Water area dynamic monitoring method and system based on unmanned aerial vehicle remote sensing technology

Abstract

The invention discloses a water area dynamic monitoring method and system based on an unmanned aerial vehicle remote sensing technology, and the method comprises the steps: adjusting the shooting angle of each camera according to the flight attitude of an unmanned aerial vehicle, and shooting a target water area photo after stabilization; performing water surface reflection compensation on the target water area picture based on the shooting angle; the compensated photos are sequenced according to the shooting time, and the photos of different shooting angles at the same shooting time are registered and spliced to obtain a panoramic image sequence; dividing the target water area into intervals, extracting an interval image sequence corresponding to each water area interval from the panoramic image sequence, extracting convolution features of each interval image in the interval image sequence, and splicing the convolution features with the shooting time of the panoramic image where the interval image is located and the central position of the corresponding water area interval to obtain a panoramic image sequence; obtaining a multi-dimensional feature vector; and constructing a water quality inversion model based on the multi-dimensional feature vector. The water quality inversion model constructed by the invention can realize dynamic monitoring of a water area and release early warning information more efficiently and accurately.

Description

Technical Field

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

[0002] Water remote sensing technology, with its non-contact and wide coverage characteristics, has become the core means of dynamic monitoring of water bodies. It captures the relationship between water quality parameters and the spectral response of remote sensing bands, builds an inversion model to achieve quantitative analysis of the water environment. The 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) wavelengths to accurately identify water pollution types, algae community distribution, and suspended sediment concentration, providing basic spectral data for water quality parameter inversion. Thermal infrared sensors dynamically monitor water temperature fields by monitoring radiation energy in the 9-12μm thermal infrared band, offering irreplaceable advantages in tracking the spread of thermal pollution from thermal power plants and analyzing offshore upwelling distribution. LiDAR uses 1064nm / 532nm wavelength laser pulses to penetrate water bodies. Combined with echo signal analysis, it can obtain centimeter-level accurate water surface elevation models and underwater terrain contours, supporting the monitoring of water level changes and riverbed evolution. Synthetic aperture radar utilizes C / X / L-band microwaves to penetrate optical obstacles such as clouds, fog, and rainfall, achieving all-weather imaging. It plays a key role in monitoring the extent of typhoon-induced river basin flooding and assessing navigation conditions in ice-covered waterways.

[0003] However, although the above remote sensing technologies have certain water monitoring capabilities, due to the complex optical properties and dynamic environmental changes of natural waters, they have the following defects in practical applications: 1. Water surface reflection interference bias: Specular reflection caused by sunlight often obscures the true information of the water body, resulting in deviations in the inversion of water quality parameters. Specifically, the water surface, as a natural reflective interface, 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 of chlorophyll a concentration will increase by 30-50%, and the inversion deviation rate of 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, resulting in a general decrease of 0.15-0.25 in the R² of the water quality parameter inversion model.

[0004] 2. Insufficient recognition accuracy in complex scenes: Turbid water bodies and shadows are difficult to distinguish, and the ability to detect 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 body surges, resulting in overlap between the reflectivity of the near-infrared band above 650nm and the spectral characteristics of the bridge / vegetation shadow area (difference <10%). The traditional threshold segmentation algorithm has an error rate of up to 25%. In addition, for floating objects with a particle size of less than 5cm (such as plastic fragments and oil film patches), although existing centimeter-level resolution sensors can capture their presence, the multispectral mixed 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 precise identification needs of emergency monitoring. Summary of the Invention

[0005] The purpose of the present invention is to provide an efficient, real-time and accurate solution for dynamic monitoring of water areas. By deploying a multi-angle imaging system and performing multi-perspective compensation for water surface reflection interference, 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 is constructed that can realize spatiotemporal dynamic monitoring of water areas, thereby improving the ability to identify pollutants in turbid water bodies and complex scenes.

[0006] To achieve the above objectives, the present invention discloses a method for dynamic monitoring of water areas based on UAV remote sensing technology, comprising the following steps: 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 posture. After stabilization, take photos of the target water area. S2. Compensate the target water area photo for water surface reflection based on the shooting angle to obtain a compensated photo; S3, sorting the compensated photos according to shooting time, and registering and stitching the photos taken at the same shooting time but different viewing angles to obtain a panoramic image sequence; S4. Divide the target water area into intervals, extract an interval image sequence corresponding to each water area interval from the panoramic image sequence, extract convolution features from each interval image in the interval image sequence, and concatenate the convolution features with the shooting time of the panoramic image in which the interval image is located and the center position of the corresponding water area interval to obtain a multidimensional feature vector; S5. Build a graph structure based on multidimensional feature vectors, use graph neural network to build a water quality inversion model, and use the water quality inversion model to dynamically monitor the target water area.

[0007] 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.

[0008] Preferably, the step S1 includes: Establish a geographic coordinate system that matches the longitude and latitude, and establish a camera coordinate system for each camera that changes synchronously with its posture; According to the flight attitude of the UAV, the shooting angle of each camera is adjusted so that the coordinate system of each camera is parallel to the geographic coordinate system.

[0009] Preferably, step S2 includes: Query the drone flight log based on the time the photo was taken to determine the drone's location in the geographic coordinate system at the time the photo was taken. ; For any point on the target water plane, calculate the direction of the camera optical axis :

[0010] in, Indicates the location of the point in the geographic coordinate system; Calculate reflection compensation angle :

[0011] in, represents the inverse trigonometric function, Indicates taking the 2nd-order norm, Indicates taking the first-order norm, represents the water surface normal vector and ; The point on the target water plane Convert to the corresponding position in the camera:

[0012] in, Indicates a point Corresponding to the position in the camera coordinate system, represents the rotation matrix, represents the translation vector; The point in the camera Convert to the corresponding pixels in the photo:

[0013] in, Indicates a point Corresponding to the pixel coordinates in the target water area photo, represents the focal length, Indicates pixel size scaling, Indicates the pixel coordinates of the center point of the photo; Based on the corresponding reflection compensation angle, the pixels in the target water area photo are Make compensation.

[0014] Preferably, step S3 includes: The compensated photos are sorted according to the shooting time, the photos taken at the same time are arranged according to the shooting angle, and the photos of adjacent angles are aligned and spliced; the alignment and splicing includes: Perform multi-scale Gaussian blur and downsampling on the photos to be registered to construct a Gaussian pyramid; At each scale level of the Gaussian pyramid, the difference Gaussian function is used to detect the local extreme points of the photos to be registered; The Taylor series expansion is used to accurately fit the position and scale of local extreme points, and the local extreme points with the lowest contrast and the local extreme points located at the edge of the photo are removed to obtain key points, and the feature descriptors of the key points are calculated; For the photos to be registered, the key point pairs that meet the similarity requirements are screened out based on the feature descriptors as matching feature points; 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 spliced, and the redundant parts are cropped.

[0015] It should be noted that, for the case where the drone is equipped with two cameras, the photos taken at the same time are arranged in the order of left view and right view, and then the photos of the left view and the right view are aligned and stitched together; for the case where the drone is equipped with four cameras, the photos taken at the same time are arranged in the order of left view, front view, right view, and rear view, and then the photos of the left view and the front view are aligned and stitched together, the photos of the front view and the right view are aligned and stitched together, and the photos of the right view and the rear view are aligned and stitched together; for the case where the drone is equipped with six cameras, the 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 photos of the front view and the right front view are aligned and stitched together, the photos of the right front view and the right rear view are aligned and stitched together, the photos of the right rear view and the rear view are aligned and stitched together, the photos of the rear view and the left rear view are aligned and stitched together, and the photos of the left rear view and the left front view are aligned and stitched together.

[0016] Preferably, the calculation of the feature descriptor of the key point includes: Taking the key point as the center, the area around it is divided into multiple sub-areas; Calculate the gradient magnitude and gradient direction of the pixels in each sub-region, and generate a gradient direction histogram. The peak value in the gradient direction histogram is taken as the main direction of the sub-region. The main directions of multiple sub-regions are combined into a high-dimensional vector as the feature descriptor of the key point.

[0017] Preferably, the process of extracting the interval image sequence in step S4 includes: The target water area is divided into intervals, and the position of each water area interval is the position of the center of the water area interval in the geographic coordinate system. Indicates that its corresponding pixel position in the panoramic image is:

[0018]

[0019] in, Indicates the center of the water area Corresponding to the pixel coordinates in the panoramic image, Indicates the center of the water area Corresponding to the position in the camera coordinate system; In the panoramic image, As the center, intercept the square image with fixed pixel length around it to get the interval image; Each panoramic image in the panoramic image sequence is intercepted in sequence to obtain an interval image sequence.

[0020] Preferably, the process of constructing the graph structure in step S5 includes: The multidimensional feature vector of each water area is used as a node of the graph structure, and the multidimensional feature vectors from adjacent shooting times and adjacent water areas are connected to form the edges of the graph structure. The weight of the edge is expressed as ,in, They represent the center position of the water area corresponding to the interval image in the two connected multidimensional feature vectors, They respectively represent the shooting time of the panoramic image where the interval image in the two connected multidimensional feature vectors is located.

[0021] The present invention also discloses a water area dynamic monitoring system based on UAV remote sensing technology, comprising: Shooting module: adjust each camera to the target shooting angle according to the UAV's flight posture, and take pictures of the target water area after stabilization; Photo compensation module: determines the position of the drone in the geographic coordinate system at the time of shooting, 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 target water area photo; Panoramic image module: sorts the compensated photos by shooting time, registers and stitches photos taken at the same shooting time but different viewing angles, and outputs a panoramic image sequence; Multidimensional feature vector module: Divide the target water area into intervals, extract the interval image sequence corresponding to each water area interval from the panoramic image sequence, extract convolution features from each interval image in the interval image sequence, and splice the convolution 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 to obtain a multidimensional feature vector; Water quality inversion module: Build a graph structure based on multi-dimensional feature vectors, use graph neural network to build a water quality inversion model, and use the water quality inversion model to dynamically monitor the target water area.

[0022] The present invention has at least the following beneficial effects: The present invention adopts a multi-angle multi-camera imaging fusion method to collect water area photos, and performs water surface reflection compensation on the water area photos, effectively suppressing the influence of light noise on water body information; at the same time, the water area photos collected by multiple cameras are aligned, spliced ​​and fused to obtain a more comprehensive and high-precision water area panorama, which is conducive to the collection and analysis of water area image information.

[0023] The water area dynamic monitoring method based on UAV remote sensing technology designed in the present invention can realize efficient, real-time and accurate water area dynamic monitoring, and timely warn of abnormal situations; the present invention embeds the shooting time and the center position of the water area into the feature vector, constructs a multidimensional feature vector containing spatial position, time series, and image convolution features, and realizes a three-dimensional expression of the spatiotemporal characteristics of the water area; the graph neural network model is used to fully utilize the spatiotemporal correlation characteristics of the water area, automatically capture the spatiotemporal propagation laws of different water quality parameters, and the constructed water quality inversion model has dynamic adaptive learning ability, which can well realize the spatiotemporal dynamic monitoring of the water area. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 This is a flow chart of a method for dynamic water area monitoring based on UAV remote sensing technology provided in Example 1 of the present invention. DETAILED DESCRIPTION

[0025] The present invention will be further described below with reference to the accompanying drawings, but the present invention is not limited in any way. Any changes or substitutions made based on the teachings of the present invention fall within the scope of protection of the present invention.

[0026] Example 1 like Figure 1 As shown, a water area dynamic monitoring method based on UAV remote sensing technology includes the following steps: 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 posture. After stabilization, take photos of the target water area. Specifically: Establish a geographic coordinate system that matches the longitude and latitude, and establish a camera coordinate system for each camera that changes synchronously with its posture; Adjust the shooting angle of each camera according to the flight attitude of the drone so that the coordinate system of each camera is parallel to the geographic coordinate system; The process of adjusting the camera 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 projection 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 projection 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 projection area on the yoz plane of the geographic coordinate system is maximized; In this embodiment, the x-axis of the geographic coordinate system points to the east, the y-axis points to the south, and the z-axis points to 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 position changes. When the drone hovers, the camera's shooting angle is adjusted to make the camera coordinate system parallel to the geographic coordinate system.

[0027] S2. Performing water surface reflection compensation on the target water area photograph based on the shooting angle to obtain a compensated photograph; specifically: Query the drone flight log based on the time the photo was taken to determine the drone's location in the geographic coordinate system at the time the photo was taken. ; For any point on the target water plane, calculate the direction of the camera optical axis :

[0028] in, Indicates the location of the point in the geographic coordinate system; Calculate reflection compensation angle :

[0029] in, represents the inverse trigonometric function, Indicates taking the 2nd-order norm, Indicates taking the first-order norm, represents the water surface normal vector and ; The point on the target water plane Convert to the corresponding position in the camera:

[0030] in, Indicates a point Corresponding to the position in the camera coordinate system, represents the rotation matrix, represents the translation vector; The point in the camera Convert to the corresponding pixels in the photo:

[0031] in, Indicates a point Corresponding to the pixel coordinates in the target water area photo, represents the focal length, Indicates pixel size scaling, Indicates the pixel coordinates of the center point of the photo; Based on the corresponding reflection compensation angle, the pixels in the target water area photo are Make compensation.

[0032] S3. Sort the compensated photos by shooting time, and register and stitch photos taken at the same shooting time but from different angles to obtain a panoramic image sequence; specifically: The compensated photos are sorted according to the shooting time, the photos taken at the same time are arranged according to the shooting angle, and the photos of adjacent angles are aligned and spliced; the alignment and splicing includes: Perform multi-scale Gaussian blur and downsampling on the photos to be registered to construct a Gaussian pyramid; At each scale level of the Gaussian pyramid, the difference Gaussian function is used to detect the local extreme points of the photos to be registered; Taylor series expansion is used to accurately fit the position and scale of local extreme points, and local extreme points with the lowest contrast and local extreme points located at the edge of the photo are removed to obtain key points; Calculate the feature descriptor of the key point: With the key point as the center, divide the area around it into multiple sub-regions; calculate the gradient amplitude and gradient direction of the pixels in each sub-region, and calculate the gradient direction histogram. The peak value in the gradient direction histogram is used 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; For the photos to be registered, the key point pairs that meet the similarity requirements are screened out based on the feature descriptors as matching feature points; 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 spliced, and the redundant parts are cropped.

[0033] S4. Divide the target water area into intervals, extract an interval image sequence corresponding to each water area interval from the panoramic image sequence, extract convolution features from each interval image in the interval image sequence, and concatenate the convolution features with the shooting time of the panoramic image in which the interval image is located and the center position of the corresponding water area interval to obtain a multidimensional feature vector; The extraction process of the interval image sequence includes: The target water area is divided into intervals, and the position of each water area interval is the position of the center of the water area interval in the geographic coordinate system. Indicates that its corresponding pixel position in the panoramic image is:

[0034]

[0035] in, Indicates the center of the water area Corresponding to the pixel coordinates in the panoramic image, Indicates the center of the water area Corresponding to the position in the camera coordinate system; In the panoramic image, Take the image of 300 pixels × 300 pixels around the center to get the interval image; Each panoramic image in the panoramic image sequence is intercepted in sequence to obtain an interval image sequence.

[0036] S5. Build a graph structure based on multidimensional feature vectors, use graph neural networks to build a water quality inversion model, and use the water quality inversion model to dynamically monitor the target water area; The process of constructing the graph structure includes: The multidimensional feature vector of each water area is used as a node of the graph structure, and the multidimensional feature vectors from adjacent shooting times and adjacent water areas are connected to form the edges of the graph structure. The weight of the edge is expressed as ,in, They represent the center position of the water area corresponding to the interval image in the two connected multidimensional feature vectors, They respectively represent the shooting time of the panoramic image where the interval image in the two connected multidimensional feature vectors is located; In this embodiment, the multidimensional feature vectors from adjacent water areas are connected to represent the mutual influence of water quality between adjacent water areas, or the impact of pollutant diffusion on adjacent water areas, and the multidimensional feature vectors from adjacent shooting times are connected to represent the dynamic changes of water quality over time.

[0037] Example 2 This example is used to verify the effectiveness of the water quality inversion model constructed in Example 1 in dynamic monitoring of water areas.

[0038] Collect data: Select Pearl River The monitoring area covers approximately 120 square kilometers of typical waters (including three core monitoring areas: the front channel, the west channel, and the back channel), covering complex locations such as urban river sections, tributary confluences, and areas near sewage outfalls. A drone equipped with a visible light camera (0.2m resolution) collected images three times daily at 8:00, 14:00, and 20:00, following a fixed route (at an altitude of 100m). This data was acquired for 30 consecutive days, generating 90 panoramic image sequences (covering clear, cloudy, and overcast weather conditions). The acquisition timestamp (accurate to the second) and the drone's position coordinates (latitude and longitude, with an accuracy of ±0.1m) were simultaneously recorded. Simultaneously, water quality parameters were collected three times daily (aligned with remote sensing acquisition times) at 15 nationally monitored sections, yielding 450 sets of measured samples.

[0039] Divide water areas: The monitoring area is divided into 50 dynamic water areas according to the direction of water flow and geographical features, and the center coordinates of each water area are marked. Construct a multidimensional feature vector: For the interval image sequence corresponding to each water area, ResNet-50 is used to extract convolutional features (2048-dimensional vectors) to capture the visual characteristics of water texture and spectral gradient; the shooting time (converted to the minute difference from the initial time, 1 dimension) and the interval center coordinates (2 dimensions) are spliced ​​to form a 2051-dimensional multidimensional feature vector.

[0040] Constructing the graph structure: The multidimensional feature vector of each water area is used as a graph node to connect geographically adjacent water areas (areas with a distance <1 km are defined as adjacent, and directed edges are constructed with the distance away from the observation station as the positive direction); nodes with adjacent shooting times in the same water area are connected (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.

[0041] Training process: The model was 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. The inputs were the node feature matrix X, the adjacency matrix A, and water quality parameter labels (five indicators). The learning rate was optimized and the training cycle was 200 rounds. The inversion accuracy was used as the evaluation index to compare the method of the present invention with the traditional multiple linear regression method. The results are shown in Table 1 below.

[0042] Table 1

[0043] It can be seen that the graph neural network model based on multidimensional feature vectors can effectively improve the water quality inversion accuracy and dynamic monitoring capabilities in complex water scenes by integrating spatial adjacency relationships and time series dependencies, verifying the advantageous effects of the technical solution of the present invention.

[0044] It should be noted that the serial numbers of the above-mentioned embodiments of the present invention are for descriptive purposes only and do not represent the advantages or disadvantages of the embodiments. In addition, the terms "including", "comprising" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, device, article or method comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "including a ..." does not exclude the presence of other identical elements in the process, device, article or method comprising the element.

[0045] Through the description of the above embodiments, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus the necessary general hardware platform, and of course can also be implemented by hardware, but in many cases the former is a better embodiment. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes a number of instructions for enabling a terminal device (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods described in each embodiment of the present invention.

[0046] The above are only preferred embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention description and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for dynamic monitoring of water areas based on UAV remote sensing technology, characterized in that: The following steps are involved: 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 posture. After stabilization, take photos of the target water area. S2. Compensate the target water area photo for water surface reflection based on the shooting angle to obtain a compensated photo; S3, sorting the compensated photos according to shooting time, and registering and stitching the photos taken at the same shooting time but different viewing angles to obtain a panoramic image sequence; S4. Divide the target water area into intervals, extract an interval image sequence corresponding to each water area interval from the panoramic image sequence, extract convolution features from each interval image in the interval image sequence, and concatenate the convolution features with the shooting time of the panoramic image in which the interval image is located and the center position of the corresponding water area interval to obtain a multidimensional feature vector; S5. Build a graph structure based on multidimensional feature vectors, use graph neural network to build a water quality inversion model, and use the water quality inversion model to dynamically monitor the target water area.

2. The method for dynamic monitoring of water areas based on UAV remote sensing technology according to claim 1, characterized in that: The step S1 comprises: Establish a geographic coordinate system that matches the longitude and latitude, and establish a camera coordinate system for each camera that changes synchronously with its posture; According to the flight attitude of the UAV, the shooting angle of each camera is adjusted so that the coordinate system of each camera is parallel to the geographic coordinate system.

3. The method for dynamic monitoring of water areas based on UAV remote sensing technology according to claim 1, characterized in that: The step S2 comprises: Query the drone flight log based on the time the photo was taken to determine the drone's location in the geographic coordinate system at the time the photo was taken. ; For any point on the target water plane, calculate the direction of the camera optical axis : in, Indicates the location of the point in the geographic coordinate system; Calculate reflection compensation angle : in, represents the inverse trigonometric function, Indicates taking the 2nd-order norm, Indicates taking the first-order norm, represents the water surface normal vector and ; The point on the target water plane Convert to the corresponding position in the camera: in, Indicates a point Corresponding to the position in the camera coordinate system, represents the rotation matrix, represents the translation vector; The point in the camera Convert to the corresponding pixels in the photo: in, Indicates a point Corresponding to the pixel coordinates in the target water area photo, represents the focal length, Indicates pixel size scaling, Indicates the pixel coordinates of the center point of the photo; Based on the corresponding reflection compensation angle, the pixels in the target water area photo are Make compensation.

4. The method for dynamic monitoring of water areas based on UAV remote sensing technology according to claim 1, characterized in that: The step S3 comprises: The compensated photos are sorted according to the shooting time, the photos taken at the same time are arranged according to the shooting angle, and the photos of adjacent angles are aligned and spliced; the alignment and splicing includes: Perform multi-scale Gaussian blur and downsampling on the photos to be registered to construct a Gaussian pyramid; At each scale level of the Gaussian pyramid, the difference Gaussian function is used to detect the local extreme points of the photos to be registered; The Taylor series expansion is used to accurately fit the position and scale of local extreme points, and the local extreme points with the lowest contrast and the local extreme points located at the edge of the photo are removed to obtain key points, and the feature descriptors of the key points are calculated; For the photos to be registered, the key point pairs that meet the similarity requirements are screened out based on the feature descriptors as matching feature points; 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 spliced, and the redundant parts are cropped.

5. The method for dynamic monitoring of water areas based on UAV remote sensing technology according to claim 4 is characterized in that: The feature descriptor of the key point calculation includes: Taking the key point as the center, the area around it is divided into multiple sub-areas; Calculate the gradient magnitude and gradient direction of the pixels in each sub-region, and generate a gradient direction histogram. The peak value in the gradient direction histogram is taken as the main direction of the sub-region. The main directions of multiple sub-regions are combined into a high-dimensional vector as the feature descriptor of the key point.

6. The method for dynamic monitoring of water areas based on UAV remote sensing technology according to claim 1, characterized in that: The extraction process of the interval image sequence in step S4 includes: The target water area is divided into intervals, and the position of each water area interval is the position of the center of the water area interval in the geographic coordinate system. Indicates that its corresponding pixel position in the panoramic image is: in, Indicates the center of the water area Corresponding to the pixel coordinates in the panoramic image, Indicates the center of the water area Corresponding to the position in the camera coordinate system; In the panoramic image, As the center, intercept the square image with fixed pixel length around it to get the interval image; Each panoramic image in the panoramic image sequence is intercepted in sequence to obtain an interval image sequence.

7. The method for dynamic monitoring of water areas based on UAV remote sensing technology according to claim 1, characterized in that: The construction process of the graph structure in step S5 includes: The multidimensional feature vector of each water area is used as a node of the graph structure, and the multidimensional feature vectors from adjacent shooting times and adjacent water areas are connected to form the edges of the graph structure. The weight of the edge is expressed as ,in, They represent the center position of the water area corresponding to the interval image in the two connected multidimensional feature vectors, They respectively represent the shooting time of the panoramic image where the interval image in the two connected multidimensional feature vectors is located.

8. A water dynamic monitoring system based on UAV remote sensing technology, characterized in that: include: Shooting module: adjust each camera to the target shooting angle according to the UAV's flight posture, and take pictures of the target water area after stabilization; Photo compensation module: determines the position of the drone in the geographic coordinate system at the time of shooting, 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 target water area photo; Panoramic image module: sorts the compensated photos by shooting time, registers and stitches photos taken at the same shooting time but different viewing angles, and outputs a panoramic image sequence; Multidimensional feature vector module: Divide the target water area into intervals, extract the interval image sequence corresponding to each water area interval from the panoramic image sequence, extract convolution features from each interval image in the interval image sequence, and splice the convolution 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 to obtain a multidimensional feature vector; Water quality inversion module: Builds a graph structure based on multi-dimensional feature vectors, uses graph neural networks to build a water quality inversion model, and uses the water quality inversion model to dynamically monitor the target water area; To realize the water area dynamic monitoring method based on UAV remote sensing technology as described in any one of claims 1 to 7.

Images