Method and system for generating a map of cyanobacteria distribution based on lake video monitoring

Abstract

The present invention provides a method and system for generating a blue algae distribution map based on lake video monitoring, which relates to the technical field of water environment monitoring, including: obtaining multispectral remote sensing image data covering the lake area, real-time video image data collected by various video monitoring points, and geographical location coordinate information corresponding to various video monitoring points; performing spectral extraction processing on the multispectral remote sensing image data to extract the spectral classification characteristics of the multispectral remote sensing image data, and determining the spatial distribution map of the blue algae remote sensing inversion according to the spectral classification characteristics through a preset blue algae inversion model; determining the real-time video image data, geographical location coordinate information, and spatial distribution map collected by various video monitoring points at the same time node as training samples, and sending the training samples to a preset diffusion model for fusion modeling processing to generate a blue algae distribution map. The present invention can significantly improve the spatiotemporal resolution and image quality of a blue algae distribution map.

Description

Technical Field

[0001] The present invention relates to the technical field of water environment monitoring, and in particular to a method and system for generating a cyanobacteria distribution map based on lake video monitoring. Background Art

[0002] In order to prevent cyanobacteria from destroying the balance of aquatic ecosystems, it is necessary to monitor the distribution range and changing trends of cyanobacteria in lakes. The traditional monitoring method is to use remote sensing monitoring, obtain multispectral remote sensing images using satellites or aviation, and classify and extract cyanobacteria based on their spectral response characteristics in different bands. However, the temporal resolution of this method is usually low. Even if MODIS remote sensing images with a temporal resolution of 1 day are used, due to obstruction by clouds, the actual unobstructed high-quality remote sensing images obtained are usually more than 1 week apart. Therefore, continuous dynamic monitoring cannot be achieved.

[0003] At present, relevant technologies have proposed that a video surveillance system can be deployed around the lake to identify areas where cyanobacteria are concentrated by analyzing video images. However, a single video surveillance system cannot cover the entire lake and cannot provide regulatory agencies with a map of the spatial distribution of cyanobacteria in geographic coordinates. Summary of the invention

[0004] In view of this, the purpose of the present invention is to provide a method and system for generating a cyanobacteria distribution map based on lake video monitoring, which can significantly improve the spatiotemporal resolution and image quality of the cyanobacteria distribution map.

[0005] In the first aspect, an embodiment of the present invention provides a method for generating a map of cyanobacteria distribution based on lake video surveillance. The method is applied to a system for generating a map of cyanobacteria distribution based on lake video surveillance, and the method includes: obtaining multispectral remote sensing image data covering the lake area, and obtaining real-time video image data collected by each video surveillance point and geographical location coordinate information corresponding to each video surveillance point through multiple video surveillance points set around the lake; performing spectral extraction processing on the multispectral remote sensing image data to extract spectral classification features of the multispectral remote sensing image data, and determining the spatial distribution map of cyanobacteria remote sensing inversion according to the spectral classification features through a preset cyanobacteria inversion model; determining the real-time video image data collected by each video surveillance point at the same time node, the geographical location coordinate information corresponding to each video surveillance point, and the spatial distribution map as training samples, and sending the training samples to a preset diffusion model for fusion modeling processing to generate a map of cyanobacteria distribution.

[0006] In one embodiment, before the step of obtaining real-time video image data collected by each video surveillance point and the geographical location coordinate information corresponding to each video surveillance point through multiple video surveillance points set up around a lake, the method includes: using a preset ray tracing method, interactively processing the preset lake monitoring layout three-dimensional model and the field of view area of ​​the video surveillance point, determining a binary mask map of the effective field of view area of ​​each video surveillance point, and constructing a video surveillance point layout combination optimization model using the binary mask map, the preset number of video surveillance points and the video surveillance point setting information, wherein the preset lake monitoring layout three-dimensional model includes: a digital elevation model, a water body range model of the lake and an obstruction model, the binary mask map includes: a visible area and an obstruction area, and the video surveillance point setting information includes: the position, height and viewing angle of the video surveillance point; iteratively processing the video surveillance point layout combination optimization model to determine the target layout plan of the video surveillance points, so as to use the target layout plan to set video surveillance points around the lake.

[0007] In one embodiment, an iterative processing is performed on the combined optimization model for the layout of video surveillance points to determine the step of a target layout plan for the video surveillance points, including: when the number of video surveillance points is not greater than the preset number of video surveillance points, and the overlap between adjacent video surveillance points is greater than the preset minimum overlap, the layout plan with the largest sum of the effective coverage areas of each video surveillance point is determined as the target layout plan.

[0008] In one embodiment, the real-time video image data collected by each video surveillance point at the same time node, the geographical location coordinate information corresponding to each video surveillance point, and the spatial distribution map are determined as training samples, and before the training samples are sent to a preset diffusion model for fusion modeling processing, the steps include: performing image correction processing and color smoothing transition processing on the real-time video image data to determine the target real-time video image data; performing feature encoding processing on the geographical location coordinates of the video surveillance points to determine the target geographical location coordinates.

[0009] In one embodiment, the steps of performing image correction processing on real-time video image data and determining target real-time video image data include: performing image correction processing on the original first real-time video image data through a preset perspective transformation algorithm, converting the tilted image in the first real-time video image data into an approximate top view to correct the image deformation caused by the pitch angle, and determining the second real-time video image data; performing color smoothing transition processing on the second real-time video image data through a preset bilinear interpolation reverse mapping algorithm to correct the image deformation that occurs in the first real-time video image data during the image correction processing, and determining the target real-time video image data.

[0010] In one embodiment, feature encoding processing is performed on the geographic location coordinates of the video surveillance point to determine the target geographic location coordinates, including: based on sine waves and cosine waves of different frequencies, coordinate normalization processing is performed on the geographic location coordinates, and periodic function encoding processing is performed on the normalized geographic location coordinates to convert the geographic location coordinates from two-dimensional coordinates to four-dimensional vectors to obtain the target geographic location coordinates.

[0011] In one embodiment, spectral extraction processing is performed on multispectral remote sensing image data to extract spectral classification features of the multispectral remote sensing image data, and a spatial distribution map of cyanobacteria remote sensing inversion is determined according to the spectral classification features through a preset cyanobacteria inversion model, including: performing image screening processing on the multispectral remote sensing image data to determine target image data whose cloud coverage is less than a preset coverage threshold; using the response characteristics of cyanobacteria in different spectral bands, performing spectral extraction processing on the target image data to extract a target spectral index, wherein the target spectral index includes: a normalized water body index, an algae index, and a phytoplankton index, and the target spectral index is used to reflect the spectral classification features related to cyanobacteria; and converting the target spectral index into a spatial distribution map of cyanobacteria density through a preset cyanobacteria inversion model.

[0012] In the second aspect, an embodiment of the present invention also provides a system for generating a map of cyanobacteria distribution based on lake video monitoring, and the system for generating a map of cyanobacteria distribution includes: a video monitoring subsystem, a remote sensing image processing subsystem and a data fusion modeling subsystem; wherein the video monitoring subsystem is used to collect real-time video image data using each video monitoring point after determining the layout plan of multiple video monitoring points around the lake, and obtain the geographical location coordinate information corresponding to the video monitoring point; the remote sensing image processing subsystem is used to obtain multispectral remote sensing image data covering the lake area, and extract the spectral classification characteristics of the multispectral remote sensing image data, so as to calculate the spatial distribution map of cyanobacteria remote sensing inversion according to the spectral classification characteristics; the data fusion modeling subsystem is used to fuse and model the real-time video image data, the geographical location coordinate information corresponding to the video monitoring points and the spatial distribution map to generate a map of cyanobacteria distribution.

[0013] In one embodiment, the video surveillance subsystem also includes: a video feature extraction module and a geographic location coordinate encoding module; wherein the video feature extraction module is used to determine the target real-time video image data by image deformation caused by the pitch angle in the real-time video image data; the geographic location coordinate encoding module is used to perform feature encoding processing on the geographic location coordinates of the video surveillance point to determine the target geographic location coordinates.

[0014] In one embodiment, the data fusion modeling subsystem includes: a preset diffusion model, which includes: an encoder and a decoder; wherein the encoder is used to extract feature information of real-time video image data, geographic location coordinate information, and a spatial distribution map, and send the feature information to the decoder; the decoder is used to gradually denoise the feature information, and use the denoised feature information to accurately reconstruct the distribution of cyanobacteria to generate a map of cyanobacteria distribution.

[0015] The embodiments of the present invention bring the following beneficial effects:

[0016] The embodiment of the present invention provides a method and system for generating a cyanobacteria distribution map based on lake video monitoring. The method first obtains multispectral remote sensing image data covering the lake area, and obtains real-time video image data collected by each video monitoring point and geographical location coordinate information corresponding to each video monitoring point through multiple video monitoring points set around the lake. Then, spectral extraction processing is performed on the multispectral remote sensing image data to extract the spectral classification characteristics of the multispectral remote sensing image data, and a spatial distribution map of cyanobacteria remote sensing inversion is determined according to the spectral classification characteristics through a preset cyanobacteria inversion model. Finally, the real-time video image data collected by each video monitoring point at the same time node, the geographical location coordinate information corresponding to each video monitoring point, and the spatial distribution map are determined as training samples, and the training samples are sent to a preset diffusion model for fusion modeling processing to generate a cyanobacteria distribution map. The embodiment of the present invention can significantly improve the spatiotemporal resolution and image quality of the cyanobacteria distribution map.

[0017] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description, claims and drawings.

[0018] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0020] Figure 1 A schematic diagram of a flow chart of a method for generating a map of cyanobacteria distribution based on lake video monitoring provided by an embodiment of the present invention;

[0021] Figure 2 A schematic diagram of a flow chart of a method for determining a target deployment scheme provided by an embodiment of the present invention;

[0022] Figure 3 A schematic diagram of the structure of a system for generating a map of cyanobacteria distribution based on lake video monitoring provided by an embodiment of the present invention;

[0023] Figure 4 A schematic diagram of the structure of a video monitoring subsystem provided by an embodiment of the present invention;

[0024] Figure 5 A schematic diagram of the structure of a remote sensing image processing subsystem provided by an embodiment of the present invention;

[0025] Figure 6 A structural schematic diagram of a data fusion modeling subsystem provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0026] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution of the present invention will be clearly and completely described in combination with the embodiments below. Obviously, the described embodiments are 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.

[0027] At present, in order to prevent cyanobacteria from destroying the balance of aquatic ecosystems, it is necessary to monitor the distribution range and change trend of cyanobacteria in lakes. The traditional monitoring method is to use remote sensing monitoring, obtain multispectral remote sensing images using satellites or aviation, and classify and extract cyanobacteria based on the spectral response characteristics of different bands. However, the transit period of remote sensing satellites is usually several days to several weeks. Even if MODIS remote sensing images with a time resolution of 1 day are used, due to the obstruction of clouds, the actual unobstructed high-quality remote sensing images obtained are usually more than 1 week apart. The cyanobacteria outbreak process often changes significantly within a few days, and it is difficult to capture it in time by remote sensing images alone. Therefore, the time resolution of this method is low, and it is impossible to monitor the dynamic changes of cyanobacteria in real time and continuously. In addition, remote sensing monitoring is greatly affected by weather. When clouds, haze and other weather conditions occur, the quality of remote sensing images will be seriously affected, resulting in missing data in some areas or inability to correctly interpret them. In addition, video surveillance systems can be deployed around lakes to identify cyanobacteria aggregation areas by analyzing video images, but single video surveillance cannot cover the entire lake and cannot provide regulatory agencies with a map of the spatial distribution of cyanobacteria in geographic spatial coordinates.

[0028] To sum up, there is a need for a technical method that can comprehensively utilize multiple video surveillance stations, which can dynamically and accurately generate a map of the distribution of cyanobacteria in geographic spatial coordinates to make up for the shortcomings of a single data source. Based on this, the present invention implements a method and system for generating a map of cyanobacteria distribution based on lake video surveillance, which can integrate two complementary data sources: video surveillance, geographic location coordinate information of video surveillance points, and remote sensing images. It overcomes the defects and shortcomings of traditional single ground video surveillance technology and single remote sensing image monitoring technology, and achieves the simultaneous improvement of the spatiotemporal resolution and accuracy of drawing a map of cyanobacteria distribution in geographic spatial coordinates.

[0029] See also Figure 1 The flowchart of a method for generating a blue algae distribution map based on lake video monitoring is shown, and the method mainly includes the following steps S102 to S106:

[0030] Step S102, obtaining multispectral remote sensing image data covering the lake area, and obtaining real-time video image data collected by each video monitoring point and the geographical location coordinate information corresponding to each video monitoring point through multiple video monitoring points set around the lake, wherein the multispectral remote sensing image data comes from different satellite sensors, such as Landsat, Sentinel, MODIS, etc. In addition to multispectral remote sensing image data, high-resolution satellite images, images taken by drones or other types of remote sensing data can also be used. For the task of cyanobacteria monitoring, it is necessary to select image data with sufficiently high spatial resolution (such as 30 meters or higher) and appropriate spectral resolution (including cyanobacteria sensitive bands), and after obtaining the original remote sensing image, preprocess the data so that the preprocessed image data provides a reliable basis for subsequent analysis.

[0031] In one embodiment, the layout plan of video surveillance points requires the pre-construction of a three-dimensional model of lake monitoring layout for iterative calculation to determine the target layout plan. The three-dimensional model of lake monitoring layout integrates the digital elevation model, the lake water range model and the obstruction model, which provides a basis for subsequent field of view analysis and layout optimization. The comprehensive three-dimensional model can fully reflect the geographical characteristics of the lake and its surrounding environment, and is the key to achieving accurate monitoring layout.

[0032] Step S104, spectral extraction processing is performed on the multispectral remote sensing image data to extract the spectral classification characteristics of the multispectral remote sensing image data, and the spatial distribution map of the cyanobacteria remote sensing inversion is determined according to the spectral classification characteristics by presetting the cyanobacteria inversion model. In one embodiment, after obtaining the multispectral remote sensing image data covering the lake area, high-quality remote sensing images with cloud coverage of less than 10-30% can be screened out, and the response characteristics of cyanobacteria in different spectral bands in the water area are used to extract multiple spectral indexes (such as normalized water body offline index NDWI, etc.) to invert the spatial distribution of cyanobacteria. In another embodiment, it is also necessary to obtain the shooting time of the high-quality historical remote sensing image, and obtain the video image data collected from different video monitoring points at the same time according to the shooting time of the historical remote sensing image. Since the time zone of the remote sensing image is usually inconsistent with the time zone of the video monitoring point, it is necessary to convert the time.

[0033] Step S106, the real-time video image data collected by each video surveillance point at the same time node, the geographical location coordinate information corresponding to each video surveillance point, and the spatial distribution map are determined as training samples, and the training samples are sent to a preset diffusion model for fusion modeling processing to generate a cyanobacteria distribution map. In one embodiment, images of multiple video surveillance points that are consistent with the multispectral remote sensing image data time can be selected, and the real-time video image data of the same time and different stations and the geographical location coordinate information of the video surveillance points are input into the diffusion model to fit the above-mentioned cyanobacteria remote sensing inversion image, and through fusion modeling, the video features and the geographical location coordinate information of the video surveillance points are used as input, and finally the two-dimensional scene in the video image is projected onto a cyanobacteria distribution map under geographic space coordinates.

[0034] The method for generating a cyanobacteria distribution map based on lake video surveillance provided in an embodiment of the present invention makes full use of the time dimension information of video data to realize dynamic tracking of cyanobacteria distribution, and generates a distribution map of cyanobacteria in geographic space coordinates (a spatiotemporal continuous image of cyanobacteria distribution), thereby improving the spatiotemporal resolution and prediction accuracy.

[0035] The embodiment of the present invention further provides an implementation method for acquiring real-time video image data and processing the real-time video image data, for details, see (1) to (3) below:

[0036] (1) First, by using a preset ray tracing method, the preset lake monitoring layout three-dimensional model and the field of view area of ​​the video monitoring point are interactively processed to determine the binary mask map of the effective field of view area of ​​each video monitoring point, so as to construct a video monitoring point layout combination optimization model using the binary mask map, the preset number of video monitoring points and the video monitoring point setting information, and then iteratively process the video monitoring point layout combination optimization model to determine the target layout plan of the video monitoring points, so as to set the video monitoring points around the lake using the target layout plan, wherein the preset lake monitoring layout three-dimensional model includes: a digital elevation model, a lake water range model and an obstruction model, the binary mask map includes: a visible area and an obstruction area, and the video monitoring point setting information includes: the position, height and viewing angle of the video monitoring point. In one embodiment, when the number of video monitoring points is not greater than the preset number of video monitoring points and the overlap between adjacent video monitoring points is greater than the preset minimum overlap, the layout plan with the largest sum of the effective coverage areas of each video monitoring point is determined as the target layout plan.

[0037] In one embodiment, see Figure 2 The flowchart of a method for determining a target layout scheme is shown in FIG. 1 . The embodiment of the present invention further provides an implementation method for determining a target layout scheme. For details, see the following steps S202 to S208:

[0038] Step S202, establish a three-dimensional terrain model of the lake. For accurate calculation and simulation, it is first necessary to obtain the three-dimensional terrain data of the study area and establish a digital elevation model. The digital elevation model is implemented using existing remote sensing measurement technology and geographic information systems. In addition, it is necessary to mark the water range of the lake to establish a water range model of the lake, and mark the location and height of obstacles such as buildings and vegetation that may block the line of sight to establish an obstruction model. Based on the above three models, the complete three-dimensional model of the lake monitoring layout can be determined.

[0039] Step S204, theoretical visible area calculation, for each pending video surveillance point position, it is necessary to calculate its field of view coverage, that is, the effective area where the lake can be seen. Specifically, it is necessary to first set the initial height of the video surveillance point, and then launch a large number of rays from the video surveillance point position in the horizontal direction, and interact with the three-dimensional scene model for detection. If the rays are not blocked by obstacles, they can directly reach the lake surface, and then the pixels in this direction belong to the effective field of view. In this way, a mask map of the effective field of view can be generated at each video surveillance point position, where pixels with a value of 1 represent visible areas, and pixels with a value of 0 represent blocked areas.

[0040] For a given video surveillance point The present invention uses ray tracing to determine the effective area of ​​its field of view and sets the initial height After that, the present invention takes P as the center in the horizontal plane, and the angle range is , with step length Emit rays, each ray can be expressed as a parametric equation:

[0041]

[0042] in, is the ray length parameter, is the angle of the ray in the horizontal plane.

[0043] For each ray, the present invention increases value to detect its intersection with each element in the 3D scene model. represents any point in the scene, then the intersection of the ray and the scene satisfies the equation:

[0044]

[0045] At this time, it is necessary to check whether the ray intersects with the obstruction: if there is an intersection, the direction is blocked; if there is no intersection, continue to check the intersection of the ray and the lake surface, and the lake surface can be simplified to a plane. ,in, is the water surface elevation, and the intersection point of the ray and the lake surface can be obtained by solving the equation:

[0046]

[0047] if ,but has finite solutions; if (The ray is parallel to the lake surface), then there is no intersection. , the ray can intersect the lake surface downwards. If there is a finite positive value, the lake surface in this direction is visible. Finally, the present invention can generate a corresponding binary mask image. This method takes into account the geometric characteristics of the terrain, obstructions and the lake surface, and can accurately simulate the actual visible range of the video surveillance point.

[0048] Step S206, traverse the candidate monitoring point positions, select the optimal combination to meet the coverage and overlap requirements, and find a set of video monitoring point layout plans among the optional candidate positions so that the plan can cover the largest possible lake area in the effective field of view while maintaining a certain degree of overlap between adjacent video monitoring points, thereby modeling it as a combinatorial optimization problem: the decision variables are the location set, installation height and viewing angle of the video monitoring points, the objective function is the total effective coverage area corresponding to these positions, and the constraints include the minimum overlap between adjacent video monitoring points and the upper limit of the total number of video monitoring points.

[0049] Furthermore, by randomly adjusting the position of the video surveillance point, the objective function value of the new solution is calculated. If the objective function value of the new solution is better or meets certain probability conditions, the position update is accepted. Through a large number of iterations, the algorithm will gradually converge to a better solution.

[0050] set up Represents a set of N candidate monitoring point locations, each of which Represents the location coordinates of a monitoring point And installation direction The decision variables are subsets , represents the final selected set of monitoring points.

[0051] Define the objective function is the total effective coverage area of ​​the selected monitoring points:

[0052]

[0053] in, Indicates monitoring point effective field of view.

[0054] The constraints can be expressed as the upper limit of the number of monitoring points and the minimum overlap of adjacent monitoring points.

[0055]

[0056] in, is the maximum number of monitoring points allowed. The minimum overlap between adjacent monitoring points:

[0057]

[0058] in, is the required minimum overlap ratio. Therefore, this combinatorial optimization problem can be formalized as:

[0059] ,when and

[0060] in, , .

[0061] Finally, through repeated iterations and adjustments, the algorithm can gradually converge to an optimal monitoring point layout plan, maximizing the effective coverage area of ​​the lake while satisfying the coverage overlap and quantity restrictions.

[0062] Step S208, on-site survey to determine the actual location. Due to the complexity of the on-site conditions around the lake, it is necessary to verify the feasibility of the target layout plan through fine-tuning and on-site survey.

[0063] (2) Perform feature encoding processing on the geographic location coordinates of the video surveillance point to determine the target geographic location coordinates. Based on sine waves and cosine waves of different frequencies, perform coordinate normalization processing on the geographic location coordinates, and perform periodic function encoding processing on the normalized geographic location coordinates to convert the geographic location coordinates from two-dimensional coordinates to four-dimensional vectors to obtain the target geographic location coordinates. For details, see the following steps (A) to (B):

[0064] (A) Constructing the video surveillance point position tensor: First, the present invention needs to convert the geographic coordinates of all video surveillance points into two-dimensional coordinates in a unified coordinate system and normalize them to the range of [0, 1]. Assuming there are N video surveillance points, and the coordinates of each video surveillance point are (x, y), a position tensor with a shape of (N, 2) is constructed.

[0065] (B) Adding position coding: In the design of this cyanobacteria monitoring point coding system, the present invention proposes a variety of coding schemes to process two-dimensional coordinate systems, such as longitude and latitude. The position coding is represented by PE, where x and y represent the normalized longitude and latitude values, respectively. By introducing the coding method using sine and cosine functions, it can be expressed as:

[0066]

[0067]

[0068] In this formula, x and y are still the normalized longitude and latitude values, i is the dimension index, and d is the encoding dimension of each coordinate. By using sine and cosine waves of different frequencies, the original values ​​are mapped to the range of [-1, 1], allowing the model to learn spatial relationships of different scales and help deal with periodic or annular spatial relationships.

[0069] In addition, considering the characteristics of two-dimensional coordinates, the present invention also proposes an optimized four-dimensional encoding scheme. For longitude X:

[0070]

[0071]

[0072] For latitude y:

[0073]

[0074]

[0075] Therefore, the use of four-dimensional vectors not only fully preserves the information of the original two-dimensional coordinates, but also provides a nonlinear representation, which helps the neural network to better learn spatial relationships. The position encoding formula aims to generate a unique vector representation for each position in the sequence, and uses the periodic characteristics of sine and cosine functions to encode position information in different dimensions. Sine functions are used in even dimensions and cosine functions are used in odd dimensions. This alternating use allows each position to obtain a unique encoding.

[0076] In addition, the formula The term increases with the increase of dimension index i, resulting in lower dimensions with lower frequencies and slower changes, while higher dimensions with higher frequencies and faster changes. This design enables the encoding to capture position information of different scales at the same time, which can represent subtle differences between adjacent positions and distinguish significant differences between distant positions. In addition, since all encoding values ​​are between -1 and 1, they can be well combined with other normalized embedding vectors (such as word embeddings).

[0077] Through the above encoding method, the encoding vectors of adjacent positions are very similar in some dimensions, while the positions farther away are different in more dimensions, which helps the model learn relative position relationships. Periodic functions in different dimensions have different frequencies and can capture position information of different scales. The encoding vector is fixed and does not need to be learned, so the number of parameters of the model can be reduced. The encoding vector can be easily extended to sequences of any length without retraining the model.

[0078] In summary, sine / cosine position coding generates a unique coding vector for each position in the sequence by cleverly designing a periodic function, which helps the model understand the relative position relationship between elements and improves the performance of the model in sequence modeling tasks. Through the above steps, the present invention successfully encodes the geographical location coordinate information of the video surveillance point into the feature expression of the model. The spatial information coding module captures the relative position relationship between the video surveillance points, which provides effective prior knowledge for the subsequent cyanobacteria distribution modeling. By introducing the spatial information coding module, the geographical location coordinates of the video surveillance points are successfully integrated into the modeling process of cyanobacteria distribution mapping.

[0079] (3) Performing image correction processing and color smoothing transition processing on the real-time video image data to determine the target real-time video image data. Specifically, the image correction processing can be performed on the original first real-time video image data through a preset perspective transformation algorithm to convert the tilted image in the first real-time video image data into an approximate top view to correct the image deformation caused by the pitch angle to determine the second real-time video image data. Then, a color smoothing transition processing is performed on the second real-time video image data through a preset bilinear interpolation reverse mapping algorithm to correct the image deformation of the first real-time video image data that occurs during the image correction processing to determine the target real-time video image data. Specifically:

[0080] First, in the training phase, real-time video image data needs to be input into the diffusion model. A decoder process is created, and the input data format is specified as an H.264-encoded video stream, and the output format is the original RGB image frame. H.264 is an efficient video compression standard that is widely used in video streaming and storage. The decoding process uses the FFmpeg library, which provides powerful multimedia processing capabilities and can efficiently decode video streams in various formats.

[0081] Furthermore, it is necessary to use a perspective transformation algorithm to correct the image deformation caused by the pitch angle and convert the tilted image into an approximate top view, which is easier to compare and fuse with remote sensing images. The image correction method based on perspective transformation solves the problem of image deformation caused by the pitch angle. This method can not only convert the tilted image into an approximate top view, but also significantly improve the comparison and fusion effect with remote sensing images. The perspective transformation is implemented through a 3x3 transformation matrix H, which defines the mapping relationship from the source image plane to the target image plane. The transformation process is expressed by the following equation:

[0082]

[0083] Where (x, y) is the coordinate in the source image, and (x', y') is the corresponding coordinate in the target image after transformation. The final target coordinate is obtained by homogeneous division:

[0084]

[0085] Where w is a scaling factor in a homogeneous coordinate system, used to perform perspective division. This step is the key to the perspective transformation, which allows the present invention to correctly project from three-dimensional space to a two-dimensional image plane.

[0086] In addition, considering the pitch angle, the present invention constructs a specific transformation matrix H. Assuming that the intrinsic parameter matrix of the camera is K and the rotation matrix is ​​R (determined by the pitch angle), the transformation matrix H is expressed as:

[0087]

[0088] Among them, taking the pitch angle of 20 degrees as an example, K and R are expressed as:

[0089]

[0090]

[0091] in, and is the focal length of the camera, is the coordinate of the principal point of the image. The principal point of the image is the intersection of the optical axis and the image plane. Ideally, it should be located at the center of the image, but in actual camera systems, due to manufacturing errors and lens distortion, the principal point may be slightly off-center. and Usually close to half the width and half the height of the image, that is, is the coordinate of the principal point of the image, reflecting the optical center of the camera; the transformation matrix H is related to the focal length because it contains the camera internal parameter information; and w is the scaling factor used for perspective division, which is not directly related to the principal point coordinates, but plays a key role in projecting a three-dimensional point onto a two-dimensional plane. These concepts together constitute the theoretical basis of perspective transformation, enabling the present invention to accurately correct image deformation caused by pitch angle.

[0092] In addition, in order to achieve high-precision image correction, the present invention adopts a reverse mapping algorithm based on bilinear interpolation. For each pixel (X, Y) in the target image, the present invention uses Calculate its corresponding position (x, y) in the source image:

[0093]

[0094] Then, the present invention uses bilinear interpolation to calculate the color value of the target pixel:

[0095]

[0096] in, and are integer coordinates around (x, y) and a and b are interpolation weights.

[0097] In one embodiment, implementing bilinear interpolation in RGB color space is a key step in correcting image deformation caused by pitch angle. The application of bilinear interpolation in RGB color model needs to consider three color channels of each pixel: red (R), green (G) and blue (B). The present invention proposes an efficient and accurate method that can process all color channels simultaneously to ensure the continuity and accuracy of color information during image deformation correction.

[0098] For each color channel c ∈ {R, G, B}, the present invention expands the bilinear interpolation formula as follows:

[0099]

[0100] Among them, (X, Y) represents the coordinates in the target image, and (x, y) is the inverse transformation matrix The corresponding floating-point coordinates in the resulting source image. and are integer coordinates around (x, y), and is the interpolation weight, Represents the intensity value of color channel c. By applying the interpolation formula to each color channel separately, the results are finally combined into RGB color values:

[0101]

[0102] The method ensures a smooth transition of colors and reduces the sawtooth effect and color distortion caused by discrete sampling. The present invention uses vectorized operations to optimize computing efficiency and uses the NumPy library to process all color channels simultaneously to improve computing speed. The color interpolation method plays a key role in the cyanobacteria monitoring system, maintains image quality, and provides a reliable data basis for cyanobacteria distribution analysis. The present invention converts oblique images into approximate bird's-eye views, improves the contrast and fusion effects with remote sensing images, and enhances the water environment monitoring capability. The perspective transformation method corrects image deformation caused by the pitch angle and maintains the geometric characteristics and texture details of the image.

[0103] Finally, all processed image frames are stored in a NumPy array and returned as the input of the diffusion model. Specifically, real-time video image data usually exists in the form of a continuous frame sequence, and each frame is an RGB image. In order to facilitate the training of the model, these image frames need to be preprocessed into a format suitable for input, such as scaling the image to a fixed size (such as 256x256) and normalizing the pixel value to the range of [0, 1]. This ensures that images between different video sources have a unified scale and value range, which is conducive to the convergence of the model.

[0104] In addition, the OpenCV library is used to read the video frame and perform preprocessing, including the specific steps of color space conversion, size scaling and pixel value normalization as follows: use cv2.VideoCapture() to open the video file; loop to read each frame of the image, using the cv2.imread() function; convert the BGR color space to RGB; use cv2.resize() to scale the image to a fixed size of 256x256; normalize the pixel value from [0, 255] to [0, 1] by dividing all pixel values ​​by 255.0; add the processed frame to the NumPy array; after the preprocessing is completed, the present invention obtains a four-dimensional NumPy array with a shape of (num_frames, height, width, channels), where num_frames is the total number of frames of the video, height and width are both 256, channels is 3 (corresponding to the three RGB channels), and this array can be directly used as the input of the diffusion model.

[0105] In one embodiment, an embodiment of the present invention further provides an implementation method for spectral remote sensing image processing, the process mainly includes several key steps of data acquisition, preprocessing, feature extraction and cyanobacteria inversion. First, image screening processing is performed on multispectral remote sensing image data to determine target image data whose cloud coverage is less than a preset coverage threshold. Then, the target image data is spectrally extracted using the response characteristics of cyanobacteria in different spectral bands to extract the target spectral index. Specifically, since cyanobacteria have unique response patterns in different spectral bands, such as strong absorption in the near-infrared band and high reflectivity in the visible light band, according to these spectral characteristics, the present invention constructs some spectral indexes to enhance cyanobacteria information and suppress interference from other land objects. Commonly used spectral indices include normalized water index (NDWI), algae index (AI), phytoplankton index (FPI), etc. These indices highlight the spectral response of cyanobacteria through mathematical operations between bands, such as difference, ratio or normalization. For example, NDWI uses the reflectivity difference between the green light band and the near-infrared band to effectively distinguish between water and land, and highlight the algae information in the water.

[0106] Furthermore, it is necessary to convert the target spectral index into a spatial distribution map of cyanobacteria density through a preset cyanobacteria inversion model, wherein the target spectral index includes: a normalized water body index, an algae index and a phytoplankton index. The target spectral index is used to reflect the spectral classification characteristics related to cyanobacteria. The preset cyanobacteria inversion model can link the spectral index with the cyanobacteria density or biomass. Through the cyanobacteria inversion model, the present invention converts the spectral index into a spatial distribution map of cyanobacteria density or biomass. Specifically, the spectral index value of each pixel corresponds to a predicted cyanobacteria concentration value, forming a spatial distribution map of the entire lake range. The distribution of cyanobacteria. In order to evaluate the reliability of the inversion results, the present invention also needs to use measured data to verify and evaluate the accuracy of the model. By collecting water samples from different locations in the lake, measuring indicators such as cyanobacteria density or chlorophyll a concentration, and comparing them with the model prediction values, the correlation coefficient, root mean square error and other accuracy indicators are calculated to quantify the uncertainty of the inversion results. In addition, in addition to obtaining the cyanobacteria distribution information of the current period, the present invention also needs to analyze the dynamic changes of the spatiotemporal distribution of cyanobacteria. By obtaining historical remote sensing image data and repeating the above inversion steps, the spatial distribution maps of cyanobacteria in different periods can be obtained.

[0107] The above-mentioned method for generating a map of cyanobacteria distribution based on lake video monitoring provided by an embodiment of the present invention makes up for the defects of single remote sensing or single video data, utilizes the complementary advantages of the two data sources, improves the spatiotemporal resolution of the map of cyanobacteria distribution, and by introducing the geographic location coordinates of the video sites, the model can well capture the spatial context associations across video sites, and realize data fusion through the diffusion model. The generation method can be widely used in different lakes, reservoirs and other water bodies, providing new technical means for refined environmental monitoring and governance of river basins.

[0108] To facilitate understanding of this embodiment, firstly, a system for generating a map of cyanobacteria distribution based on lake video monitoring disclosed in an embodiment of the present invention is described in detail, see Figure 3 The structure diagram of a system for generating a blue algae distribution map based on lake video monitoring is shown. The system for generating a blue algae distribution map includes: a video monitoring subsystem, a remote sensing image processing subsystem and a data fusion modeling subsystem.

[0109] Among them, the video monitoring subsystem is used to collect real-time video image data using various video monitoring points after determining the layout plan of multiple video monitoring points around the lake, and obtain the geographical location coordinate information corresponding to the video monitoring points. Specifically, the field of view coverage of the video monitoring points can be calculated through ray tracing technology, that is, by generating a mask map of the effective area of ​​the field of view, the effective coverage range of each monitoring point (that is, the effective area of ​​the field of view) is quantitatively characterized, providing accurate evaluation indicators for subsequent optimization. After that, the monitoring point layout problem is modeled as a combinatorial optimization problem, in which the decision variables include the location set, installation height and viewing angle of the monitoring points, and the objective function is the total effective coverage area. At the same time, constraints such as the minimum overlap between adjacent monitoring points and the upper limit of the total number of monitoring points are considered, thereby comprehensively considering various requirements in practical applications, making the optimization results more practical, and solving the combinatorial optimization problem to determine the target layout plan of the monitoring points.

[0110] In one embodiment, see Figure 4 A structural schematic diagram of a video monitoring subsystem is shown, and the video monitoring subsystem also includes: a video feature extraction module and a geographic location coordinate encoding module; wherein the video feature extraction module is used to correct the image deformation caused by the pitch angle in the real-time video image data, and determine the target real-time video image data. Specifically, the image correction algorithm based on perspective transformation is used to correct the image, and the perspective transformation algorithm is used to solve the image deformation problem caused by the pitch angle. By constructing a specific 3x3 transformation matrix, combined with the camera intrinsic parameters and the rotation matrix, the accurate conversion of the inclined image to the approximate top view is achieved, thereby improving the contrast and fusion effect of the video image and the remote sensing image, and providing a more reliable data basis for water environment monitoring. Then, the NumPy array is used as the input data structure of the diffusion model, which realizes efficient memory management and calculation optimization, and by organizing the processed image frames into a four-dimensional array (number of frames, height, width, color channel), a unified and easy-to-process data format is provided for the subsequent deep learning model.

[0111] In addition, the geographic location coordinate encoding module is used to perform feature encoding processing on the geographic location coordinates of the video surveillance points to determine the target geographic location coordinates. Specifically, by proposing an efficient location encoding method suitable for a two-dimensional coordinate system, combined with the basic normalized coordinate representation and periodic function encoding, a four-dimensional vector representation is formed; this encoding scheme not only completely retains the original two-dimensional coordinate information, but also provides a nonlinear representation, which helps the neural network to better learn spatial relationships, thereby being able to simultaneously capture location information of different scales, representing both subtle differences between adjacent locations and significant differences between distant locations.

[0112] The remote sensing image processing subsystem is used to obtain multispectral remote sensing image data covering the lake area and extract the spectral classification features of the multispectral remote sensing image data to calculate the spatial distribution map of cyanobacteria remote sensing inversion according to the spectral classification features. In one embodiment, see Figure 5 The schematic diagram of the structure of a remote sensing image processing subsystem is shown. The remote sensing image processing subsystem mainly includes two parts: extracting the spectral characteristics of multispectral remote sensing image data and determining the spatial distribution map of cyanobacteria density or biomass by remote sensing inversion of cyanobacteria. In addition, it is necessary to preprocess the remote sensing image and evaluate the reliability of the inversion result during the remote sensing inversion of cyanobacteria. By collecting water samples from different locations of the lake, measuring indicators such as cyanobacteria density or chlorophyll a concentration and comparing them with the model prediction value, calculating accuracy indicators such as correlation coefficient and root mean square error to quantify the uncertainty of the inversion result. In addition, in addition to obtaining the cyanobacteria distribution information of the current period, it is also necessary to analyze the dynamic changes of the spatiotemporal distribution of cyanobacteria, and by obtaining historical remote sensing image data, repeat the above inversion steps to obtain the spatial distribution map of cyanobacteria in different periods.

[0113] The data fusion modeling subsystem is used to fuse and model the real-time video image data, the geographic location coordinate information corresponding to the video monitoring point, and the spatial distribution map to generate a map of the cyanobacteria distribution. In one embodiment, see Figure 6 A structural schematic diagram of a data fusion modeling subsystem is shown, the data fusion modeling subsystem includes: a preset diffusion model, the preset diffusion model includes: an encoder and a decoder; wherein the encoder is used to extract real-time video image data, geographic location coordinate information and feature information of a spatial distribution map, and send the feature information to the decoder; the decoder is used to gradually denoise the feature information, and use the denoised feature information to accurately reconstruct the distribution of cyanobacteria to generate a cyanobacteria distribution map. Therefore, when it is necessary to monitor and analyze cyanobacteria in an inland lake, and the lake area is large, a single remote sensing image is difficult to cover the entire picture, and the time resolution of the remote sensing image cannot meet the needs of real-time monitoring, the above system can realize the automatic generation of cyanobacteria distribution maps, and provide technical support for the monitoring and management of cyanobacteria blooms.

[0114] To sum up, the spatial distribution map of cyanobacteria generated by the embodiment of the present invention based on ground video surveillance can continuously collect lake surface data in real time, fully capture the dynamic change process of cyanobacteria outbreaks, thereby greatly improving the temporal resolution. In addition, compared with remote sensing images, the spatial distribution map of cyanobacteria generated by the embodiment of the present invention based on ground video surveillance is less affected by weather and can continue to work in severe weather conditions such as rainy and cloudy weather. In addition, by fusing multiple video surveillance data around the lake to generate a spatial distribution map of cyanobacteria in geographic spatial coordinates, the problem that a single video surveillance cannot cover the entire lake is also solved.

[0115] Finally, it should be noted that the above-described embodiments are only specific implementations of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them. The protection scope of the present invention is not limited thereto. Although the present invention is described in detail with reference to the above-described embodiments, ordinary technicians in the field should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the above-described embodiments within the technical scope disclosed by the present invention, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A method for generating a map of cyanobacteria distribution based on lake video monitoring, characterized in that: The method is applied to a system for generating a map of cyanobacteria distribution based on lake video monitoring, and the method comprises: Acquire multispectral remote sensing image data covering the lake area, and through multiple video monitoring points set up around the lake, obtain real-time video image data collected by each of the video monitoring points and geographical location coordinate information corresponding to each of the video monitoring points; Performing spectral extraction processing on the multispectral remote sensing image data to extract spectral classification features of the multispectral remote sensing image data, and determining a spatial distribution map of cyanobacteria remote sensing inversion according to the spectral classification features by using a preset cyanobacteria inversion model; The real-time video image data collected by each of the video monitoring points at the same time node, the geographical location coordinate information corresponding to each of the video monitoring points, and the spatial distribution map are determined as training samples, and the training samples are sent to a preset diffusion model for fusion modeling processing to generate a map of cyanobacteria distribution; Wherein, before the step of obtaining the real-time video image data collected by each of the video surveillance points and the geographical location coordinate information corresponding to each of the video surveillance points through the multiple video surveillance points set around the lake, it includes: interactively processing the preset lake monitoring layout three-dimensional model and the field of view area of ​​the video surveillance point through a preset ray tracing method, and determining a binary mask map of the effective field of view area of ​​each video surveillance point, so as to construct a video surveillance point layout combination optimization model by using the binary mask map, the preset number of video surveillance points and the video surveillance point setting information, wherein the preset lake monitoring layout three-dimensional model includes: a digital elevation model, a water range model of the lake and an obstruction model, the binary mask map includes: a visible area and an obstruction area, and the video surveillance point setting information includes: the position, height and viewing angle of the video surveillance point; iteratively processing the video surveillance point layout combination optimization model to determine the target layout scheme of the video surveillance points, so as to set the video surveillance points around the lake by using the target layout scheme; The candidate monitoring point positions are traversed to select the optimal combination that meets the coverage and overlap requirements. Among the optional candidate positions, a set of video monitoring point layout schemes are found so that the scheme can maintain the preset overlap between adjacent video monitoring points while maximizing the lake area covered by the effective field of view. The above scheme is modeled as a combinatorial optimization problem, in which the decision variables of the combinatorial optimization problem are the position set, installation height and viewing angle of the video monitoring points. The objective function represents the total effective coverage area corresponding to the layout positions of different video monitoring points. The constraints include the minimum overlap between adjacent video monitoring points and the upper limit of the total number of video monitoring points. The positions of the video monitoring points are randomly adjusted to calculate the objective function value of the new scheme. If the objective function value of the new scheme is better or meets the preset probability conditions, the position update is accepted. Through iterative calculation, the algorithm gradually converges to the target solution. Among them, Represents a set of N candidate monitoring point locations, each Represents the location coordinates of a monitoring point And installation direction , the decision variables are subsets , represents the set of monitoring points finally selected; Define the objective function is the total effective coverage area of ​​the selected monitoring points: in, Indicates monitoring point The effective field of view area; Among them, the constraints are expressed as the upper limit of the number of monitoring points and the minimum overlap of adjacent monitoring points. The upper limit of the number of monitoring points is: in, The maximum number of monitoring points allowed and the minimum overlap of adjacent monitoring points are: in, To obtain the minimum overlap ratio, the combinatorial optimization problem is formalized as: ,when and in, , ; Among them, through iteration and adjustment, the algorithm gradually converges to the target monitoring point layout plan to maximize the effective coverage area of ​​the lake surface under the conditions of meeting the coverage overlap and quantity restrictions; Among them, the step of iteratively processing the combined optimization model for the layout of the video surveillance points to determine the target layout plan of the video surveillance points includes: when the number of the video surveillance points is not greater than the preset number of video surveillance points, and the overlap between adjacent video surveillance points is greater than the preset minimum overlap, the layout plan with the largest sum of the effective coverage areas of each of the video surveillance points is determined as the target layout plan.

2. The method for generating a blue algae distribution map based on lake video monitoring according to claim 1, characterized in that: Before the step of determining the real-time video image data collected by each of the video surveillance points at the same time node, the geographical location coordinate information corresponding to each of the video surveillance points, and the spatial distribution map as training samples, and sending the training samples to a preset diffusion model for fusion modeling processing, the step includes: Performing image correction processing and color smoothing transition processing on the real-time video image data to determine target real-time video image data; The geographic location coordinates of the video surveillance point are subjected to feature coding processing to determine the target geographic location coordinates.

3. The method for generating a blue algae distribution map based on lake video monitoring according to claim 2 is characterized in that: The step of performing image correction processing on the real-time video image data to determine target real-time video image data comprises: Performing image correction processing on the original first real-time video image data by using a preset perspective transformation algorithm, converting the tilted image in the first real-time video image data into an approximate top view to correct the image deformation caused by the pitch angle, and determining the second real-time video image data; The second real-time video image data is subjected to color smoothing transition processing by using a preset bilinear interpolation reverse mapping algorithm to correct image deformation occurring during image correction processing of the first real-time video image data, thereby determining the target real-time video image data.

4. The method for generating a blue algae distribution map based on lake video monitoring according to claim 2, characterized in that: The step of performing feature coding processing on the geographic location coordinates of the video surveillance point to determine the target geographic location coordinates includes: Based on sine waves and cosine waves of different frequencies, the geographic location coordinates are normalized, and the normalized geographic location coordinates are encoded with a periodic function so that the geographic location coordinates are converted from two-dimensional coordinates to four-dimensional vectors to obtain the target geographic location coordinates.

5. The method for generating a blue algae distribution map based on lake video monitoring according to claim 1, characterized in that: The step of performing spectral extraction processing on the multispectral remote sensing image data to extract spectral classification features of the multispectral remote sensing image data, and determining a spatial distribution map of cyanobacteria remote sensing inversion according to the spectral classification features by presetting a cyanobacteria inversion model comprises: Performing image screening processing on the multispectral remote sensing image data to determine target image data that is not covered by clouds with a coverage degree less than a preset coverage threshold; Using the response characteristics of cyanobacteria in different spectral bands, the target image data is subjected to spectral extraction processing to extract a target spectral index, wherein the target spectral index includes: a normalized water body index, an algae index and a phytoplankton index, and the target spectral index is used to reflect the spectral classification characteristics related to cyanobacteria; The target spectral index is converted into a spatial distribution map of cyanobacteria density by presetting a cyanobacteria inversion model.

6. A system for generating a map of cyanobacteria distribution based on lake video monitoring, characterized in that: The blue algae distribution map generation system includes: a video monitoring subsystem, a remote sensing image processing subsystem and a data fusion modeling subsystem; wherein, The video monitoring subsystem is used to collect real-time video image data using each of the video monitoring points after determining the layout plan of multiple video monitoring points around the lake, and obtain the geographical location coordinate information corresponding to the video monitoring points; The remote sensing image processing subsystem is used to obtain multispectral remote sensing image data covering the lake area, and extract spectral classification features of the multispectral remote sensing image data, so as to calculate the spatial distribution map of cyanobacteria remote sensing inversion according to the spectral classification features; The data fusion modeling subsystem is used to perform fusion modeling processing on the real-time video image data, the geographic location coordinate information corresponding to the video monitoring point and the spatial distribution map to generate a map of cyanobacteria distribution; Wherein, before the step of obtaining the geographic location coordinate information corresponding to the video surveillance point, it includes: interactively processing the preset lake monitoring layout three-dimensional model and the field of view area of ​​the video surveillance point through a preset ray tracing method, determining a binary mask map of the effective field of view area of ​​each video surveillance point, so as to construct a video surveillance point layout combination optimization model using the binary mask map, the preset number of video surveillance points and the video surveillance point setting information, wherein the preset lake monitoring layout three-dimensional model includes: a digital elevation model, a water range model of the lake and an obstruction model, the binary mask map includes: a visible area and an obstruction area, and the video surveillance point setting information includes: the position, height and viewing angle of the video surveillance point; iteratively processing the video surveillance point layout combination optimization model to determine the target layout plan of the video surveillance points, so as to use the target layout plan to set video surveillance points around the lake; The candidate monitoring point positions are traversed to select the optimal combination that meets the coverage and overlap requirements. Among the optional candidate positions, a set of video monitoring point layout schemes are found so that the scheme can maintain the preset overlap between adjacent video monitoring points while maximizing the lake area covered by the effective field of view. The above scheme is modeled as a combinatorial optimization problem, in which the decision variables of the combinatorial optimization problem are the position set, installation height and viewing angle of the video monitoring points. The objective function represents the total effective coverage area corresponding to the layout positions of different video monitoring points. The constraints include the minimum overlap between adjacent video monitoring points and the upper limit of the total number of video monitoring points. The positions of the video monitoring points are randomly adjusted to calculate the objective function value of the new scheme. If the objective function value of the new scheme is better or meets the preset probability conditions, the position update is accepted. Through iterative calculation, the algorithm gradually converges to the target solution. Among them, Represents a set of N candidate monitoring point locations, each Represents the location coordinates of a monitoring point And installation direction , the decision variables are subsets , represents the set of monitoring points finally selected; Define the objective function is the total effective coverage area of ​​the selected monitoring points: in, Indicates monitoring point The effective field of view area; Among them, the constraints are expressed as the upper limit of the number of monitoring points and the minimum overlap of adjacent monitoring points. The upper limit of the number of monitoring points is: in, The maximum number of monitoring points allowed and the minimum overlap of adjacent monitoring points are: in, To obtain the minimum overlap ratio, the combinatorial optimization problem is formalized as: ,when and in, , ; Among them, through iteration and adjustment, the algorithm gradually converges to the target monitoring point layout plan to maximize the effective coverage area of ​​the lake surface under the conditions of meeting the coverage overlap and quantity restrictions; Among them, the step of iteratively processing the combined optimization model for the layout of the video surveillance points to determine the target layout plan of the video surveillance points includes: when the number of the video surveillance points is not greater than the preset number of video surveillance points, and the overlap between adjacent video surveillance points is greater than the preset minimum overlap, the layout plan with the largest sum of the effective coverage areas of each of the video surveillance points is determined as the target layout plan.

7. The system for generating a blue algae distribution map based on lake video monitoring according to claim 6, characterized in that: The video monitoring subsystem also includes: a video feature extraction module and a geographic location coordinate encoding module; wherein, The video feature extraction module is used to correct the image deformation caused by the pitch angle in the real-time video image data to determine the target real-time video image data; The geographic location coordinate encoding module is used to perform feature encoding processing on the geographic location coordinates of the video surveillance point to determine the target geographic location coordinates.

8. The system for generating a blue algae distribution map based on lake video monitoring according to claim 6, characterized in that: The data fusion modeling subsystem includes: a preset diffusion model, and the preset diffusion model includes: an encoder and a decoder; wherein, The encoder is used to extract the real-time video image data, the geographic location coordinate information and the feature information of the spatial distribution map, and send the feature information to the decoder; The decoder is used to perform step-by-step denoising on the characteristic information, and use the denoised characteristic information to accurately reconstruct the distribution of cyanobacteria to generate a map of the cyanobacteria distribution.

Images