A method for hydrological monitoring based on a drone

Abstract

The present application relates to a kind of unmanned aerial vehicle-based hydrological monitoring method, it is related to hydrological monitoring technical field, the present application is by constructing bottom characteristics library and is combined with radiation transfer equation to execute deconvolution operation, can accurately exclude background noise, extract pure water body reflection signal, to eliminate the pollution inversion phenomenon of shallow water area, simultaneously through water depth spatial probability distribution model, realizes the depth coupling of spectral feature and physical depth data, by integrated multi-source data acquisition unit, multi-spectral, high-resolution RGB texture and elevation / depth physical data are integrated into multidimensional image set, so that it can still maintain stable recognition rate and inversion robustness under complex light and shadow conditions, different bottom types and wave interference environment, finally output is the visual monitoring data after regional feature mark and signal correction.

Description

Technical Field

[0001] This invention relates to the field of hydrological monitoring technology, and in particular to a hydrological monitoring method based on unmanned aerial vehicles (UAVs). Background Technology

[0002] With the increasing frequency of global climate change and extreme weather events, hydrological disasters, such as floods, droughts, and landslides, have become major challenges affecting human social development and the ecological environment. Traditional hydrological monitoring methods mainly rely on ground stations, sensors, and manual inspections. While these methods provide some hydrological data, their monitoring efficiency is low in wide areas, complex terrain, and extreme weather conditions, their data coverage is limited, and they are slow to respond to rapidly changing hydrological conditions. Furthermore, traditional methods cannot provide comprehensive real-time monitoring and forecasting in a short period of time, making it difficult to issue early warnings for hydrological disasters and significantly limiting preventative capabilities.

[0003] With the development of drone technology, its advantages such as high efficiency, flexibility, and low cost have led to its widespread application in various fields. In the field of hydrological monitoring, drones can achieve real-time monitoring of large areas of water by carrying high-definition cameras, infrared sensors, lidar, and other sensors, obtaining accurate hydrological data such as water level, flow velocity, and flow rate. Especially in areas with complex terrain and difficult access, the aerial monitoring advantages of drones are particularly prominent, avoiding the limitations of traditional monitoring methods and providing comprehensive and real-time hydrological information.

[0004] However, in clear, shallow water areas (such as river shallows and lake shores), incident light penetrates the water and reflects off the bottom of the riverbed. The total incident and reflected intensity received by the fiber optic sensor on the drone includes surface reflection, water scattering, and bottom reflection. Because the intensity of bottom reflection often far exceeds that of water scattering, traditional inversion models cannot accurately extract signals representing water quality components (such as turbidity and chlorophyll), resulting in significant monitoring errors.

[0005] To address the aforementioned technical deficiencies, a solution is proposed. Summary of the Invention

[0006] The purpose of this invention is to accurately remove background noise and extract pure water reflection signals by constructing a sediment feature library and performing deconvolution operations in conjunction with the radiative transfer equation, thereby eliminating pollution inversion phenomena in shallow water areas. At the same time, through the water depth spatial probability distribution model, deep coupling of spectral features and physical depth data is achieved. Through the integrated multi-source data acquisition unit, multispectral, high-resolution RGB textures and elevation / depth physical data are integrated into a multi-dimensional image set, which can maintain a stable recognition rate and inversion robustness even under complex lighting conditions, different sediment types, and wave interference environments. The final output is visualized monitoring data after regional feature labeling and signal correction.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: a hydrological monitoring method based on unmanned aerial vehicles (UAVs), comprising a multi-source data acquisition unit, a water area zoning unit, a sediment feature database, a dynamic parameter compensation engine, and a water quality inversion unit, wherein:

[0008] The multi-source data acquisition unit acquires multispectral image data, visible light RGB image data, and elevation / depth detection data of the target water area simultaneously through a drone equipped with multiple sensor groups. After preprocessing, the data is integrated into a hydrological image set and sent to the water area partitioning unit.

[0009] The water area zoning unit is used to acquire and process hydrological image sets. Based on the preset band ratio of the multispectral image data and combined with elevation / depth detection data, a water depth spatial probability distribution model is established. Visible light RGB image data is input into the water depth spatial probability distribution model for feature division. The target water area is divided into deep water area, shallow water area and transition area. A water area zoning map is generated and sent to the dynamic parameter compensation engine and water quality inversion unit.

[0010] The sediment feature library is used to store standard reflectance spectral curves of different types of riverbeds at different depths, and to establish an indexing mechanism based on the feature vectors of different types of riverbeds;

[0011] The dynamic parameter compensation engine is used to obtain the water area division map and mark the shallow water area distribution area and the corresponding feature vector in the water area division map. For the pixels of the shallow water area distribution area, the radiative transfer equation deconvolution operation is performed according to the feature vector by calling the bottom sediment feature library to remove the bottom sediment reflection component and extract the pure water body reflection signal to be sent to the water quality inversion unit.

[0012] The water quality inversion unit is used to mark regional features based on the acquired water area delineation map and pure water body reflection signal, and to fuse the corrected pure water body reflection signal with the corresponding regional features to output visualized hydrological environment monitoring data.

[0013] Furthermore, the sensor group includes a multispectral imager, a high-resolution visible light camera, and a single-green lidar, wherein:

[0014] The multispectral imager includes four standard bands: blue, green, red, and near-infrared, used to capture the spectral characteristics of water bodies.

[0015] A high-resolution visible light camera is used to acquire RGB texture images with centimeter-level spatial resolution, providing fine features for subsequent semantic segmentation;

[0016] The single-green lidar uses 532nm green light and takes advantage of its water penetration characteristics to obtain dual echo signals from the water surface and bottom for water depth measurement.

[0017] Furthermore, the specific process of integrating the hydrological images is as follows:

[0018] S101. The sensor group uses the GNSS / IMU module on the UAV carrier board for unified timing, and records the current latitude and longitude, flight altitude and attitude angle when the shutter is triggered.

[0019] S102. Acquire multispectral image data, visible light RGB image data and elevation / depth detection data of the target water area. Using the high-resolution visible light RGB image as a reference, resample the multispectral image using a projection transformation matrix to ensure that each pixel is completely superimposed in spatial position.

[0020] S103. Real-time downlink irradiance recorded by the airborne light sensor equipped on the UAV is used to convert the raw digital quantities in the multispectral image data and visible light RGB image data into water reflectance with physical meaning, and the path radiation caused by atmospheric scattering is removed by the dark pixel method.

[0021] S104. The elevation / depth detection data acquired by the single green lidar is specifically a discrete depth point cloud. The discrete point cloud is transformed into a depth map of the same size as the RGB image through the Kriging interpolation algorithm, where each pixel value represents the real-time water depth data of the corresponding location.

[0022] S105. The water reflectance and depth maps are integrated into a hydrological image set using a multi-dimensional matrix form. The structure of the hydrological image set is defined as follows:

[0023] The first to third layers are R, G, and B color components, used to provide texture and shape features;

[0024] The fourth to seventh layers are multispectral reflectance, used to provide spectral response characteristics of chemical components;

[0025] The eighth layer is a normalized elevation / depth channel, used to provide physical space constraints.

[0026] Furthermore, the specific process for generating the water area delineation map is as follows:

[0027] S201. Obtain a set of hydrological images and extract multidimensional features based on the set of hydrological images. The multidimensional features include spectral feature terms, physical depth constraints, and visual texture features, wherein:

[0028] The spectral feature is obtained by calculating the ratio of a preset band in the multispectral image data. The preset band ratio is highly sensitive to changes in real-time water depth data and is used as an indirect indicator of real-time water depth data.

[0029] Physical depth constraints are obtained by extracting elevation / depth survey data and are used as hard constraint anchor points in the spatial probability distribution model of water depth.

[0030] Visual texture features are extracted from visible light RGB image data using feature operators to characterize the visual consistency of water surfaces.

[0031] S202. Using elevation / depth survey data as the real label, and the ratio of multispectral bands at the same location as the independent variable, a conditional probability distribution is established using Gaussian process regression. The spatial probability distribution model of the depth is defined as P(Hi|S,T), where Hi is the depth to be predicted, S is the spectral ratio feature, and T is the spatial location coordinate. Markov random field is used to spatially constrain the probability distribution.

[0032] S203. Input high-resolution visible light RGB image data into the water depth spatial probability distribution model. The water depth spatial probability distribution model assigns each pixel to the corresponding depth feature vector space based on the hue and saturation characteristics of the visible light RGB pixels and the mean of the probability density function output by the water depth spatial probability distribution model.

[0033] S204. Obtain the preset depth judgment interval (T1, T2). If the mean of the probability density function is less than T1, it is judged as a shallow water area, that is, an area with high visibility of bottom texture.

[0034] If the mean of the probability density function is less than T2 and greater than T1, it is judged to be a transition zone, where the spectral signal is dominated by water scattering, but the bottom sediment still contributes.

[0035] If the mean of the probability density function is greater than T2, it is judged to be a deep water area, which conforms to the optical characteristics of a semi-infinite water body, and the reflection of the bottom sediment can be ignored.

[0036] S205. Generate a multi-channel water area delineation map through a pixel classifier, generate binary masks for deep water area, shallow water area and transition area respectively, and perform morphological operations to smooth the boundaries. In the water area delineation map, mark the feature vector center value and its corresponding average confidence for each partition. Perform layer mapping between the water area delineation map and the original hydrological image set, and send it to the subsequent dynamic parameter compensation engine and water quality inversion unit.

[0037] Furthermore, the specific process for constructing the sediment feature library is as follows:

[0038] S301. Obtain geographic information and geological survey data of the target water area, specifically:

[0039] In a controlled experimental environment or during the dry season in the field, the initial reflectance spectra of various substrates under waterless conditions were collected using a high-precision spectroradiometer.

[0040] Repeated sampling was performed on each type of sediment with different moisture contents and particle size distributions to obtain the spectral fluctuation range of that sediment type, forming a sediment benchmark dataset;

[0041] Based on the sediment baseline dataset, the riverbed sediments are divided into several typical categories, including fine sand, gravel, silt, submerged plants, and mixed sediments.

[0042] S302. Introduce water attenuation coefficients of different levels based on the bio-optical model. The attenuation process of the seabed reflectance spectrum as it changes with real-time water depth data is simulated by the following formula:

[0043] Where H represents real-time water depth data. This represents a set of standard reflectance spectral curves, where e is a preset scaling factor. This is the initial reflectance spectrum;

[0044] S303. Store a set of standard reflectance spectral curves that vary with depth for each type of substrate. When H increases by a preset step size, the corresponding apparent spectral features are automatically generated.

[0045] S304. Extract the physical features of the standard reflectance spectral curve cluster and the texture features based on visible light images. The physical features include the position of the characteristic absorption peak, the spectral slope, and the band ratio. Integrate the physical features and the texture features to generate a unique feature vector V={f1, f2, f3, ..., fn} for each substrate type. Here, fn is the spectral shape descriptor and spatial distribution statistics.

[0046] S305. The indexing mechanism based on high-dimensional clustering is as follows:

[0047] Primary index: Coarse classification of substrate based on texture features of visible light RGB images;

[0048] Secondary index: Based on the detected real-time water depth data, the search scope is narrowed within the same type of bottom sediment;

[0049] Three-level index: Real-time feature vectors extracted by the dynamic parameter compensation engine are used to perform cosine similarity matching in the reduced sub-database.

[0050] Furthermore, the specific process for extracting the reflection signal from pure water is as follows:

[0051] S401. Obtain the water area division map and extract the set of pixel coordinates belonging to the shallow water area marked in the division map. Then, perform point-to-point mapping between the set of pixel coordinates of the shallow water area and the multispectral band reflectance data and corresponding real-time water depth data in the hydrological image set to obtain the pixel points of the shallow water area.

[0052] S402. For each pixel in the shallow water area, extract the substrate feature vector associated with the pixel, use the substrate feature vector to perform a similarity search in the substrate feature library through an indexing mechanism to determine the substrate type below the pixel, and retrieve the standard reflectance reference curve corresponding to the substrate type.

[0053] S403. Construct a radiative transfer model based on bio-optics. In this model, the total reflected signal received by the sensor is the superposition of contributions from water components and geological background.

[0054] Contribution of water components: This represents the effective signal generated by the scattering of suspended matter and organic matter in the water body and attenuated by the water layer;

[0055] Substrate background contribution: This indicates the interference signal that penetrates the water layer and is reflected back to the sensor by the riverbed;

[0056] in, For wavelength, K( ) represents the average diffuse decay coefficient of the water body, which is positively correlated with the turbidity of the water body itself;

[0057] S404 and K( All parameters are affected by the water quality parameters to be measured and are interdependent. Initial values ​​are established using a preset infrared band that is insensitive to the sediment, in order to estimate the average diffuse attenuation coefficient K of the water body in the area to be measured. The retrieved sediment background contribution and measured water depth data are substituted into the equation in S403, and K is continuously optimized through a recursive iterative algorithm. This allows for the precise extraction of the exponential decay term belonging to the substrate from multispectral reflectance data.

[0058] S405, Introduction of the Modification Operator Energy compensation is applied to the second-order reflection path, and the final inverse solution is obtained: R i ( ) represents multispectral reflectance data, Re( This means that the pure water body reflection signal, which excludes the influence of water depth and bottom sediment type, is repackaged into a pure spectral feature stream after correction of each band, and sent to the water quality inversion unit in real time for final parameter calculation.

[0059] Furthermore, the specific process for obtaining visualized hydrological environment monitoring data is as follows:

[0060] S501. Obtain the water area delineation map and the pure water reflection signal. Based on the mask information in the water area delineation map, assign a corresponding region label to each pixel.

[0061] S502. Logically integrate the original reflectance signal of the deep water area with the pure water reflectance signal of the shallow water area after compensation and correction to form a unified, interference-free spectral dataset across the entire domain.

[0062] S503, Partition-based differentiated inversion model matching, specifically implementing differentiated model invocation strategies:

[0063] Deepwater inversion: A standard bio-optical semi-analysis model is used to calculate water composition using multi-band combination;

[0064] Shallow water and transition zone inversion: Invokes an inversion operator specifically optimized for the correction signal and applies a mathematical model of preset water quality indicators to calculate the water composition.

[0065] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0066] This UAV-based hydrological monitoring system, by constructing a sediment feature database and performing deconvolution operations using the radiative transfer equation, can accurately remove background noise and extract pure water reflection signals, thereby eliminating pollution inversion phenomena in shallow water areas. Simultaneously, through a water depth spatial probability distribution model, it achieves deep coupling of spectral features and physical depth data. Through an integrated multi-source data acquisition unit, it integrates multispectral, high-resolution RGB textures and elevation / depth physical data into a multi-dimensional image set, maintaining stable recognition rates and inversion robustness even under complex lighting conditions, different sediment types, and wave interference. The final output is visualized monitoring data after regional feature labeling and signal correction, containing not only pollutant concentration distribution but also incorporating three-dimensional water topological features. This provides environmental protection departments with a scientific and intuitive decision-making basis for real-time early warning and precise management in small and medium-sized rivers and shallow water areas. Attached Figure Description

[0067] Figure 1 A schematic diagram of the overall method flow of the present invention is shown. Detailed Implementation

[0068] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0069] Example:

[0070] like Figure 1As shown, a hydrological monitoring method based on unmanned aerial vehicles (UAVs) includes a multi-source data acquisition unit, a water area zoning unit, a sediment characteristic database, a dynamic parameter compensation engine, and a water quality inversion unit, wherein:

[0071] The multi-source data acquisition unit acquires multispectral image data, visible light RGB image data, and elevation / depth detection data of the target water area simultaneously through a drone equipped with multiple sensor groups. After preprocessing, the data is integrated into a hydrological image set and sent to the water area partitioning unit.

[0072] The sensor suite includes a multispectral imager, a high-resolution visible light camera, and a single-green lidar, among which:

[0073] A multispectral imager, containing four standard bands—blue, green, red, and near-infrared—is used to capture the spectral characteristics of water bodies.

[0074] A high-resolution visible light camera is used to acquire RGB texture images with centimeter-level spatial resolution, providing fine features for subsequent semantic segmentation;

[0075] The single-green lidar uses 532nm green light and takes advantage of its water penetration characteristics to obtain dual echo signals from the water surface and bottom for water depth measurement.

[0076] The specific process of integrating and obtaining the hydrological image set is as follows:

[0077] S101: The sensor group uses the GNSS / IMU module on the UAV carrier board for unified timing, and records the current latitude and longitude, flight altitude and attitude angle when the shutter is triggered.

[0078] S102. Acquire multispectral image data, visible light RGB image data and elevation / depth detection data of the target water area. Using the high-resolution visible light RGB image as a reference, resample the multispectral image using a projection transformation matrix to ensure that each pixel is completely superimposed in spatial position.

[0079] S103. Real-time downlink irradiance recorded by the airborne light sensor equipped on the UAV is used to convert the raw digital quantities in the multispectral image data and visible light RGB image data into water reflectance with physical meaning, and the path radiation caused by atmospheric scattering is removed by the dark pixel method.

[0080] The elevation / depth detection data acquired by the S104 single-green lidar is specifically a discrete depth point cloud. Through the Kriging interpolation algorithm, the discrete point cloud is transformed into a depth map of the same size as the RGB image, where each pixel value represents the real-time water depth data of the corresponding location.

[0081] S105. The water reflectance and depth maps are integrated into a hydrological image set using a multidimensional matrix form. The structure of the hydrological image set is defined as follows:

[0082] The first to third layers are R, G, and B color components, used to provide texture and shape features;

[0083] The fourth to seventh layers are multispectral reflectance, used to provide spectral response characteristics of chemical components;

[0084] The eighth layer is a normalized elevation / depth channel, used to provide physical space constraints.

[0085] The water area zoning unit is used to acquire and process hydrological image sets. Based on the preset band ratio of multispectral image data and combined with elevation / depth detection data, a water depth spatial probability distribution model is established. Visible light RGB image data is input into the water depth spatial probability distribution model for feature division. The target water area is divided into deep water area, shallow water area and transition area. A water area zoning map is generated and sent to the dynamic parameter compensation engine and water quality inversion unit.

[0086] The specific process for generating a water area delineation map is as follows:

[0087] S201. Obtain a set of hydrological images and extract multidimensional features from the set. The multidimensional features include spectral features, physical depth constraints, and visual texture features, among which:

[0088] The spectral feature is obtained by calculating the ratio of preset bands in multispectral image data. The preset band ratio is highly sensitive to changes in real-time water depth data and is used as an indirect indicator of real-time water depth data.

[0089] Physical depth constraints are obtained by extracting elevation / depth survey data and are used as hard constraint anchor points in the spatial probability distribution model of water depth.

[0090] Visual texture features are extracted from visible light RGB image data using feature operators to characterize the visual consistency of water surfaces.

[0091] S202. Using elevation / depth survey data as the real label and the multispectral band ratio at the same location as the independent variable, a conditional probability distribution is established using Gaussian process regression. The spatial probability distribution model of depth is defined as P(Hi|S,T), where Hi is the depth to be predicted, S is the spectral ratio feature, and T is the spatial coordinate. The spatial probability distribution model of depth not only outputs the predicted depth but also the confidence interval of that depth. Markov random fields are used to spatially constrain the probability distribution to ensure that the partitioning results of adjacent pixels have logical continuity and avoid isolated misjudgment points.

[0092] S203. Input high-resolution visible light RGB image data into the water depth spatial probability distribution model. The water depth spatial probability distribution model assigns each pixel to the corresponding depth feature vector space based on the hue and saturation characteristics of the visible light RGB pixels and the mean of the probability density function output by the water depth spatial probability distribution model.

[0093] S204. Obtain the preset depth judgment interval (T1, T2). If the mean of the probability density function is less than T1, it is judged as a shallow water area, that is, an area with high visibility of bottom texture.

[0094] If the mean of the probability density function is less than T2 and greater than T1, it is judged to be a transition zone, where the spectral signal is dominated by water scattering, but the bottom sediment still contributes.

[0095] If the mean of the probability density function is greater than T2, it is judged to be a deep water area, which conforms to the optical characteristics of a semi-infinite water body, and the reflection of the bottom sediment can be ignored.

[0096] S205. Generate a multi-channel water area delineation map through a pixel classifier, generate binary masks for deep water area, shallow water area and transition area respectively, and perform morphological operations to smooth the boundaries. In the water area delineation map, mark the feature vector center value and its corresponding average confidence for each partition. Perform layer mapping between the water area delineation map and the original hydrological image set, and send it to the subsequent dynamic parameter compensation engine and water quality inversion unit.

[0097] The sediment feature library is used to store standard reflectance spectral curves of different types of riverbeds at different depths, and to establish an indexing mechanism based on the feature vectors of different types of riverbeds;

[0098] The specific process of constructing the sediment feature library is as follows:

[0099] S301. Obtain geographic information and geological survey data of the target water area, specifically:

[0100] In a controlled experimental environment or during the dry season in the field, the initial reflectance spectra of various substrates under waterless conditions were collected using a high-precision spectroradiometer.

[0101] Repeated sampling was performed on each type of sediment with different moisture contents and particle size distributions to obtain the spectral fluctuation range of that sediment type, forming a sediment benchmark dataset;

[0102] Based on the sediment baseline dataset, the riverbed sediments are divided into several typical categories, including fine sand, gravel, silt, submerged plants, and mixed sediments.

[0103] S302. Introduce water attenuation coefficients of different levels based on the bio-optical model. The attenuation process of the seabed reflectance spectrum as it changes with real-time water depth data is simulated by the following formula:

[0104] Where H represents real-time water depth data. This represents a set of standard reflectance spectral curves, where e is a preset scaling factor. This is the initial reflectance spectrum;

[0105] S303. Store a set of standard reflectance spectral curves that vary with depth for each type of substrate. When H increases by a preset step size, the corresponding apparent spectral characteristics are automatically generated to ensure that there is a reference benchmark under different water depth conditions.

[0106] S304. Extract the physical features of the standard reflectance spectral curve cluster and the texture features based on visible light images. The physical features include the position of the characteristic absorption peak, the spectral slope, and the band ratio. Integrate the physical features and texture features to generate a unique feature vector V={f1, f2, f3, ..., fn} for each substrate type. Here, fn is the spectral shape descriptor and spatial distribution statistics.

[0107] S305. The indexing mechanism based on high-dimensional clustering is as follows:

[0108] Primary index: Coarse classification of substrate based on texture features of visible light RGB images;

[0109] Secondary index: Based on the detected real-time water depth data, the search scope is narrowed within the same type of bottom sediment;

[0110] Three-level index: Real-time feature vectors extracted by the dynamic parameter compensation engine are used to perform cosine similarity matching in the reduced sub-database.

[0111] The dynamic parameter compensation engine is used to obtain the water area division map and mark the shallow water area distribution area and the corresponding feature vector in the water area division map. For the pixels in the shallow water area distribution area, the radiative transfer equation deconvolution operation is performed according to the feature vector by calling the bottom sediment feature library to remove the bottom sediment reflection component and extract the pure water body reflection signal to send to the water quality inversion unit.

[0112] The specific process for extracting the reflection signal from pure water is as follows:

[0113] S401. Obtain the water area division map and extract the set of pixel coordinates belonging to the shallow water area marked in the division map. Then, perform point-to-point mapping between the set of pixel coordinates of the shallow water area and the multispectral band reflectance data and corresponding real-time water depth data in the hydrological image set to obtain the pixel points of the shallow water area.

[0114] S402. For each pixel in the shallow water area, extract the substrate feature vector associated with the pixel, use the substrate feature vector to perform a similarity search in the substrate feature library through an indexing mechanism, determine the substrate type below the pixel, and retrieve the standard reflectance baseline curve corresponding to the substrate type.

[0115] S403. Construct a radiative transfer model based on bio-optics. In this model, the total reflected signal received by the sensor is the superposition of contributions from water components and geological background.

[0116] Contribution of water components: This represents the effective signal generated by the scattering of suspended matter and organic matter in the water body and attenuated by the water layer;

[0117] Substrate background contribution: This indicates the interference signal that penetrates the water layer and is reflected back to the sensor by the riverbed;

[0118] in, For wavelength, K( ) represents the average diffuse decay coefficient of the water body, which is positively correlated with the turbidity of the water body itself;

[0119] S404 and K( All parameters are affected by the water quality parameters to be measured and are interdependent. Initial values ​​are established using a preset infrared band that is insensitive to the sediment, in order to estimate the average diffuse attenuation coefficient K of the water body in the area to be measured. The retrieved sediment background contribution and measured water depth data are substituted into the equation in S403, and K is continuously optimized through a recursive iterative algorithm. This allows for the precise extraction of the exponential decay term belonging to the substrate from multispectral reflectance data.

[0120] S405, Introduction of the Modification Operator Energy compensation is applied to the second-order reflection path, and the final inverse solution is obtained: R i ( ) represents multispectral reflectance data, Re( This means that the pure water body reflection signal, which excludes the influence of water depth and bottom sediment type, is repackaged into a pure spectral feature stream after correction of each band, and sent to the water quality inversion unit in real time for final parameter calculation.

[0121] The extracted pure water reflection signal needs to pass through the self-checking module of the dynamic parameter compensation engine to ensure data quality:

[0122] If the calculated pure reflectance is negative or exceeds the physically reasonable range of this type of water, the pixel will be marked as an abnormal acquisition area, and the neighborhood spatial autocorrelation algorithm will be automatically invoked for smoothing compensation.

[0123] The water quality inversion unit is used to mark regional features based on the acquired water area delineation map and pure water body reflection signal, and to fuse the corrected pure water body reflection signal with the corresponding regional features to output visualized hydrological environment monitoring data.

[0124] The specific process for obtaining visualized hydrological environment monitoring data is as follows:

[0125] S501. Obtain the water area delineation map and the pure water reflection signal. Based on the mask information in the water area delineation map, assign a corresponding region label to each pixel.

[0126] S502. Logically integrate the original reflectance signal of the deep water area with the pure water reflectance signal of the shallow water area after compensation and correction to form a unified, interference-free spectral dataset across the entire domain.

[0127] S503, Partition-based differentiated inversion model matching, specifically implementing differentiated model invocation strategies:

[0128] Deepwater inversion: A standard bio-optical semi-analysis model is used to calculate water composition using multi-band combination;

[0129] Shallow water and transition zone inversion: Invokes an inversion operator specifically optimized for the correction signal and applies a mathematical model of preset water quality indicators to calculate the water composition.

[0130] The size of the interval and threshold is set to facilitate comparison. The size of the threshold depends on the amount of sample data and the number of bases set by those skilled in the art for each set of sample data; as long as it does not affect the ratio between the parameter and the quantized value.

[0131] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0132] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A hydrological monitoring method based on unmanned aerial vehicles (UAVs), characterized in that, Includes the following steps: Step 1: Simultaneously acquire multispectral image data, visible light RGB image data, and elevation / depth detection data of the target water area using a drone equipped with multiple sensor groups, and integrate them into a hydrological image set after preprocessing; Step 2: Acquire and process the hydrological image set. Based on the preset band ratio of the multispectral image data and the elevation / depth detection data, establish a water depth spatial probability distribution model. Input the visible light RGB image data into the water depth spatial probability distribution model for feature division. Divide the target water area into deep water area, shallow water area and transition area to generate a water area division map. Step 3: Pre-set a sediment feature library, which is used to store the standard reflectance spectral curves of different types of riverbeds at different depths, and to establish an indexing mechanism based on the feature vectors of different types of riverbeds; Step 4: Obtain the water area division map and mark the shallow water area distribution area and the corresponding feature vector in the water area division map. For the pixels of the shallow water area distribution area, call the bottom sediment feature library to perform radiative transfer equation deconvolution operation according to the feature vector to remove the bottom sediment reflection component and extract the pure water body reflection signal. Step 5: Based on the acquired water area delineation map and pure water body reflection signal, mark the regional features according to the zoning results in the water area delineation map, and fuse the corrected pure water body reflection signal with the corresponding regional features to output visualized hydrological environment monitoring data.

2. The hydrological monitoring method based on unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, The sensor group involved in step one includes a multispectral imager, a high-resolution visible light camera, and a single-green lidar, wherein: The multispectral imager includes four standard bands: blue, green, red, and near-infrared, used to capture the spectral characteristics of water bodies. A high-resolution visible light camera is used to acquire RGB texture images with centimeter-level spatial resolution, providing fine features for subsequent semantic segmentation; The single-green lidar uses 532nm green light and takes advantage of its water penetration characteristics to obtain dual echo signals from the water surface and bottom for water depth measurement.

3. The hydrological monitoring method based on unmanned aerial vehicles (UAVs) according to claim 2, characterized in that, The specific process of integrating and obtaining the hydrological image set is as follows: S101. The sensor group uses the GNSS / IMU module on the UAV carrier board for unified timing, and records the current latitude and longitude, flight altitude and attitude angle when the shutter is triggered. S102. Acquire multispectral image data, visible light RGB image data and elevation / depth detection data of the target water area. Using the high-resolution visible light RGB image as a reference, resample the multispectral image using a projection transformation matrix to ensure that each pixel is completely superimposed in spatial position. S103. Real-time downlink irradiance recorded by the airborne light sensor equipped on the UAV is used to convert the raw digital quantities in the multispectral image data and visible light RGB image data into water reflectance with physical meaning, and the path radiation caused by atmospheric scattering is removed by the dark pixel method. S104. The elevation / depth detection data acquired by the single green lidar is specifically a discrete depth point cloud. The discrete point cloud is transformed into a depth map of the same size as the RGB image through the Kriging interpolation algorithm, where each pixel value represents the real-time water depth data of the corresponding location. S105. Integrate the water-leaving reflectance and depth map into a hydrological image set in a multi-dimensional matrix form, and the structure of the hydrological image set is defined as follows: The first to third layers are the R, G, and B color components, which are used to provide texture and shape features; The fourth to seventh layers are multi-spectral band reflectances, which are used to provide spectral response features of chemical components; The eighth layer is the normalized elevation / water depth channel, which is used to provide physical space constraints.

4. The hydrological monitoring method based on unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, The specific process of generating the water area division map is as follows: S201. Obtain the hydrological image set and perform multi-dimensional feature extraction according to the hydrological image set. The multi-dimensional features include spectral feature terms, physical depth constraints, and visual texture features, where: The spectral feature terms are obtained by calculating the preset band ratios in the multi-spectral image data. The preset band ratios are highly sensitive to changes in real-time water depth data and are used as indirect indicator factors for real-time water depth data; The physical depth constraints are obtained by extracting elevation / water depth detection data and are used as hard constraint anchor points in the water depth spatial probability distribution model; The visual texture features are obtained by extracting feature operators from visible light RGB image data and are used to characterize the visual consistency of the water surface; S202. Using the elevation / water depth detection data as the true label and the multi-spectral band ratio at the same location as the independent variable, establish a conditional probability distribution using Gaussian process regression. Define the water depth spatial probability distribution model as P(Hi|S, T), where Hi is the depth to be predicted, S is the spectral ratio feature, and T is the spatial position coordinate. Use the Markov random field to perform spatial constraints on the probability distribution; S203. Input the high-resolution visible light RGB image data into the water depth spatial probability distribution model. The water depth spatial probability distribution model assigns each pixel to the corresponding depth feature vector space according to the hue and saturation features of the visible light RGB pixels and combines the mean value of the probability density function output by the water depth spatial probability distribution model; S204. Obtain the preset depth judgment interval (T1, T2), where T1 and T2 are any positive numbers preset, and T1 < T2. If the mean value of the probability density function is less than T1, it is judged as a shallow water area, that is, an area with high visibility of bottom texture; If the mean value of the probability density function is less than T2 and greater than T1, it is judged as a transition area, an area where the spectral signal is dominated by water body scattering but the bottom still makes a contribution; If the mean value of the probability density function is greater than T2, it is judged as a deep water area, which conforms to the optical characteristics of a semi-infinite water body and the bottom reflection can be ignored; S205. Generate a multi-channel water area division map through a pixel classifier, generate binary masks for the deep water area, shallow water area, and transition area respectively, perform morphological operations to smooth the boundaries, and mark the central value of the feature vector and its corresponding average confidence for each partition in the water area division map. Layer and map the water area division map with the original hydrological image set and send it to the subsequent dynamic parameter compensation engine and water quality inversion unit.

5. A hydrological monitoring method based on unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, The specific process of constructing the bottom texture feature library is as follows: S301. Obtain the geographical information and geological survey data of the target water area. Specifically: In a controlled experimental environment or during the dry season in the field, the initial reflectance spectra of various substrates under waterless conditions were collected using a high-precision spectroradiometer. Repeated sampling was performed on each type of sediment with different moisture contents and particle size distributions to obtain the spectral fluctuation range of that sediment type, forming a sediment benchmark dataset; Based on the sediment baseline dataset, the riverbed sediments are divided into several typical categories, including fine sand, gravel, silt, submerged plants, and mixed sediments. S302. Introduce water attenuation coefficients of different levels based on the bio-optical model. The attenuation process of the seabed reflectance spectrum as it changes with real-time water depth data is simulated by the following formula: Where H represents real-time water depth data. This represents a set of standard reflectance spectral curves, where e is a preset scaling factor. This is the initial reflectance spectrum; S303. Store a set of standard reflectance spectral curves that vary with depth for each type of substrate. When H increases by a preset step size, the corresponding apparent spectral features are automatically generated. S304. Extract the physical features of the standard reflectance spectral curve cluster and the texture features based on visible light images. The physical features include the position of the characteristic absorption peak, the spectral slope, and the band ratio. Integrate the physical features and the texture features to generate a unique feature vector V={f1, f2, f3, ..., fn} for each substrate type. Here, fn is the spectral shape descriptor and spatial distribution statistics. S305. The indexing mechanism based on high-dimensional clustering is as follows: Primary index: Coarse classification of substrate based on texture features of visible light RGB images; Secondary index: Based on the detected real-time water depth data, the search scope is narrowed within the same type of bottom sediment; Three-level index: Real-time feature vectors extracted by the dynamic parameter compensation engine are used to perform cosine similarity matching in the reduced sub-database.

6. The hydrological monitoring method based on unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, The specific process for extracting the reflection signal from pure water is as follows: S401. Obtain the water area division map and extract the set of pixel coordinates belonging to the shallow water area marked in the division map. Then, perform point-to-point mapping between the set of pixel coordinates of the shallow water area and the multispectral band reflectance data and corresponding real-time water depth data in the hydrological image set to obtain the pixel points of the shallow water area. S402. For each pixel in the shallow water area, extract the substrate feature vector associated with the pixel, use the substrate feature vector to perform a similarity search in the substrate feature library through an indexing mechanism to determine the substrate type below the pixel, and retrieve the standard reflectance reference curve corresponding to the substrate type. S403. Construct a radiative transfer model based on bio-optics. In this model, the total reflected signal received by the sensor is the superposition of contributions from water components and geological background. Contribution of water components: This represents the effective signal generated by scattering from suspended solids and organic matter in the water body and attenuated by the water layer, where... The data consists of multispectral band reflectance data from a hydrological image set, where EX is a preset weighting coefficient and H is real-time water depth data. Substrate background contribution: This represents the interference signal that penetrates the water layer and is reflected back to the sensor by the riverbed. The standard reflectance reference curve is retrieved from the substrate feature library to correspond to the substrate type. in, For wavelength, K( ) represents the average diffuse decay coefficient of the water body, which is positively correlated with the turbidity of the water body itself; S404, and K( All parameters are affected by the water quality parameters to be measured and are interdependent. Initial values ​​are established using a preset infrared band that is insensitive to the sediment, in order to estimate the average diffuse attenuation coefficient K of the water body in the area to be measured. The retrieved sediment background contribution and measured water depth data are substituted into the equation in S403, and K is continuously optimized through a recursive iterative algorithm. This allows for the precise extraction of the exponential decay term belonging to the substrate from multispectral reflectance data. S405, Introduction of the Modification Operator Energy compensation is applied to the second-order reflection path, and the final inverse solution is obtained: , where R i ( ) represents multispectral reflectance data, Re( This means that the pure water body reflection signal, which excludes the influence of water depth and bottom sediment type, is repackaged into a pure spectral feature stream after correction of each band, and sent to the water quality inversion unit in real time for final parameter calculation.

7. A hydrological monitoring method based on unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, The specific process for obtaining visualized hydrological environment monitoring data is as follows: S501. Obtain the water area delineation map and the pure water reflection signal. Based on the mask information in the water area delineation map, assign a corresponding region label to each pixel. S502. Logically integrate the original reflectance signal of the deep water area with the pure water reflectance signal of the shallow water area after compensation and correction to form a unified, interference-free spectral dataset across the entire domain. S503, Partition-based differentiated inversion model matching, specifically implementing differentiated model invocation strategies: Deepwater inversion: A standard bio-optical semi-analysis model is used to calculate water composition using multi-band combination; Shallow water and transition zone inversion: Invokes an inversion operator specifically optimized for the correction signal and applies a mathematical model of preset water quality indicators to calculate the water composition.

Images