A method and system for dynamic water monitoring based on UAV remote sensing technology

Abstract

The application discloses a kind of water area dynamic monitoring method and system based on unmanned aerial vehicle remote sensing technology, which comprises: adjusting the shooting angle of each camera according to the flight attitude of unmanned aerial vehicle, and shooting target water area photo after stabilizing;Water surface reflection compensation is carried out on the target water area photo based on the shooting angle;The photos after compensation are sorted according to shooting time, and the photos of different shooting angles under the same shooting time are registered and spliced to obtain a panoramic image sequence;The target water area is divided into intervals, and the interval image sequence corresponding to each water area interval is extracted in the panoramic image sequence, the convolution feature of each interval image in the interval image sequence is extracted, the convolution feature is spliced with the shooting time of the panoramic image where the interval image is located and the center position of the corresponding water area interval to obtain a multi-dimensional feature vector;A water quality inversion model is constructed based on the multi-dimensional feature vector.The water quality inversion model constructed by the application can realize water area dynamic monitoring and more efficiently and accurately release early warning information.

Description

Technical Field

[0001] This invention relates to the field of water area monitoring technology, and in particular, to a method and system for dynamic water area monitoring based on unmanned aerial vehicle (UAV) remote sensing technology. Background Technology

[0002] Water remote sensing technology, with its non-contact and wide-coverage characteristics, has become a core means of dynamic water monitoring. It constructs inversion models by capturing the relationship between water quality parameters and the spectral response of remote sensing bands to achieve quantitative analysis of the aquatic environment. Current mainstream remote sensing technologies include multispectral / hyperspectral imaging, thermal infrared sensors, lidar, and synthetic aperture radar, among which: multispectral / Hyperspectral imaging utilizes a combination of blue (450-520nm), green (520-590nm), and infrared (760-1100nm) bands to accurately identify water pollution types, algal community distribution, and suspended sediment concentration, providing fundamental spectral data for water quality parameter inversion. Thermal infrared sensors monitor the radiation energy in the 9-12μm thermal infrared band to achieve dynamic monitoring of the water temperature field, offering irreplaceable advantages in tracking the spread of thermal pollution from power plants and analyzing nearshore upwelling distribution. LiDAR uses 1064nm / 532nm wavelength laser pulses to penetrate water bodies, and combined with echo signal analysis, can obtain centimeter-level accuracy water surface elevation models and underwater topographic contours, supporting the monitoring of water level changes and riverbed evolution. Synthetic aperture radar uses C / X / L band microwaves to penetrate optical obstacles such as clouds and rainfall, achieving all-weather imaging, playing a crucial role, especially in monitoring the extent of typhoon-induced floods and assessing navigation conditions in ice-covered waterways.

[0003] However, although the aforementioned remote sensing technologies have a certain ability to monitor water areas, the complex optical characteristics and dynamic environmental changes of natural water bodies lead to the following shortcomings in their practical applications:

[0004] 1. Water surface reflection interference bias: Specular reflection caused by sunlight often obscures the true information of water bodies, leading to deviations in water quality parameter inversion. Specifically, as a natural reflective interface, the water surface is prone to specular reflection (BRDF effect) when the solar altitude angle is less than 60°, resulting in strong noise signals in the 350-800nm ​​visible light band. Taking multispectral imaging as an example, when the water surface reflectivity exceeds 20%, the inversion error regarding chlorophyll a concentration will increase by 30-50%, and the inversion deviation rate regarding suspended sediment content can reach more than 40%. This interference has a more significant impact on hyperspectral technology that relies on narrow band fine interpretation, causing the R² of the water quality parameter inversion model to generally decrease by 0.15-0.25.

[0005] 2. Insufficient accuracy in complex scenes: Turbid water is difficult to distinguish from shadows, and the detection capability for small pollutants is insufficient, which limits the improvement of monitoring accuracy. For example, in waters with high sediment content, such as the Shanxi section of the Yellow River, the backscattering coefficient of the water increases dramatically, causing the reflectance of the near-infrared band above 650nm to overlap with the spectral characteristics of the bridge / vegetation shadow area (difference <10%), and the misjudgment rate of traditional threshold segmentation algorithms is as high as 25%. In addition, for floating objects with a particle size <5cm (such as plastic fragments, oil film patches, etc.), although existing centimeter-level resolution sensors can detect their existence, the multispectral mixing pixel effect will lead to incomplete spectral feature extraction. The detection accuracy based on a single band is only 60-70%, which is difficult to meet the accurate identification needs of emergency monitoring. Summary of the Invention

[0006] The purpose of this invention is to provide an efficient, real-time and accurate dynamic monitoring solution for water bodies. By deploying a multi-angle imaging system and compensating for water surface reflection interference from multiple perspectives, the interference of light noise on water body information is effectively suppressed. Furthermore, by utilizing multi-camera data fusion technology, a water quality inversion model that can realize spatiotemporal dynamic monitoring of water bodies is constructed, thereby improving the ability to identify pollutants in turbid water bodies and complex scenarios.

[0007] To achieve the above objectives, this invention discloses a method for dynamic monitoring of water areas based on unmanned aerial vehicle (UAV) remote sensing technology, comprising the following steps:

[0008] S1. Equip the drone with at least two cameras with different shooting angles, adjust each camera to the target shooting angle according to the drone's flight attitude, and take photos of the target water area after stabilization.

[0009] S2. Based on the shooting angle, perform water surface reflection compensation on the target water area photo to obtain the compensated photo;

[0010] S3. Sort the compensated photos according to the shooting time, and register and stitch photos from different shooting angles at the same shooting time to obtain a panoramic image sequence.

[0011] S4. Divide the target water area into intervals, extract the interval image sequence corresponding to each water area from the panoramic image sequence, extract convolutional features for each interval image in the interval image sequence, and stitch the convolutional features with the shooting time of the panoramic image where the interval image is located and the center position of the corresponding water area to obtain a multi-dimensional feature vector.

[0012] S5. Construct a graph structure based on multidimensional feature vectors, use a graph neural network to build a water quality inversion model, and use the water quality inversion model to dynamically monitor the target water area.

[0013] It should be noted that the number of cameras carried on a drone is usually an even number and symmetrically distributed, such as 2, 4 or 6 cameras.

[0014] Preferably, step S1 includes:

[0015] Establish a geographic coordinate system that matches latitude and longitude, and establish a camera coordinate system for each camera that changes synchronously with its pose;

[0016] Adjust the shooting angles of each camera to match the flight attitude of the drone, so that the coordinate system of each camera is parallel to the geographic coordinate system.

[0017] Preferably, step S2 includes:

[0018] By checking the drone's flight log based on the photo's capture time, the drone's location in the geographic coordinate system at the time of the capture can be determined. ;

[0019] For any point on the target water surface, calculate the direction of the camera's optical axis. :

[0020]

[0021] in, This indicates the location of the point in the geographic coordinate system;

[0022] Calculate the reflection compensation angle :

[0023]

[0024] in, Represents inverse trigonometric functions, This indicates taking the 2nd norm. This indicates taking the first-order norm. Denotes the water surface normal vector and ;

[0025] Points on the target water surface Convert to the corresponding position in the camera:

[0026]

[0027] in, Point Corresponding to the position in the camera coordinate system, Represents the rotation matrix. Represents the translation vector;

[0028] Points in the camera Convert to the corresponding pixels in the photo:

[0029]

[0030] in, Point Corresponding pixel coordinates in the target water area photo Indicates focal length. Indicates pixel size scaling. Represents the pixel coordinates of the center point of the photo;

[0031] Based on the corresponding reflection compensation angle, the pixels in the target water area photograph... Compensation will be provided.

[0032] Preferably, step S3 includes:

[0033] The compensated photos are sorted by shooting time, and photos taken at the same time are arranged by shooting angle. Photos from adjacent angles are then registered and stitched together. The registration and stitching includes:

[0034] Multi-scale Gaussian blurring and downsampling are applied to the photos to be registered to construct a Gaussian pyramid;

[0035] Local extrema of the image to be registered are detected using the difference Gaussian function at various scale levels of the Gaussian pyramid.

[0036] Taylor series expansion is used to accurately fit the location and scale of local extrema, and local extrema with the lowest contrast and those located at the edge of the photo are removed to obtain key points. Feature descriptors of key points are then calculated.

[0037] For the photos to be registered, key point pairs that meet the similarity requirements are selected based on feature descriptors and used as matching feature points;

[0038] The spatial transformation relationship between the photos to be registered is calculated by matching feature points, and the photos to be registered are rotated, flipped and stitched to remove excess parts.

[0039] It should be noted that, for drones equipped with two cameras, photos taken at the same time are arranged in the order of left view, then right view, and then the left and right views are registered and stitched together. For drones equipped with four cameras, photos taken at the same time are arranged in the order of left view, front view, right view, and rear view, and then the left and front views are registered and stitched together, the front and right views are registered and stitched together, and the right and rear views are registered and stitched together. For drones equipped with six cameras, photos taken at the same time are arranged in the order of front view, right front view, right rear view, rear view, left rear view, and left front view, and then the front and right front views are registered and stitched together, the right front and right rear views are registered and stitched together, the right rear and rear views are registered and stitched together, the rear and left rear views are registered and stitched together, and the left rear and left front views are registered and stitched together.

[0040] Preferably, the feature descriptor for calculating key points includes:

[0041] Divide the area around the key point into multiple sub-regions.

[0042] Calculate the gradient magnitude and gradient direction of each pixel in each sub-region, and compile a gradient direction histogram. The peak value in the gradient direction histogram is taken as the main direction of the sub-region.

[0043] The main directions of multiple sub-regions are combined into a high-dimensional vector, which serves as the feature descriptor for the key points.

[0044] Preferably, the process of extracting the interval image sequence in step S4 includes:

[0045] The target water area is divided into sections, and the location of each section is determined by the position of its center in the geographic coordinate system. This indicates that its corresponding pixel position in the panoramic image is:

[0046]

[0047]

[0048] in, Indicates the center of the water area Corresponding pixel coordinates in a panoramic image Indicates the center of the water area The corresponding position in the camera coordinate system;

[0049] In panoramic images, with Centered on a point, a square image with a fixed side length of pixels is extracted from the surrounding area to obtain the interval image;

[0050] Each panoramic image in the panoramic image sequence is cropped sequentially to obtain a range image sequence.

[0051] Preferably, the process of constructing the graph structure in step S5 includes:

[0052] The multidimensional feature vector of each water area is used as a node in the graph structure. Multidimensional feature vectors from adjacent shooting times and adjacent water areas are connected to form edges in the graph structure. The weight of each edge is represented as... ,in, These represent the center positions of the corresponding water area intervals in the interval images of the two connected multidimensional feature vectors. These represent the capture time of the panoramic image containing the interval image in the two connected multidimensional feature vectors.

[0053] This invention also discloses a water dynamic monitoring system based on UAV remote sensing technology, comprising:

[0054] Shooting module: Adjust each camera to the target shooting angle according to the drone's flight attitude, and take photos of the target water area after stabilization;

[0055] Photo compensation module: Determines the location of the drone in the geographic coordinate system at the shooting time, calculates the direction of the camera's optical axis, determines the reflection compensation angle, and performs water surface reflection compensation on each pixel in the photo of the target water area;

[0056] Panoramic Image Module: Sorts the compensated photos according to the shooting time, registers and stitches photos from different shooting angles at the same shooting time, and outputs a panoramic image sequence;

[0057] Multidimensional feature vector module: Divide the target water area into intervals, extract the interval image sequence corresponding to each water area from the panoramic image sequence, extract convolutional features for each interval image in the interval image sequence, and concatenate the convolutional features with the shooting time of the panoramic image where the interval image is located and the center position of the corresponding water area to obtain a multidimensional feature vector;

[0058] Water quality inversion module: Based on multi-dimensional feature vectors, a graph structure is constructed, and a graph neural network is used to build a water quality inversion model. The water quality inversion model is then used to dynamically monitor the target water area.

[0059] The present invention has at least the following beneficial effects:

[0060] This invention employs a multi-angle, multi-camera imaging fusion method to acquire water area photos and performs water surface reflection compensation on the photos, effectively suppressing the impact of light noise on water body information. Simultaneously, by registering and stitching together the water area photos acquired by multiple cameras, a more comprehensive and high-precision panoramic view of the water area can be obtained, which is beneficial for the acquisition and analysis of water area image information.

[0061] The water dynamic monitoring method based on UAV remote sensing technology designed in this invention can achieve efficient, real-time and accurate water dynamic monitoring and timely early warning for abnormal situations. This invention embeds the shooting time and the center position of the water area into a feature vector to construct a multi-dimensional feature vector containing spatial location, time series and image convolution features, realizing a three-dimensional expression of the spatiotemporal characteristics of the water area. By using a graph neural network model to make full use of the spatiotemporal correlation characteristics of the water area, it automatically captures the spatiotemporal propagation law of different water quality parameters. The constructed water quality inversion model has dynamic adaptive learning ability and can effectively realize the spatiotemporal dynamic monitoring of water areas. Attached Figure Description

[0062] Figure 1 This is a flowchart illustrating the water dynamic monitoring method based on UAV remote sensing technology provided in Embodiment 1 of the present invention. Detailed Implementation

[0063] The present invention will be further described below with reference to the accompanying drawings, but this is not intended to limit the present invention in any way. Any modifications or substitutions made based on the teachings of the present invention shall fall within the protection scope of the present invention.

[0064] Example 1

[0065] like Figure 1 As shown, a method for dynamic monitoring of water areas based on UAV remote sensing technology includes the following steps:

[0066] S1: Equip the drone with at least two cameras with different shooting angles. Adjust each camera to the target shooting angle according to the drone's flight attitude, and then take photos of the target water area after stabilization; specifically:

[0067] Establish a geographic coordinate system that matches latitude and longitude, and establish a camera coordinate system for each camera that changes synchronously with its pose;

[0068] Adjust the shooting angle of each camera to match the flight attitude of the drone, so that the coordinate system of each camera is parallel to the geographic coordinate system;

[0069] The process of adjusting the camera's shooting angle is as follows: First, adjust the angle between the camera and the z-axis of the geographic coordinate system until the camera's projected area on the xoy plane of the geographic coordinate system is maximized; then, adjust the angle between the camera and the y-axis of the geographic coordinate system until the camera's projected area on the zox plane of the geographic coordinate system is maximized; finally, adjust the angle between the camera and the x-axis of the geographic coordinate system until the camera's projected area on the yoz plane of the geographic coordinate system is maximized.

[0070] In this embodiment, the x-axis of the geographic coordinate system points east, the y-axis points south, and the z-axis points in the direction of the gravity vector. At the initial moment, the camera coordinate system coincides with the geographic coordinate system. As the drone flies, the camera's pose changes. When the drone hovers, the camera's shooting angle is adjusted to make the camera coordinate system parallel to the geographic coordinate system.

[0071] S2. Based on the shooting angle, perform water surface reflection compensation on the target water area photograph to obtain the compensated photograph; specifically:

[0072] By checking the drone's flight log based on the photo's capture time, the drone's location in the geographic coordinate system at the time of the capture can be determined. ;

[0073] For any point on the target water surface, calculate the direction of the camera's optical axis. :

[0074]

[0075] in, This indicates the location of the point in the geographic coordinate system;

[0076] Calculate the reflection compensation angle :

[0077]

[0078] in, Represents inverse trigonometric functions, This indicates taking the 2nd norm. This indicates taking the first-order norm. Denotes the water surface normal vector and ;

[0079] Points on the target water surface Convert to the corresponding position in the camera:

[0080]

[0081] in, Point Corresponding to the position in the camera coordinate system, Represents the rotation matrix. Represents the translation vector;

[0082] Points in the camera Convert to the corresponding pixels in the photo:

[0083]

[0084] in, Point Corresponding pixel coordinates in the target water area photo Indicates focal length. Indicates pixel size scaling. Represents the pixel coordinates of the center point of the photo;

[0085] Based on the corresponding reflection compensation angle, the pixels in the target water area photograph... Compensation will be provided.

[0086] S3. Sort the compensated photos according to their shooting time, and register and stitch photos taken at the same shooting time but from different angles to obtain a panoramic image sequence; specifically:

[0087] The compensated photos are sorted by shooting time, and photos taken at the same time are arranged by shooting angle. Photos from adjacent angles are then registered and stitched together. The registration and stitching includes:

[0088] Multi-scale Gaussian blurring and downsampling are applied to the photos to be registered to construct a Gaussian pyramid;

[0089] Local extrema of the image to be registered are detected using the difference Gaussian function at various scale levels of the Gaussian pyramid.

[0090] By using Taylor series expansion to accurately fit the location and scale of local extrema, and removing local extrema with the lowest contrast and those located at the edge of the image, key points are obtained.

[0091] Calculate the feature descriptor of the key point: Divide the area around the key point into multiple sub-regions; calculate the gradient magnitude and gradient direction of the pixels in each sub-region, and calculate the gradient direction histogram. Take the peak value in the gradient direction histogram as the main direction of the sub-region; combine the main directions of multiple sub-regions into a high-dimensional vector as the feature descriptor of the key point.

[0092] For the photos to be registered, key point pairs that meet the similarity requirements are selected based on feature descriptors and used as matching feature points;

[0093] The spatial transformation relationship between the photos to be registered is calculated by matching feature points, and the photos to be registered are rotated, flipped and stitched to remove excess parts.

[0094] S4. Divide the target water area into intervals, extract the interval image sequence corresponding to each water area from the panoramic image sequence, extract convolutional features for each interval image in the interval image sequence, and stitch the convolutional features with the shooting time of the panoramic image where the interval image is located and the center position of the corresponding water area to obtain a multi-dimensional feature vector.

[0095] The extraction process of the interval image sequence includes:

[0096] The target water area is divided into sections, and the location of each section is determined by the position of its center in the geographic coordinate system. This indicates that its corresponding pixel position in the panoramic image is:

[0097]

[0098]

[0099] in, Indicates the center of the water area Corresponding pixel coordinates in a panoramic image Indicates the center of the water area The corresponding position in the camera coordinate system;

[0100] In panoramic images, with Centered on a point, extract a 300-pixel x 300-pixel image from the surrounding area to obtain the interval image;

[0101] Each panoramic image in the panoramic image sequence is cropped sequentially to obtain a range image sequence.

[0102] S5. Construct a graph structure based on multidimensional feature vectors, use graph neural networks to construct a water quality inversion model, and use the water quality inversion model to dynamically monitor the target water area;

[0103] The process of constructing the graph structure includes:

[0104] The multidimensional feature vector of each water area is used as a node in the graph structure. Multidimensional feature vectors from adjacent shooting times and adjacent water areas are connected to form edges in the graph structure. The weight of each edge is represented as... ,in, These represent the center positions of the corresponding water area intervals in the interval images of the two connected multidimensional feature vectors. These represent the capture time of the panoramic image containing the interval image in the two connected multidimensional feature vectors;

[0105] In this embodiment, multidimensional feature vectors from adjacent water areas are connected to represent the mutual influence of water quality between neighboring water areas, or the impact of pollutant diffusion on neighboring water areas. Multidimensional feature vectors from adjacent shooting times are connected to represent the dynamic changes of water quality over time.

[0106] Example 2

[0107] This embodiment is used to verify the effectiveness of the water quality inversion model constructed in Embodiment 1 in dynamic monitoring of water areas.

[0108] Data collection:

[0109] Select Pearl River The typical waterway section (including three core monitoring areas: the front channel, the west channel, and the rear channel) covers a total area of ​​approximately 120 km², encompassing complex scenarios such as urban river sections, tributary confluences, and areas near sewage outlets. A drone equipped with a visible light camera (0.2m resolution) collected images three times daily at 8:00, 14:00, and 20:00 along a fixed flight path (100m altitude), continuously acquiring data for 30 days to form a sequence of 90 panoramic images (including sunny / partly cloudy / overcast weather conditions). The timestamp of each image capture (accurate to the second) and the drone's location coordinates (latitude and longitude, accuracy ±0.1m) were recorded simultaneously. Meanwhile, water quality parameters were collected simultaneously at 15 national-level monitoring sections three times daily (aligned with the remote sensing data collection time), obtaining 450 sets of measured samples.

[0110] Divide the water area into sections:

[0111] The monitoring area was divided into 50 dynamic water zones based on water flow direction and geographical features, and the center coordinates of each water zone were marked.

[0112] Constructing multidimensional feature vectors:

[0113] For each water area, the corresponding image sequence is extracted using ResNet-50 to extract convolutional features (2048-dimensional vector) to capture water texture and spectral gradient visual features; the shooting time (converted to the minute difference from the initial time, 1-dimensional) and the center coordinates of the area (2-dimensional) are stitched together to form a 2051-dimensional multidimensional feature vector.

[0114] Constructing a graph structure:

[0115] The multidimensional feature vector of each water area is used as a graph node to connect geographically adjacent water areas (areas with a distance of <1km are defined as adjacent, and directed edges are constructed with the direction away from the observation station as the positive direction); nodes in the same water area with adjacent shooting times (interval ≤6 hours, directed edges are constructed, and the time order is the edge direction); an adjacency matrix (recording edge connection relationships) and a node feature matrix (multidimensional feature vector) are generated.

[0116] Training process:

[0117] The dataset is divided into a training set (first 20 days, 3000 samples), a validation set (middle 5 days, 750 samples), and a test set (last 5 days, 750 samples) in chronological order. Inputs include node feature matrix X, adjacency matrix A, and water quality parameter labels (5 indicators). The learning rate is optimized, and the training cycle is 200 rounds.

[0118] Using inversion accuracy as an evaluation index, the method of this invention and the traditional multiple linear regression method are compared, and the results are shown in Table 1 below.

[0119] Table 1

[0120]

[0121] It is evident that the graph neural network model based on multidimensional feature vectors, by integrating spatial adjacency relationships and temporal series dependencies, can effectively improve the accuracy of water quality inversion and dynamic monitoring capabilities in complex water scenarios, thus verifying the advantages of the technical solution of this invention.

[0122] It should be noted that the sequence numbers of the above embodiments of the present invention are merely for descriptive purposes and do not represent the superiority or inferiority of the embodiments. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, apparatus, article, or method. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, apparatus, article, or method that includes that element.

[0123] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0124] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A water area dynamic monitoring method based on unmanned aerial vehicle remote sensing technology, characterized in that, The method comprises the following steps: S1, at least two cameras with different shooting angles are carried on the unmanned aerial vehicle, and each camera is adjusted to a target shooting angle according to the flight attitude of the unmanned aerial vehicle, and then the target water area is photographed after stabilization; S2, water surface reflection compensation is performed on the target water area photograph based on the shooting angle to obtain a compensated photograph; S3, the compensated photograph is sorted according to the shooting time, and the photographs with different shooting angles at the same shooting time are registered and spliced to obtain a panoramic image sequence; S4, the target water area is divided into intervals, and an interval image sequence corresponding to each water area interval is extracted from the panoramic image sequence, and a convolution feature is extracted from each interval image in the interval image sequence, and the convolution feature is spliced with the shooting time of the panoramic image where the interval image is located and the center position of the corresponding water area interval to obtain a multi-dimensional feature vector; S5, a graph structure is constructed based on the multi-dimensional feature vector, a water quality inversion model is constructed by using a graph neural network, and the target water area is dynamically monitored by using the water quality inversion model; The step S2 comprises: According to the photograph shooting time, the unmanned aerial vehicle flight log is inquired, and the position of the unmanned aerial vehicle in a geographic coordinate system when the photograph is shot is determined ; For any point on the plane of the target water area, the camera optical axis direction is calculated : ； wherein represents the position of the point in the geographical coordinate system; Computing reflection compensation angles : ； wherein denotes the inverse trigonometric function, denotes the 2-norm, denotes the 1-norm, denotes the water surface normal vector and ; converting points on a plane of a target water area to corresponding positions in the camera: ； wherein representing a point corresponding to a position in the camera coordinate system, representing a rotation matrix, representing a translation vector; Converting a point in the camera to the corresponding pixel in the photo: ； wherein, representing a point corresponding to a pixel coordinate in a photo of the target water area, representing a focal length, representing a pixel size scaling, representing a pixel coordinate of a photo center point; Compensate the pixels in the target water area photo based on the corresponding reflection compensation angle. Compensate the pixels in the target water area photo based on the corresponding reflection compensation angle. 2.The water area dynamic monitoring method based on the UAV remote sensing technology according to claim 1, characterized in that, The step S1 comprises: A geographic coordinate system matched with the latitude and longitude is established, and a camera coordinate system synchronous with the pose of each camera is established for each camera; The shooting angle of each camera is adjusted according to the flight attitude of the unmanned aerial vehicle, so that the coordinate system of each camera is parallel to the geographic coordinate system. 3.The water area dynamic monitoring method based on the UAV remote sensing technology according to claim 1, characterized in that, The step S3 comprises: The compensated photograph is sorted according to the shooting time, and the photographs at the same shooting time are arranged according to the shooting angle, and the photographs at adjacent angles are registered and spliced; the registration and splicing comprises: Multi-scale Gaussian blur and down-sampling are performed on the photographs to be registered to construct a Gaussian pyramid; Local extreme points of the photographs to be registered are detected on each scale layer of the Gaussian pyramid by using a difference Gaussian function; The position and scale of the local extreme points are accurately fitted by using Taylor series expansion, the local extreme points with the lowest contrast and the local extreme points located at the edge of the photograph are removed, key points are obtained, and feature descriptors of the key points are calculated; For the photographs to be registered, key point pairs satisfying the similarity requirement are selected based on the feature descriptors as matched feature points; The spatial transformation relationship between the photographs to be registered is calculated through the matched feature points, and the photographs to be registered are rotated, flipped and spliced, and the redundant part is cut off. 4.The water area dynamic monitoring method based on the UAV remote sensing technology according to claim 3, characterized in that, The calculation of the feature descriptors of the key points comprises: The area around the key point is divided into a plurality of sub-areas; The gradient amplitude and gradient direction of the pixel points in each sub-area are calculated, and the gradient direction histogram is counted, and the peak value in the gradient direction histogram is taken as the main direction of the sub-area; The main directions of the plurality of sub-areas are combined into a high-dimensional vector as the feature descriptor of the key point. 5.The water area dynamic monitoring method based on the UAV remote sensing technology according to claim 1, characterized in that, The extraction process of the interval image sequence in the step S4 comprises: The target water area is divided into intervals, and the position of each water area interval is represented by the position of the center of the water area interval in the geographic coordinate system The pixel position corresponding to the panoramic image is represented as: ； ； wherein, represents the water area section center corresponding to the pixel coordinate in the panoramic image, represents the water area section center corresponding to the position in the camera coordinate system; In panoramic images, with Centered on a point, a square image with a fixed side length of pixels is extracted from the surrounding area to obtain the interval image; Each panoramic image in the panoramic image sequence is sequentially intercepted to obtain an interval image sequence. 6.The water area dynamic monitoring method based on the UAV remote sensing technology according to claim 1, characterized in that, The construction process of the graph structure in the step S5 comprises: The multi-dimensional feature vector of each water area interval is taken as a node of the graph structure, and the multi-dimensional feature vectors from adjacent shooting times and adjacent water area intervals are connected to form edges of the graph structure, and the weight of the edge is represented as wherein, respectively represent the center positions of the water area intervals corresponding to the interval images in the two connected multi-dimensional feature vectors, respectively represent the shooting times of the panoramic images in which the interval images in the two connected multi-dimensional feature vectors are located.

7. A water area dynamic monitoring system based on unmanned aerial vehicle remote sensing technology, characterized in that, It comprises: A photographing module: each camera is adjusted to a target shooting angle according to the flight attitude of the unmanned aerial vehicle, and then the target water area is photographed after stabilization; Photo compensation module: determine the position of the shooting time unmanned aerial vehicle in the geographical coordinate system, calculate the camera optical axis direction, determine the reflection compensation angle, and compensate the water surface reflection of each pixel in the target water area photo; Panoramic image module: sort the compensated photos according to the shooting time, and register and splice the photos with different shooting angles at the same shooting time, and output a panoramic image sequence; Multi-dimensional feature vector module: divide the target water area into intervals, extract the interval image sequence corresponding to each water area interval in the panoramic image sequence, extract the convolutional features of each interval image in the interval image sequence, splice the convolutional features with the shooting time of the panoramic image where the interval image is located, and the center position of the corresponding water area interval, and obtain a multi-dimensional feature vector; Water quality inversion module: construct a graph structure based on the multi-dimensional feature vector, construct a water quality inversion model using a graph neural network, and dynamically monitor the target water area using the water quality inversion model; To realize the water area dynamic monitoring method based on unmanned aerial vehicle remote sensing technology as claimed in any one of claims 1-6.

Images