination with the drawings and specific embodiments.

[0050] With reference to Figure 1 The present application provides a multi-spectral image fusion water pollutant accurate identification system, which comprises a data acquisition module 1, a cloud processing module 2, a boundary processing module 3, a pollution identification module 4, a diffusion prediction module 5 and a visualization module 6.

[0051] The data acquisition module 1 is used for acquiring multi-spectral image data of a water area to be detected, and transmitting the multi-spectral image data to the cloud processing module 2; the cloud processing module 2 is in communication connection with the data acquisition module 1, and is used for receiving the multi-spectral image data, pre-processing the multi-spectral image data, and transmitting the pre-processed data to the boundary processing module 3; the boundary processing module 3 is in communication connection with the cloud processing module 2, and is used for obtaining the pre-processed data, constructing a sub-pixel representation of a water pollutant boundary based on a differential geometric manifold theory, generating a boundary fine depiction result, and transmitting the boundary fine depiction result to the pollution identification module 4; the pollution identification module 4 is in communication connection with the boundary processing module 3, and is used for receiving the boundary fine depiction result, classifying and identifying water pollutants based on the boundary fine depiction result, and generating a pollutant type identification result; the diffusion prediction module 5 is in communication connection with the boundary processing module 3 and the pollution identification module 4, and is used for receiving the boundary fine depiction result and the pollutant type identification result, predicting a diffusion trend of the water pollutants based on the boundary fine depiction result and the pollutant type identification result; the visualization module 6 is in communication connection with the pollution identification module 4 and the diffusion prediction module 5, and is used for receiving the pollutant type identification result and the diffusion trend, and generating a visual display result of a water pollution state.

[0052] With reference to Figure 2In a preferred embodiment of the present application, the data acquisition module 1 comprises a first acquisition unit 11, a second acquisition unit 12 and a data transmission unit 13. The first acquisition unit 11 is used to acquire multi-band remote sensing images of the water area to be detected, including visible red band, visible green band, visible blue band, visible near-infrared band and short-wave infrared band. Preferably, the first acquisition unit 11 can be a satellite remote sensing sensor, a multi-spectral camera carried by a drone or a high-precision video camera device on a water quality monitoring ship. The second acquisition unit 12 is used to acquire remote sensing reflection radiation data of the water area to be detected, providing spectral feature information. The data transmission unit 13 is used to transmit the multi-band remote sensing images and the remote sensing reflection radiation data to the cloud processing module 2, which can use wireless transmission methods such as 4G / 5G network, satellite communication, etc., or wired transmission methods such as optical fiber network, etc.

[0053] Referring to Figure 3 In a preferred embodiment of the present application, the cloud processing module 2 comprises an image processing unit 21, an image denoising unit 22, a multi-spectral stitching unit 23 and an optimal band selection unit 24. The image processing unit 21 is used to perform radiation calibration, atmospheric correction and geometric correction on the multi-spectral image data to obtain preprocessed images. Specifically, radiation calibration converts the DN value in the image into remote sensing reflection radiation value, atmospheric correction eliminates atmospheric influence to obtain ground surface reflection spectrum, and geometric correction preserves spatial position accuracy. The image denoising unit 22 is used to correct the speckle noise in the preprocessed image to generate a denoised image. The multi-spectral stitching unit 23 is used to perform multi-spectral stitching on the denoised image to generate a false color image. The optimal band selection unit 24 is used to select the optimal band combination from the multi-spectral image data to provide high-quality data basis for subsequent processing.

[0054] In a preferred embodiment of the present application, the multi-spectral stitching unit 23 performs image selection and stitching based on four criteria of cloud coverage, cloud-free water body, spectral variation and image clarity, and combines the features of the four stitched images. Preferably, the cloud coverage criterion is used to select areas with little or no cloud coverage; the cloud-free water body criterion is used to ensure that the water body area is visible; the spectral variation criterion is used to select areas with large spectral variation between bands to enhance contrast; and the image clarity criterion is used to select clear images with rich details. In addition, the multi-spectral stitching uses wavelet transform theory to compare and fuse the pixels of different spectral band images, achieving information complementation.

[0055] In a preferred embodiment of the present application, the image denoising unit 22 extracts the speckle noise in the preprocessed image, classifies the speckle noise based on morphological features, divides the speckle noise into circular speckle noise and fan-shaped speckle noise, and corrects different types of speckle noise using different correction strategies. Preferably, for circular speckle noise, its area is first calculated, then a circular mask layer is generated based on the area, the circular mask layer is superimposed on the image, the average value of the pixels in the mask is calculated, and the average value is used to replace the noise pixel; for fan-shaped speckle noise, its diagonal value is calculated, and the image value of the speckle noise is replaced with the diagonal value. This differential processing strategy significantly improves the image quality and provides high-quality input data for subsequent boundary processing.

[0056] Reference Figure 4 In a preferred embodiment of the present application, the boundary processing module 3 includes a boundary initialization unit 31, a manifold modeling unit 32, a sub-pixel optimization unit 33, and a boundary feature extraction unit 34. The boundary initialization unit 31 is used to perform multi-scale boundary detection on the preprocessed data to generate an initial boundary point set. The manifold modeling unit 32 is used to model the initial boundary point set as a one-dimensional smooth manifold in a two-dimensional Euclidean space, construct a parameterized mapping, and generate a boundary manifold representation. The sub-pixel optimization unit 33 is used to construct a boundary energy functional based on the curvature flow theory in differential geometry, perform variational optimization on the boundary manifold representation in a multi-scale space, and generate a sub-pixel precision boundary representation. The boundary feature extraction unit 34 is used to extract morphological features, curvature distribution, and dynamic characteristics from the sub-pixel precision boundary representation to generate a boundary feature description.

[0057] The core innovation of the manifold modeling unit 32 is to model the pollutant boundary as a continuous differential manifold. Traditional pollutant boundary detection only stops at pixel-level precision, while the present system represents the boundary as a parameterized curve through the manifold theory. The specific implementation process includes the following steps:

[0058] First, the multi-spectral fusion image is preliminarily segmented to obtain an initial boundary point set , where represents a point on the boundary. Each point contains sub-pixel coordinates , a normal vector , a curvature , and a confidence , etc.

[0059] Secondly, based on the curvature variation, the boundary is divided into multiple segments , each segment contains parameter range , point set index range , etc. When the curvature variation exceeds the threshold When the value is preferably set to 0.15, segmentation is performed while ensuring the minimum segment length. (Preferred setting: 10 pixels).

[0060] Then, a parameterized representation is constructed for each segment. , Cubic B-spline representation is preferred, with control point intervals set at 5-10 pixels to ensure curve smoothness and accuracy. Adjacent segments are constrained using continuity constraints. Smooth connections ensure the continuity of the overall boundary.

[0061] Based on the manifold representation, the differential geometric properties of the boundary are calculated, including the tangent vector T(t), the normal vector N(t), and the curvature κ(t):

[0062] ,

[0063] Where T(t) is the unit tangent vector along the boundary curve at parameter t, γ(t) is the parameterized boundary curve, dγ(t) / dt represents the derivative of the curve at t (velocity vector), and ||dγ(t) / dt|| represents the magnitude of the velocity vector (Euclidean norm).

[0064] ,

[0065] Where N(t) is the unit normal vector at the boundary curve parameter t. This represents the operation of rotating 90 degrees counterclockwise in a two-dimensional plane, which rotates the tangent vector to obtain the normal vector.

[0066] ,

[0067] Where κ(t) is the curvature of the boundary curve at parameter t, and x(t) and y(t) are the components of the parametric curve γ(t) in the x and y directions, respectively. and They are respectively and For parameters The first derivative, and They are respectively and For parameters The second derivative of . The curvature κ(t) describes the degree of curvature of the curve at that point; the larger the value, the greater the curvature.

[0068] This boundary representation method based on manifold theory maps the boundary in the discrete pixel space to the continuous parameter space, achieving sub-pixel level accurate positioning (accuracy up to 1 / 20 of a pixel), overcoming the boundary jaggedness and discontinuity problems in traditional methods.

[0069] Sub-pixel optimization unit 33 employs multi-scale analysis and curvature flow theory to optimize the boundary. In multi-scale space, the boundary has different representations at different scales:

[0070] ,

[0071] in, The scale parameter is The boundary representation, For the original boundary representation, The standard deviation is Gaussian kernel function, This represents the convolution operation. Gaussian kernel function. Defined as:

[0072] ,

[0073] in, The standard deviation of the Gaussian kernel controls the smoothing degree. This is the base of the natural logarithm (approximately 2.718). Convolution operation. Indicates along the parameter Integrals:

[0074] ,

[0075] In practice, this integral is usually calculated using a discrete approximation.

[0076] Preferably, the system sets a scale sequence. It covers different levels of features, from detailed to global. It analyzes boundary features and topological structures at each scale, identifies cross-scale stable feature points, and extracts essential features. Feature point response threshold. The preferred setting is 0.05, which represents the cross-scale matching tolerance. Set to 2.5 pixels to ensure the stability and consistency of feature points.

[0077] Boundary optimization employs a variational method, defining the energy functional:

[0078] ,

[0079] in, For the total energy functional, For each data item, the degree of matching between the boundary and image features is represented. For smoothing terms, control the smoothness of the boundaries; As a priori term, prior knowledge of the morphology of specific pollutants is introduced. , , These are the corresponding weight parameters, preferably set as follows: , , . The sum of these weight parameters equals 1, ensuring the balance of energy terms.

[0080] Data item is defined as the matching degree of boundary and image gradient:

[0081] ,

[0082] where, is the image gradient at the boundary position, is the boundary normal vector. This term encourages the boundary to be located in the area with large image gradient.

[0083] Smoothness term is defined as the squared integral of boundary curvature:

[0084] ,

[0085] where, is the boundary curvature. This term penalizes the excessive bending of boundary, promoting the boundary to be smoother.

[0086] Prior term is set according to different types of pollutants, for example, for oil film type, it can be defined as:

[0087] ,

[0088] where, is the reference boundary shape, determined by the prior knowledge of pollutant shape. This term encourages the boundary shape to conform to the typical characteristics of a specific pollutant.

[0089] The energy functional is minimized by gradient descent method:

[0090] ,

[0091] where, is the boundary representation at the t-th iteration, is the updated boundary representation, is the learning rate (initial value set to 0.01, gradually decaying), is the gradient of energy functional E with respect to boundary . The gradient is calculated as the variational derivative of energy functional with respect to boundary:

[0092] ,

[0093] During the iteration process, the tangential translation method is used to realize the local deformation of the boundary, while maintaining the topological structure of the curve unchanged. When the difference between adjacent iteration results is less than the convergence threshold The optimization process is terminated when the error e reaches a minimum value e_min (set as 1e-4) or the maximum iteration number N_max (set as 100) is reached:

[0094] ,

[0095] wherein, denotes the Euclidean distance norm.

[0096] The multi-scale optimization method can effectively handle boundary features at different scales and adapt to the morphological characteristics of various pollutants, especially showing excellent processing capability for complex boundary morphology (such as irregularly diffused edges of oil film, dispersed and aggregated structures of algae, etc.).

[0097] The boundary feature extraction unit 34 extracts key features from the optimized sub-pixel boundary, including morphological features (such as perimeter, area, complexity, etc.), curvature distribution features (such as average curvature, maximum curvature, curvature change rate, etc.), and dynamic characteristics (such as boundary change rate, diffusion direction, etc.). These features provide important basis for subsequent pollutant identification and diffusion prediction.

[0098] With reference to Figure 5 In a preferred embodiment of the present application, the pollution identification module 4 uses a support vector machine model to classify water pollutants, and the identified water pollution state types include oil film, oil spot, oil film and oil spot, algae, green tide, red tide, brown tide, colorless heterotrophic bacteria bloom, and their combination types.

[0099] The support vector machine model is trained and classified based on the boundary features and morphological features extracted from the boundary processing module 3. Preferably, a radial basis function is used as the kernel function:

[0100] ,

[0101] wherein, is the kernel function, and denotes the similarity between two points and y in the feature space, and y are feature vectors, is the kernel parameter (controls the width of the radial basis function), which is preferably set to 0.1, and exp denotes the exponential function, denotes the Euclidean distance between x and y.

[0102] For different types of water pollutants, the system establishes a detailed feature description library:

[0103] 1) Oil film: in the form of an oil film floating on the water surface, with a clear oil film boundary, small boundary curvature, and relatively continuous morphology;

[0104] 2) Oil spots: There are obvious black, brown, yellow, and gray spots on the water surface, with large variations in the curvature of the boundaries and relatively irregular shapes;

[0105] 3) Algae material: It appears in the form of blue-green plant blooms, with relatively blurred boundaries and rich internal texture;

[0106] 4) Green tide: It appears on the water surface as a light green or grayish-green algal bloom, with relatively clear boundaries and a relatively uniform color inside;

[0107] 5) Red tide: It appears on the water surface as light red, green, or yellow algal blooms with relatively clear boundaries and reddish internal color.

[0108] 6) Brown tide: It appears on the water surface as a grayish-brown or yellow algal bloom with relatively clear boundaries and a brownish hue inside;

[0109] 7) Colorless heterotrophic bacterial bloom: It appears as a white, foamy bloom with a clear boundary in the middle. The boundary is relatively clear but irregular.

[0110] 8) Mixed types: such as algal oil film, red tide oil spots, etc., which have the characteristics of a mixture of two pollutants.

[0111] Support Vector Machine (SVM) models achieve multi-class classification through a one-to-many classification strategy, with an accuracy of over 90%, and exhibit excellent classification performance, especially when dealing with mixed pollution types.

[0112] Reference Figure 6 In a preferred embodiment of the present invention, the diffusion prediction module 5 establishes a pollutant diffusion model based on the temporal changes of the boundary fine characterization results, combined with historical flow data, wind direction data and wind speed data, calculates the diffusion speed and diffusion direction of pollutants, and generates a pollutant diffusion trend prediction result.

[0113] The diffusion prediction method combines physical models with data-driven approaches. First, the diffusion velocity and direction are calculated based on the boundary changes at two consecutive time points. Second, historical flow data (such as average flow velocity) and meteorological data (such as wind direction and wind speed) are used for correction. Finally, the diffusion trend in the future is predicted based on the corrected diffusion parameters.

[0114] The formula for calculating the pollutant diffusion rate is:

[0115] ,

[0116] Where V is the diffusion velocity vector (m / s), representing the actual movement velocity of the pollutant; U is the water flow velocity vector (m / s), representing the flow velocity of the water body itself; P is the pollutant density (kg / m³), representing the mass of the pollutant per unit volume; and ρ is the density of the polluted area (kg / m³). 3The density of water is typically represented by P, and the wind speed vector (m / s) represents the wind speed and direction. P / ρ represents the density ratio of pollutants to water, influencing the wind's contribution to pollutant movement. Water flow velocity is usually obtained from historical data, typically ranging from 0.1 to 2.0 m / s; pollutant density is determined based on the type of pollutant, for example, oil film is approximately 850 kg / m³. 3 The algae content is approximately 1050 kg / m³. 3 Regional density is typically taken from the density of water, which is 1000 kg / m³; wind speed is obtained from meteorological data and is generally in the range of 0-30 m / s.

[0117] The diffusion direction is the same as the diffusion velocity vector direction, and the calculation formula is:

[0118] ,

[0119] Where θ is the diffusion direction angle (rad), indicating the direction of pollutant diffusion; and These are the components of the diffusion velocity vector V along the x-axis (east-west direction) and y-axis (north-south direction), respectively; arctan is the arctangent function, which converts the ratio of the velocity components into an angle.

[0120] Based on diffusion velocity and direction, the system uses diffusion equations to predict the future distribution of pollutants:

[0121] ,

[0122] in, Indicates position ( The pollutant concentration at time t; The time step represents the time interval for prediction. and The diffusion rates are respectively at and Component of direction; Let h⁻¹ be the decay coefficient, representing the natural decay rate of pollutants over time; e is the base of the natural logarithm. This equation represents the position ( In time The concentration equals the position In time The concentration after attenuation reflects the movement and decay of pollutants over time. Attenuation coefficient. Depending on the type of contaminant, the oil film duration is approximately 0.005–0.02 h. -1 Algae consumption is approximately 0.001–0.01 h. -1 .

[0123] The system supports different time scale predictions: short-term prediction (within 6 hours, average error <15%), medium-term prediction (within 24 hours, average error <25%) and long-term trend prediction (direction accuracy >70%). This multi-time scale prediction capability provides strong decision support for pollution emergency disposal.

[0124] In a preferred embodiment of the present application, the visualization module 6 displays the pollution diffusion trend in the form of coordinate axes, including a time coordinate axis and a space coordinate axis. The time coordinate axis includes a time axis and a clock control for representing the change of the pollutant over time; the space coordinate axis includes a color representation control and a size adjustment control for representing the spatial distribution and severity of the pollutant.

[0125] The visualization module 6 also includes a pollution diffusion trend chart, which includes a pollution source schematic diagram, a diffusion direction and a diffusion speed prediction chart. The pollution source schematic diagram identifies the origin position of the pollutant; the diffusion direction chart shows the main direction of the pollutant diffusion, usually represented by arrows; the diffusion speed prediction chart shows the diffusion rate of the pollutant in different directions, usually represented by vectors of different lengths.

[0126] The system uses different colors to represent different pollution indexes, such as green for light pollution (pollution index <0.3), yellow for moderate pollution (pollution index 0.3-0.6) and red for severe pollution (pollution index >0.6). The color coding scheme is based on the internationally recognized environmental pollution index standard, which intuitively reflects the pollution level.

[0127] Preferably, the visualization module 6 also supports a variety of interactive operations such as zooming, panning, time axis adjustment, etc., allowing users to observe the pollution situation from different angles and scales. At the same time, the system supports a variety of output formats such as real-time screen display, picture saving, video recording and report generation, meeting the use requirements in different scenarios.

[0128] The multi-spectral image fusion water pollutant accurate identification system of the present application can be applied to various scenarios:

[0129] 1) Lake pollution monitoring: The system obtains multi-spectral images through satellite remote sensing and unmanned aerial vehicle cruising, realizing large-area and periodic pollution monitoring, especially suitable for monitoring blue-green algae blooms in large lakes such as Taihu Lake and Dongting Lake.

[0130] 2) River pollution tracking: The system combines data collected by shore-based monitoring stations and water quality monitoring ships to realize real-time tracking of mobile pollutants, suitable for monitoring pollution incidents in important rivers such as the Yangtze River and the Yellow River.

[0131] 3) Marine oil pollution monitoring: The system realizes large-scale monitoring of marine oil pollution through multispectral data collected by satellites and marine monitoring ships, especially suitable for responding to emergencies such as oil tanker leaks at sea, oil platform accidents, etc.

[0132] In practical applications, the system exhibits excellent performance: recognition accuracy exceeds 90%, sub-pixel positioning accuracy reaches 1 / 20 pixels, diffusion prediction error is less than 15% (short-term) and 25% (medium-term), and processing efficiency is improved by 3-5 times compared to similar systems. These performance indicators make the system have significant technical advantages and application value in the field of water pollution monitoring.

[0133] The above embodiments are only preferred embodiments of the present application and are not intended to limit the present application. Those skilled in the art can make various modifications and improvements without departing from the spirit and scope of the present application, and these modifications and improvements should also be considered as within the scope of the present application.

Claims

1. A precise identification system of water pollutants by multispectral image fusion, characterized in that, The application relates to a water pollution detection system and method. The system comprises: a data acquisition module for acquiring multi-spectral image data of a water area to be detected and transmitting the multi-spectral image data to a cloud processing module; a cloud processing module in communication connection with the data acquisition module, configured to receive the multi-spectral image data, pre-process the multi-spectral image data, and transmit the pre-processed data to a boundary processing module; a boundary processing module in communication connection with the cloud processing module, configured to obtain the pre-processed data, construct a sub-pixel representation of a water body pollutant boundary based on a differential geometric manifold theory, generate a boundary fine delineation result, and transmit the boundary fine delineation result to a pollution identification module; a pollution identification module in communication connection with the boundary processing module, configured to receive the boundary fine delineation result, classify and identify water body pollutants based on the boundary fine delineation result, and generate a pollutant type identification result; a diffusion prediction module in communication connection with the boundary processing module and the pollution identification module, configured to receive the boundary fine delineation result and the pollutant type identification result, and predict a diffusion trend of the water body pollutants based on the boundary fine delineation result and the pollutant type identification result; a visualization module in communication connection with the pollution identification module and the diffusion prediction module, configured to receive the pollutant type identification result and the diffusion trend, and generate a visual display result of a water body pollution state. The boundary processing module comprises: a boundary initialization unit configured to perform multi-scale boundary detection on the pre-processed data to generate an initial boundary point set; a manifold modeling unit configured to model the initial boundary point set into a one-dimensional smooth manifold in a two-dimensional Euclidean space, construct a parameterized mapping, and generate a boundary manifold representation; a sub-pixel optimization unit configured to construct a boundary energy functional based on a curvature flow theory in differential geometry, perform variational optimization on the boundary manifold representation in a multi-scale space, and generate a sub-pixel precision boundary representation; 2. The multispectral image fusion-based precise identification system of water pollutants according to claim 1, characterized in that, a boundary feature extraction unit configured to extract morphological features, curvature distribution and dynamic characteristics from the sub-pixel precision boundary representation, and generate a boundary feature description. The data acquisition module comprises: a first acquisition unit configured to acquire multi-band remote sensing images of a water area to be detected; a second acquisition unit configured to acquire remote sensing reflection radiation data of the water area to be detected; 3. The multispectral image fusion-based precise identification system of water pollutants according to claim 1, characterized in that, a data transmission unit configured to transmit the multi-band remote sensing images and the remote sensing reflection radiation data to the cloud processing module. The cloud processing module comprises: an image processing unit configured to perform radiation scaling, atmospheric correction and geometric correction on the multi-spectral image data to obtain a pre-processed image; an image noise reduction unit configured to correct speckle noise in the pre-processed image to generate a noise-reduced image; a multi-spectral stitching unit configured to perform multi-spectral stitching on the noise-reduced image to generate a false color image; 4. The multispectral image fusion-based precise identification system of water pollutants according to claim 3, characterized in that, an optimal band selection unit configured to select an optimal band combination from the multi-spectral image data. The multi-spectral stitching unit performs image selection and stitching based on four criteria of cloud coverage, cloud-free water body, spectral variation and image clarity, and combines features of the four stitched images.

5. The multispectral image fusion-based precise identification system of water pollutants according to claim 3, characterized in that, The image denoising unit extracts the speckle noise in the preprocessed image, classifies the speckle noise based on morphological features, divides the speckle noise into circular speckle noise and fan-shaped speckle noise, and corrects different types of speckle noise by using different correction strategies.

6. The multispectral image fusion-based precise identification of water pollutant system according to claim 1, characterized in that, The multi-scale space of the sub-pixel optimization unit includes a scale parameter sequence, and the boundary features and topological structures are analyzed at different scales to extract the essential features that are invariant to scale. The sub-pixel optimization unit uses the gradient descent method to minimize the energy functional, realizes local deformation of the boundary by tangential translation, and keeps the curve topological structure unchanged in the iteration process.

7. The multispectral image fusion-based precise identification of water pollutant system according to claim 1, characterized in that, The pollution identification module uses a support vector machine model to classify water pollutants, and the identified water pollution state types include oil film, oil spot, oil film and oil spot, algae material, green tide, red tide, brown tide, colorless heterotrophic bacteria bloom, and their combination types.

8. The multispectral image fusion-based precise identification of water pollutant system according to claim 1, characterized in that, The diffusion prediction module establishes a pollutant diffusion model according to the time sequence variation of the boundary fine description result, combines historical flow data, wind direction data and wind speed data, calculates the diffusion speed and diffusion direction of the pollutant, and generates a pollutant diffusion trend prediction result. 9.The multispectral image fusion-based precise identification system of water body pollutants according to claim 1, characterized in that, The visualization module displays the pollutant diffusion trend in the form of coordinate axes, including a time coordinate axis and a space coordinate axis; the time coordinate axis includes a time axis and a clock control; the space coordinate axis includes a color representation control and a size adjustment control; the visualization module also includes a pollutant diffusion trend graph, which includes a pollution source schematic diagram, a diffusion direction and a diffusion speed prediction graph; the visualization module uses different colors to represent different pollution indexes.

Images