Method, system and equipment for studying and judging river-lake four-disorder problem of water conservancy

Abstract

The invention discloses a research and judgment method, system and equipment for four disordered problems of water conservancy rivers and lakes, and particularly relates to the field of research and judgment of water conservancy rivers and lakes, and the method comprises the steps: constructing a multi-source heterogeneous data set, and extracting dynamic environment features to generate a feature data set of a target water area; deploying a lightweight recognition engine at an edge computing node to perform dimension reduction on the features to generate four types of lightweight feature packets; a multi-dimensional decision graph is generated through space-time correlation analysis; and finally, the visual early warning platform is driven to realize dynamic visualization of the four disordered problems. According to the method, the system and the equipment for studying and judging the four-disorder problem of the river and the lake in the water conservancy, automatic alignment and efficient fusion of multi-source data are realized by controlling the space-time deviation within a threshold value, and the manual intervention requirement and the preprocessing computing resource consumption are remarkably reduced; the model parameter adjustment frequency is reduced through a radiation correction coefficient and a dynamic weight distribution strategy; and through space-time correlation analysis, a multi-dimensional scientific basis is provided for decision making, and the treatment efficiency is optimized.

Description

Technical Field

[0001] The present invention relates to the technical field of water conservancy river and lake research and determination, and more specifically, the present invention relates to a method, system and device for judging the "Four Irregularities" problems of water conservancy rivers and lakes. Background Art

[0002] The "Four Irregularities" problems in rivers and lakes have developed into key challenges threatening water resource security and ecological balance; the traditional manual inspection mode generally has limitations of low efficiency and narrow coverage. Although unmanned aerial vehicle (UAV) aerial photography and satellite remote sensing technologies have been gradually applied to the field of river and lake monitoring, there are still obvious shortcomings in the intelligent identification, accurate judgment and rapid disposal of the "Four Irregularities" problems.

[0003] At present, at the data collection level, high-precision aerial photography monitoring is carried out on key areas through UAVs, and automated screening of the "Four Irregularities" problems is carried out in combination with computer vision algorithms, effectively improving the monitoring efficiency and coverage; at the data analysis level, basic geographical information of rivers and lakes, real-time monitoring data and historical treatment records are integrated, and machine learning algorithms are used to realize intelligent classification of problems and risk prediction; at the decision support level, an early warning model is established relying on spatio-temporal big data analysis technology to provide a scientific basis for the dynamic optimization of treatment plans.

[0004] However, in actual use, there are still some disadvantages. For example, due to differences in spatio-temporal resolution and format standards among UAV aerial photography, satellite remote sensing and geographic information system data, manual intervention is required to complete alignment and association, resulting in high consumption of computing resources in the data preprocessing link; the adaptability to dynamic environments is weak, and existing algorithms are prone to feature misjudgment in scenarios of complex lighting, vegetation occlusion or seasonal hydrological changes, and model parameters need to be repeatedly adjusted to maintain the recognition accuracy; the real-time processing ability at the edge is limited, and problems such as large bandwidth occupancy and high response latency occur when high-resolution image data is transmitted back to the cloud, restricting the on-site rapid judgment ability. Therefore, a method, system and device for judging the "Four Irregularities" problems of water conservancy rivers and lakes are needed to improve the timeliness of monitoring and early warning and the scientific nature of decision-making optimization. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, the present invention provides a method, system and device for judging the "Four Irregularities" problems of water conservancy rivers and lakes, and through the following solutions, the problems raised in the above-mentioned background art are solved.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A method for judging the "Four Irregularities" problems of water conservancy rivers and lakes, comprising:

[0008] S1: Obtain the real-time images of the target water area by using a multi-spectral sensor carried by a drone, and simultaneously receive satellite remote sensing data, geographic information system vector data, and real-time environmental perception data including light intensity, vegetation coverage index, and hydrological parameters, and construct a multi-source heterogeneous data set of the target water area;

[0009] S2: Extract the dynamic environmental features from the multi-source heterogeneous data set to generate the feature data set of the target water area;

[0010] S3: Pre-deploy a lightweight recognition engine on the edge computing node, perform region segmentation and feature dimensionality reduction on the feature data set to generate four types of lightweight feature packets;

[0011] S4: Perform spatio-temporal correlation analysis on the four types of lightweight feature packets and the historical governance database to generate the multi-dimensional decision-making map of the target water area;

[0012] S5: Drive the visualization warning platform based on the multi-dimensional decision-making map of the target water area to realize the dynamic visualization of the four chaos problems in the target water area.

[0013] Preferably, in S1, constructing the multi-source heterogeneous data set of the target water area specifically includes:

[0014] Control the time deviation of multi-source data acquisition within a preset threshold range through the time synchronization protocol of the drone flight control software and the satellite data receiving end;

[0015] Perform coordinate transformation on the multi-source data based on the WGS84 geographic coordinate system, and when the grid scale difference between the GIS vector data and the satellite remote sensing data exceeds the preset threshold, call the spatial resampling function to perform interpolation calculation. In the spatial resampling function, dynamically select the interpolation algorithm type according to the spectral characteristics of the real-time environmental perception data;

[0016] Generate a multi-source heterogeneous data set represented by a multi-source data fusion matrix.

[0017] Preferably, in S2, extracting the dynamic environmental features from the multi-source heterogeneous data set specifically includes:

[0018] Construct an adaptive spatial registration weight matrix based on the light intensity and vegetation coverage index of the real-time environmental perception data to register the multi-source data;

[0019] Use the Gaussian-Laplacian pyramid fusion model to perform multi-scale denoising on the registered multi-source data;

[0020] Generate an anti-interference environmental feature map through the dynamic environmental feature extraction model. The anti-interference environmental feature map includes a radiation correction multi-spectral feature layer, a dynamic mask feature layer, and a temporal change feature layer;

[0021] Fuse the data of each layer of the anti-interference environment feature map to generate a feature dataset.

[0022] Preferably, in S2, the dynamic environment feature extraction model is a composite neural network architecture including a light intensity compensation layer, a vegetation occlusion attention mechanism layer, and a hydrological cycle feature embedding layer, where:

[0023] The light intensity compensation layer calculates a radiation correction coefficient based on the light intensity and performs adaptive brightness compensation on the multispectral image;

[0024] The vegetation occlusion attention mechanism layer generates an occlusion area mask according to the vegetation coverage index and dynamically assigns attention weights to the feature extraction channels;

[0025] The hydrological cycle feature embedding layer encodes hydrological parameters into periodic feature vectors and performs feature-level fusion with satellite remote sensing data.

[0026] Preferably, in S2, in the light intensity compensation layer, based on the band reflectance R b of the multispectral image, calculate the radiation correction coefficient γ b , which is specifically expressed as:

[0027]

[0028] Among them, L ref represents the calibrated light intensity, L represents the current light intensity, β represents the environmental sensitivity coefficient, and R avg represents the current scene average reflectance;

[0029] In the vegetation occlusion attention mechanism layer, generate a binary mask M(l,v) according to the vegetation coverage index; calculate the dynamic weight w c of the convolution channels in the mask area, which is specifically expressed as:

[0030]

[0031] Among them, F c (l,v) represents the feature map of the c-th channel, γ1 represents the reference weight, and γ2 represents the weight adjustment coefficient.

[0032] Preferably, in S3, perform regional segmentation and feature dimensionality reduction on the feature dataset, specifically including:

[0033] Based on the geographical grid coding rule of the target water area, segment the high-resolution image into block units of a preset size through a dynamic block algorithm;

[0034] Perform sparse representation on each block unit to generate an optimized sparse basis for discrete cosine transform;

[0035] Construct an adaptive measurement matrix based on real-time environmental perception data;

[0036] Generate four types of lightweight feature packets through compressive sampling.

[0037] Preferably, in step S4, the multi-dimensional decision-making map includes the risk levels, disposal priorities, associated geographical grid codes, and quarterly timestamps corresponding to the types of problems such as illegal occupation, illegal mining, illegal piling, and illegal construction.

[0038] Preferably, in step S5, the dynamic visualization method for the four types of illegal problems includes the flood discharge capacity impact coefficient matrix, grid-based risk probability, and two-dimensional gradient color scale.

[0039] To achieve the above object, the present invention also provides the following technical solution: A judgment system for four types of illegal problems in water conservancy rivers and lakes, applicable to a judgment method for four types of illegal problems in water conservancy rivers and lakes, includes:

[0040] Water area data acquisition module: Obtain real-time images of the target water area through a drone equipped with a multi-spectral sensor, and simultaneously receive satellite remote sensing data, geographic information system vector data, and real-time environmental perception data including light intensity, vegetation coverage index, and hydrological parameters, and construct a multi-source heterogeneous data set of the target water area;

[0041] Dynamic environmental feature extraction module: Extract dynamic environmental features from the multi-source heterogeneous data set to generate a feature data set of the target water area;

[0042] Four types of illegal feature segmentation module: Pre-deploy a lightweight recognition engine at the edge computing node, perform region segmentation and feature dimensionality reduction on the feature data set, and generate four types of lightweight feature packets;

[0043] Spatio-temporal analysis module: Perform spatio-temporal correlation analysis on the four types of lightweight feature packets and the historical governance database to generate a multi-dimensional decision-making map of the target water area;

[0044] Visualization warning module: Drive a visualization warning platform based on the multi-dimensional decision-making map of the target water area to realize the dynamic visualization of the four types of illegal problems in the target water area.

[0045] To achieve the above object, the present invention also provides an electronic device, including a system central processing unit, a user information terminal, and a system operation database, characterized in that the system operation database is used to store one or more computer program instructions, wherein the one or more computer program instructions are executed by the system central processing unit to implement a judgment method for four types of illegal problems in water conservancy rivers and lakes.

[0046] The technical effects and advantages of the present invention:

[0047] 1. Through the time synchronization protocol and dynamic interpolation algorithm, the present invention controls the spatio-temporal deviation of UAV images, satellite remote sensing, and GIS data within a threshold, and dynamically selects the type of interpolation algorithm according to real-time environmental data, achieving automatic alignment and efficient fusion of multi-source data, significantly reducing the need for manual intervention and the consumption of preprocessing computing resources;

[0048] 2. Through the radiation correction coefficient and dynamic weight allocation strategy, the present invention improves the robustness of feature extraction in scenarios of complex illumination, vegetation occlusion, and seasonal hydrological changes, reducing the frequency of model parameter adjustment;

[0049] 3. By generating the risk level, disposal priority, and geocoding of the four-disorder problems based on spatio-temporal correlation analysis, the present invention provides a multi-dimensional scientific basis for decision-making and optimizes the governance efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 It is a flowchart of the steps of a method for judging four-disorder problems of water conservancy rivers and lakes provided according to an embodiment of the present application.

[0051] Figure 2 It is a decision tree diagram for dynamically selecting an interpolation algorithm in a method for judging four-disorder problems of water conservancy rivers and lakes provided according to an embodiment of the present application.

[0052] Figure 3 It is a structural block diagram of a system for judging four-disorder problems of water conservancy rivers and lakes provided according to an embodiment of the present application.

[0053] Figure 4 It is a schematic structural diagram of an electronic device provided according to an embodiment of the present application.

[0054] Description of the reference numerals: 400, schematic structural diagram of an electronic device; 401, system central processing unit; 402, communication bus; 403, system operation database; 404, user information terminal. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0055] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0056] The terms used in the following embodiments of the present application are only for the purpose of describing specific embodiments and are not intended to limit the present application. As used in the specification of the present application, the singular forms "a", "an", "the", "above-mentioned", "said", and "this" are also intended to include the plural forms, unless there is a clear indication to the contrary in the context. It should also be understood that the term "and / or" used in the present application refers to and includes any or all possible combinations of one or more of the listed items; in the description of the embodiments of the present application, unless otherwise specified, the meaning of "plural" is two or more.

[0057] In a first aspect, refer to Figure 1 A method for judging the four chaos problems of water conservancy rivers and lakes shown in the figure, which is applicable to the judging system of the four chaos problems of water conservancy rivers and lakes. The method includes:

[0058] S1: Obtain the real-time image of the target water area through a drone equipped with a multispectral sensor, and synchronously receive satellite remote sensing data, geographic information system vector data, and real-time environmental perception data including light intensity, vegetation coverage index, and hydrological parameters, and construct a multi-source heterogeneous data set of the target water area;

[0059] S2: Extract the dynamic environmental characteristics of the multi-source heterogeneous data set to generate a characteristic data set of the target water area;

[0060] S3: Pre-deploy a lightweight recognition engine at the edge computing node, perform region segmentation and feature dimensionality reduction on the characteristic data set, and generate four types of lightweight feature packages;

[0061] S4: Perform spatio-temporal correlation analysis on the four types of lightweight feature packages and the historical governance database to generate a multi-dimensional decision-making map of the target water area;

[0062] S5: Drive a visualization warning platform based on the multi-dimensional decision-making map of the target water area to realize the dynamic visualization of the four chaos problems of the target water area.

[0063] It should be noted that the judging system for the four chaos problems of water conservancy rivers and lakes can be the judging system for the four chaos problems of water conservancy rivers and lakes described later, or an existing judging system for the four chaos problems of water conservancy rivers and lakes. Optionally, the judging system for the four chaos problems of water conservancy rivers and lakes can be a judging system for the four chaos problems of water conservancy rivers and lakes that has been built and is in operation, or the model data of the judging system for the four chaos problems of water conservancy rivers and lakes. The judging method shown in this embodiment can analyze the application of the judging system for the four chaos problems of water conservancy rivers and lakes during the experiment or simulation process, so as to improve the structure and operation mode of the existing judging system for the four chaos problems of water conservancy rivers and lakes.

[0064] Specifically, before performing the foregoing method, corresponding sensors can be installed in the judgment system for the four types of chaos problems in water conservancy rivers and lakes, including light sensors, flow velocity sensors, etc. The specific installation locations can be set according to actual needs. In this embodiment, they can be distributed in the judgment system for the four types of chaos problems in water conservancy rivers and lakes in a grid-like form, that is, a plurality of sensors are arranged at equal intervals in the target water area where the judgment system for the four types of chaos problems in water conservancy rivers and lakes is applied, so as to comprehensively collect the multiple structural data in the judgment system for the four types of chaos problems in water conservancy rivers and lakes by the water flow in space; the specific distribution density of the sensors can be set according to actual needs.

[0065] Specifically, in S1, the construction of the multi-source heterogeneous data set of the target water area includes:

[0066] S101: Through the time synchronization protocol of the UAV flight control software and the satellite data receiving end, control the time deviation of multi-source data collection within a preset threshold range. The multi-source data includes real-time image data obtained by the UAV, satellite remote sensing data, geographic information system vector data, and real-time environment perception data;

[0067] In this embodiment, a clock synchronization module based on improved PTP is embedded in the UAV flight control software. The clock synchronization module obtains the GNSS timing signal through the API interface of the satellite data receiving end and establishes a time reference system that coordinates software and hardware. Specifically, the UAV sends a time calibration request packet to the satellite data receiving end every 50 ms, and the receiving end returns a response data packet with a time stamp. By calculating the two-way transmission delay, dynamically compensate for the network jitter error, so that the time deviation between the UAV aerial photography image and the satellite remote sensing data collection time is stably controlled within the range of ±30 ms;

[0068] Further, calculate the two-way transmission delay to dynamically compensate for the network jitter error Δt, which is specifically expressed as:

[0069]

[0070] Among them, T1 represents the time when the UAV sends a request, T2 represents the time when the satellite end receives the request, T3 represents the time when the satellite end sends a response, and T4 represents the time when the UAV receives the response;

[0071] S102: Perform coordinate transformation on the multi-source data based on the WGS84 geographic coordinate system, and when the grid scale difference between the GIS vector data and the satellite remote sensing data exceeds the preset threshold, call the spatial resampling function to perform interpolation calculation;

[0072] Based on the WGS84 geographic coordinate system, this embodiment uniformly converts the local coordinate system of the UAV aerial images, the UTM projection coordinate system of the satellite remote sensing data, and the CGCS2000 coordinate system of the GIS vector data into longitude and latitude coordinates under the WGS84 ellipsoid. By using the spatial resolution difference detection algorithm to calculate the grid scale difference rate between the satellite remote sensing data and the GIS vector data, when the difference rate exceeds 8%, the spatial resampling function is triggered;

[0073] Further, the grid scale difference rate δ between the satellite remote sensing data and the GIS vector data is calculated by the spatial resolution difference detection algorithm G , specifically expressed as:

[0074]

[0075] Among them, G sat represents the satellite data grid size, and G GIS represents the GIS vector data grid size;

[0076] Furthermore, referring to Figure 2 , in the spatial resampling function, the interpolation algorithm type is dynamically selected according to the spectral characteristics of the real-time environment perception data. The selection strategies of the interpolation algorithm include but are not limited to the water area priority strategy, the vegetation coverage area strategy, the bare area strategy, etc. Among them, the water area priority strategy is that when the normalized difference water index > 0.2, the bilinear interpolation algorithm is used to reduce the sawtooth effect of the water body boundary; the vegetation coverage area strategy is that when the normalized difference vegetation index > 0.3, switch to the cubic convolution interpolation algorithm to retain the vegetation texture details, and the bare area strategy is that when the average reflectance of the visible light band < 0.15, the nearest neighbor interpolation algorithm is enabled to reduce the computational complexity; the default strategy in this embodiment uses bilinear interpolation, which ensures a balance between computational quality and computational efficiency;

[0077] S103: Generate a multi-source heterogeneous data set represented by a multi-source data fusion matrix;

[0078] Based on the UAV image acquisition time, this embodiment performs a time sliding window compensation with a window length of 5s on the satellite remote sensing data to eliminate the timing misalignment caused by transmission delay. At the same time, the multi-source data is encoded with a 1m×1m geographic grid unit, and each grid unit is associated with the metadata corresponding to the multi-source data. The metadata corresponding to the multi-source data includes the RGB-IR four-channel pixel values of the UAV image, the thermal infrared band radiation value of the satellite remote sensing data, the land use type code of the GIS vector data, the light intensity, vegetation coverage index, and hydrological parameters of the real-time environment perception data. Finally, the fusion outliers are eliminated through the spatial-spectral consistency verification algorithm to generate a spatio-temporally consistent multi-source data fusion matrix.

[0079] Specifically, in S2, data cleaning and spatio-temporal registration operations are performed on the multi-source heterogeneous data set, and dynamic environmental feature extraction and fusion are realized to generate a feature data set including;

[0080] S201: Based on the light intensity and vegetation coverage index of real-time environmental perception data, construct an adaptive spatial registration weight matrix, and perform non-rigid registration on the multi-source heterogeneous data set through an improved SURF algorithm, where the feature point matching threshold is dynamically adjusted according to the light intensity;

[0081] In this embodiment, based on the light intensity L and vegetation coverage index NDVI in the real-time environmental perception data, a spatial registration weight matrix W(l, v) is constructed, which is specifically expressed as:

[0082]

[0083] Among them, L represents the current light intensity, L0 represents the reference light threshold, k represents the adjustment coefficient, and l and v respectively represent the indices of the light intensity L and vegetation coverage index NDVI in the real-time environmental perception data. In this embodiment, L0 = 10 4 Lux, k = 0.1; when L < L0, the feature point matching threshold MT is dynamically reduced to compensate for feature loss under low light conditions;

[0084] Furthermore, the calculation formula for reducing the feature point matching threshold MT is specifically expressed as:

[0085]

[0086] Among them, MT0 represents the reference threshold, k0 represents the attenuation coefficient, L represents the current light intensity, and L0 represents the reference light threshold; it should be noted that the value of the reference threshold reaches the best matching accuracy when L = L0; the attenuation coefficient is the rate of controlling the change of the threshold with the light intensity;

[0087] In this embodiment, the CUDA parallel computing architecture is used to accelerate the matching of SURF descriptors of multi-source data, and the average error of non-rigid registration is controlled within 1.2 pixels;

[0088] S202: Use the Gaussian-Laplacian pyramid fusion model to perform multi-scale denoising on the registered multi-source data, where the number of pyramid layers is adaptively selected according to the image resolution, and the fusion coefficient is jointly calculated by the vegetation coverage index and hydrological parameters;

[0089] In this embodiment, the rule for adaptively selecting the number of pyramid layers according to the image resolution is as follows: when the image resolution R ≥ 1 m / pixel, construct a 5-layer pyramid; when 0.5 m / pixel ≤ R < 1 m / pixel, construct a 7-layer pyramid; when R < 0.5 m / pixel, construct a 9-layer pyramid;

[0090] In this embodiment, the hydrological parameters are calculated using flow velocity data;

[0091] Furthermore, a fusion coefficient α is jointly calculated from the vegetation cover index NDVI and the flow velocity vf in the hydrological parameters IF , specifically expressed as:

[0092]

[0093] where α IF1 and α IF2 respectively represent the weights of the vegetation cover index and the flow velocity in the hydrological parameters, vf max represents the preset maximum flow velocity, represents the ratio of the current flow velocity to the preset maximum flow velocity calculated through the activation function tanh;

[0094] S203: Generate an anti-interference environmental feature map from the denoised data through a dynamic environmental feature extraction model;

[0095] It should be noted that the dynamic environmental feature extraction model is a composite neural network architecture including a light intensity compensation layer, a vegetation occlusion attention mechanism layer, and a hydrological cycle feature embedding layer; the light intensity compensation layer calculates a radiation correction coefficient based on the light intensity and performs adaptive brightness compensation on the multispectral image; the vegetation occlusion attention mechanism layer generates an occlusion area mask according to the vegetation cover index and dynamically allocates the attention weights of the feature extraction channels; the hydrological cycle feature embedding layer encodes the hydrological parameters into periodic feature vectors and performs feature-level fusion with the satellite remote sensing data; the anti-interference environmental feature map includes a radiation-corrected multispectral feature layer, a dynamic mask feature layer, and a temporal change feature layer;

[0096] Furthermore, in the light intensity compensation layer, based on the band reflectance R b of the multispectral image, calculate the radiation correction coefficient γ b , specifically expressed as:

[0097]

[0098] where L ref represents the calibrated light intensity, L represents the current light intensity, β represents the environmental sensitivity coefficient, and R avg represents the current scene average reflectance; it should be noted that L is setref = 2 × 10 4 Lux, β = 0.1;

[0099] Furthermore, in the vegetation occlusion attention mechanism layer, a binary mask M(l, v) is generated according to the vegetation coverage index. When the vegetation coverage index ≥ 0.3, it is the vegetation area with a value of 1, otherwise 0; the dynamic weight w of the convolutional channel in the masked area is calculated c , specifically expressed as:

[0100]

[0101] where F c (l, v) represents the feature map of the c-th channel, γ1 represents the reference weight, and γ2 represents the weight adjustment coefficient; it should be noted that the value of the reference weight is where NDVI avg represents the average vegetation coverage index of the current scene, which is calculated from the vegetation masked area of the multi-source data fusion matrix in S1; the value of the weight adjustment coefficient is PN total represents the number of pixels, and max(F c ) represents the maximum activation value of the current feature channel;

[0102] Furthermore, in the hydrological cycle feature embedding layer, the hydrological parameters are encoded into periodic feature vectors. In this embodiment, the flow velocity v and water depth d in the hydrological parameters are encoded, specifically expressed as:

[0103]

[0104] where TH represents the hydrological cycle, and t h represents the current timestamp; through the fully connected layer, h vd is mapped to the same dimension as the satellite remote sensing features and then feature-level fusion is performed;

[0105] S204: The data of each layer of the anti-interference environment feature map are fused in the frequency domain space through a cross-modal feature alignment algorithm to generate a feature dataset;

[0106] Specifically, feature alignment is performed in the frequency domain space. By performing two-dimensional fast Fourier transform on the radiometric correction multi-spectral feature layer, the dynamic mask feature layer, and the temporal change feature layer respectively, the amplitude spectrum and phase spectrum of each feature layer are obtained; based on the phase spectrum of the radiometric correction multi-spectral feature layer, the phase alignment of the dynamic mask feature layer and the temporal change feature layer is performed; the aligned phase spectrum is combined with the original amplitude spectrum, and the inverse fast Fourier transform is performed to generate a fused feature map, that is, the feature dataset.

[0107] Specifically, in S3, a lightweight recognition engine is pre-deployed on the edge computing nodes in the target water area to perform regional segmentation and feature dimensionality reduction on the feature dataset, generating a lightweight feature package containing four types of problem identifications: illegal occupation, illegal mining, illegal piling, and illegal construction, that is, four types of lightweight feature packages.

[0108] In some embodiments, the steps of performing regional segmentation and feature dimensionality reduction on the feature dataset transmitted in S2 to generate four types of lightweight feature packages are as follows:

[0109] S301: Based on the geographic grid coding rule of the target water area, the high-resolution image is segmented into block units of a preset size through a dynamic chunking algorithm, and a 10% overlapping area is set at the block boundary to avoid feature truncation;

[0110] It should be noted that the segmentation rule of the block unit of the preset size includes: when the image resolution R ≤ 1m / pixel, the size of the preset block unit is 256×256 pixels; when 0.5m / pixel ≤ R < 1m / pixel, the size of the preset block unit is 512×512 pixels; when R < 0.5m / pixel, the size of the preset block unit is 1024×1024 pixels;

[0111] Furthermore, a 10% overlapping area is set at the block boundary, and the overlapping area pixels are smoothed and fused through a cosine weighting function to eliminate the block edge feature truncation effect;

[0112] S302: Perform sparse representation based on K-SVD dictionary learning on each block unit to generate an optimized sparse basis for discrete cosine transform;

[0113] Specifically, extract 100,000 block samples from the historical four-disorder problem data stored in the system operation database to construct a training set; initialize the discrete cosine transform dictionary, update the dictionary atoms through an iterative optimization algorithm to minimize the sparse representation error, and at the same time constrain the number of non-zero elements of the sparse coefficient not to exceed a preset threshold; output the optimized discrete cosine transform sparse basis; perform sparse coding on the real-time block data, solve the coefficient vector that satisfies the minimum reconstruction error and sparsity constraints, and use the orthogonal matching pursuit algorithm to accelerate the calculation process.

[0114] S303: Construct an adaptive measurement matrix according to the vegetation coverage index and hydrological parameters in the real-time environment perception data;

[0115] It should be noted that an adaptive measurement matrix is dynamically generated according to real-time environmental perception data, specifically including: when the vegetation coverage index is not less than 0.3, the sampling rate is increased from the baseline of 20% to 30%, which is achieved by increasing the number of rows of the adaptive measurement matrix, and a partial orthogonal matrix structure is adopted to ensure sampling stability; when the flow velocity reaches or exceeds 0.5 m / s, the row vectors of the adaptive measurement matrix are orthonormalized to reduce the matrix condition number and enhance the reconstruction quality; according to the spatial distribution ratio of the vegetation and water body areas, the measurement matrices corresponding to different regions are dynamically fused to generate the final environment-adaptive measurement matrix.

[0116] S304: Four types of lightweight feature packets are generated through the compressive sampling formula. The four types of lightweight feature packets include lightweight feature packets for illegal occupation problems, illegal mining problems, illegal piling problems, and illegal construction problems. Among them, each lightweight feature packet is associated with the probability distribution vectors of illegal occupation problems, illegal mining problems, illegal piling problems, and illegal construction problems.

[0117] Specifically, in S4, the steps of the spatio-temporal correlation analysis and multi-dimensional decision-making graph generation include:

[0118] S401: The data of different basins in the historical governance data stored in the system operation database are divided into the source domain and the target domain. Among them, the source domain data contains complete annotations of the four illegal problems, and the target domain data is the lightweight feature packet of the current basin;

[0119] The source domain data of this embodiment is from 3 adjacent basins that have been completed in governance, including historical annotations of the four illegal problems and governance effect scores, with a time span from 2015 to 2024; the target domain data is the four types of lightweight feature packets of the target water area;

[0120] S402: The historical scores of the source domain governance are mapped to the probability space of the target domain through a linear weighting strategy, and the distributions of the source domain and the target domain in the probability space are aligned through the KL divergence loss function;

[0121] In this embodiment, the difference degree between the feature distributions of the source domain and the target domain is calculated, and the probability density curves of the features of the two domains are respectively fitted through kernel density estimation to minimize the information entropy difference between the two curves;

[0122] S403: The probability space of the target domain is passed through a temporal convolutional network to obtain a risk weight matrix;

[0123] In this embodiment, the aligned feature vectors and encoded hydrological parameters are received, and through a temporal convolutional network, a 4×4 risk weight matrix is mapped and generated in the fully connected layer. The elements of the risk weight matrix represent the risk levels of the four illegal problems in the four quarters;

[0124] S404: Weight the probability distribution vector of the four types of violations in the target water area and the risk weight matrix by quarter dimension to generate a multi-dimensional decision-making map, which includes the risk levels, disposal priorities, associated geographical grid codes, and quarterly timestamps corresponding to the problem types of illegal occupation, illegal mining, illegal piling, and illegal construction.

[0125] Specifically, in S5, the dynamic visualization method of the four types of violations includes a flood discharge capacity impact coefficient matrix, a grid-based risk probability, and a two-dimensional gradient color scale. The implementation steps of the dynamic visualization of the four types of violations are as follows:

[0126] S501: Calculate the cross-sectional flow velocity distribution and water level change of the river channel according to the spatial distribution of the illegal piling and illegal construction problems, and generate a flood discharge capacity impact coefficient matrix;

[0127] In this embodiment, the two-dimensional shallow water equation is used to describe the river channel water flow movement. The illegal piling and illegal construction problems are modeled as the cross-sectional blockage rate. The illegal piling area is calculated by the coverage rate of the illegal piling mask in the UAV image, and the illegal construction area is determined according to the proportion of the building projection area; the flood discharge capacity impact coefficient is calculated by the shallow water equation solver based on the finite volume method, and then the flood discharge capacity impact coefficient matrix is obtained. The value range of the elements in the flood discharge capacity impact coefficient matrix is [0, 1], and 1 represents complete blockage;

[0128] S502: Construct a spatio-temporal heat map, simulate the risk diffusion path of the four types of violations within the preset governance cycle, and calculate the grid-based risk probability value;

[0129] In some embodiments, an improved random walk algorithm is used to simulate the diffusion path of the four types of violations, and the random walk path is iteratively calculated 1000 times to statistically calculate the grid-based risk probability value; the risk probability value is mapped to a 256-level gray scale gradient to generate a spatio-temporal heat map; when the detection result of a new four-type violation problem is input, the heat map is updated every 5 minutes;

[0130] S503: Generate a two-dimensional gradient color scale through HSV color space mapping according to the flood discharge capacity impact coefficient and the risk probability value, where red corresponds to the high-risk area;

[0131] In some embodiments, the flood discharge capacity impact dimension is represented by a gradient from blue to purple, and the risk probability dimension is represented by a gradient from green to red. The two-dimensional data is fused through the HSV color space to display the two-dimensional gradient color scale.

[0132] In the second aspect, this embodiment also discloses a research and judgment system for the four types of violations in water conservancy rivers and lakes, referring to Figure 3 , applicable to the method described in the first aspect. The system includes:

[0133] Water area data acquisition module: Obtain real-time images of the target water area through a drone equipped with a multispectral sensor, and synchronously receive satellite remote sensing data, geographic information system vector data, and real-time environmental perception data including light intensity, vegetation coverage index, and hydrological parameters to construct a multi-source heterogeneous dataset of the target water area;

[0134] Dynamic environmental feature extraction module: Extract dynamic environmental features from the multi-source heterogeneous dataset to generate a feature dataset of the target water area;

[0135] Four-disorder feature segmentation module: Pre-deploy a lightweight recognition engine on the edge computing node, perform regional segmentation and feature dimensionality reduction on the feature dataset to generate four types of lightweight feature packets;

[0136] Spatio-temporal analysis module: Perform spatio-temporal correlation analysis on the four types of lightweight feature packets and the historical governance database to generate a multi-dimensional decision-making map of the target water area;

[0137] Visualization and early warning module: Drive a visualization early warning platform based on the multi-dimensional decision-making map of the target water area to realize the dynamic visualization of the four-disorder problems in the target water area.

[0138] In a third aspect, this embodiment also discloses an electronic device. Referring to Figure 4 , the electronic device may include: at least one system central processing unit 401, at least one communication bus 402, at least one system operation database 403, and a user information terminal 404; the system operation database is used to store one or more computer program instructions, wherein the one or more computer program instructions are executed by the system central processing unit to implement the method described in the first aspect.

[0139] It should be noted that the system central processing unit 401 is the core operation and control unit of the entire research and judgment system for the four-disorder problems of water conservancy rivers and lakes; it includes one or more processing cores, and connects various parts within the entire system through various interfaces and lines; by running or executing instructions, programs, code sets, or instruction sets stored in the system operation database, and being able to call the data stored therein, it can execute various functions for the research and judgment of the four-disorder problems of water conservancy rivers and lakes, including real-time monitoring of the river and lake water area shorelines, intelligent identification of illegal and irregular behaviors, analysis and statistics of relevant data, etc.

[0140] Among them, the communication bus 402 is used to realize connection and communication between components.

[0141] Among them, the system operation database 403 is used to store a large amount of data related to the judgment of the four chaos problems in water conservancy rivers and lakes, including the geographical information of rivers and lakes, water level and flow data, and a large amount of historical operation data, including monitoring data and processing result data in different time periods, etc. When the system central processor executes various functions, it will frequently call these data from the system operation database for operations such as data comparison, model training, and decision-making, so as to achieve precise control and efficient management of the judgment of the four chaos problems in water conservancy rivers and lakes.

[0142] Among them, the user information terminal 404 is an information output device for receiving the judgment system of the four chaos problems in water conservancy rivers and lakes. It provides an interface for users to interact with the system by connecting external devices such as a display screen and a camera through a standard wired interface or wireless interface.

[0143] Secondly, in the accompanying drawings of the disclosed embodiments of the present invention, only the structures related to the disclosed embodiments are involved. For other structures, reference can be made to the general design. Without conflict, the same embodiment and different embodiments of the present invention can be combined with each other.

[0144] Finally, the above are only the preferred embodiments of the present invention and are not used to limit the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for judging the four chaos problems in water conservancy rivers and lakes, including: S1: Obtain real-time images of the target water area through a drone equipped with a multispectral sensor, and simultaneously receive satellite remote sensing data, geographic information system vector data, and real-time environmental perception data including light intensity, vegetation coverage index, and hydrological parameters, and construct a multi-source heterogeneous data set of the target water area; S2: Extract dynamic environmental features from the multi-source heterogeneous data set to generate a feature data set of the target water area; S3: Pre-deploy a lightweight recognition engine on the edge computing node, perform regional segmentation and feature dimensionality reduction on the feature data set, and generate four types of lightweight feature packets; S4: Perform spatio-temporal correlation analysis on the four types of lightweight feature packets and the historical governance database to generate a multi-dimensional decision-making map of the target water area; S5: Drive a visualization warning platform based on the multi-dimensional decision-making map of the target water area to realize the dynamic visualization of the four chaos problems in the target water area.

2. The research and judgment method for the four chaos problems of water conservancy rivers and lakes according to claim 1 is characterized in that: In S1, constructing the multi-source heterogeneous data set of the target water area specifically includes: Control the time deviation of multi-source data acquisition within a preset threshold range through the time synchronization protocol of the drone flight control software and the satellite data receiving end; Perform coordinate transformation on multi-source data based on the WGS84 geographic coordinate system, and when the grid scale difference between GIS vector data and satellite remote sensing data exceeds the preset threshold, call the spatial resampling function to perform interpolation calculation. In the spatial resampling function, dynamically select the interpolation algorithm type according to the spectral characteristics of real-time environmental perception data; Generate a multi-source heterogeneous data set represented by a multi-source data fusion matrix.

3. The research and judgment method for the four chaos problems of water conservancy rivers and lakes according to claim 1, characterized in that: In S2, extracting dynamic environmental features from the multi-source heterogeneous data set specifically includes: Construct an adaptive spatial registration weight matrix based on the light intensity and vegetation coverage index of real-time environmental perception data to register multi-source data; Use a Gaussian-Laplacian pyramid fusion model to perform multi-scale denoising on the registered multi-source data; Generate an anti-interference environmental feature map through a dynamic environmental feature extraction model. The anti-interference environmental feature map includes a radiation correction multispectral feature layer, a dynamic mask feature layer, and a temporal change feature layer; Fuse the data of each layer of the anti-interference environmental feature map to generate a feature data set.

4. The research and judgment method for the four chaos problems of water conservancy rivers and lakes according to claim 3, characterized in that: In S2, the dynamic environmental feature extraction model is a composite neural network architecture including a light intensity compensation layer, a vegetation occlusion attention mechanism layer, and a hydrological cycle feature embedding layer, where: The light intensity compensation layer calculates the radiation correction coefficient through the light intensity and performs adaptive brightness compensation on the multispectral image; The vegetation occlusion attention mechanism layer generates an occlusion area mask according to the vegetation coverage index and dynamically assigns attention weights to the feature extraction channels; The hydrological cycle feature embedding layer encodes hydrological parameters into periodic feature vectors and performs feature-level fusion with satellite remote sensing data.

5. The research and judgment method for the four chaos problems of water conservancy rivers and lakes according to claim 4, characterized in that: S2. In the light intensity compensation layer, based on the band reflectance R of the multispectral image b , calculate the radiation correction coefficient γ b , which is specifically expressed as: Among them, L ref represents the calibrated light intensity, L represents the current light intensity, β represents the environmental sensitivity coefficient, and R avg represents the average reflectance of the current scene; In the vegetation occlusion attention mechanism layer, a binary mask M(l, v) is generated according to the vegetation coverage index; the dynamic weight w of the convolutional channels in the masked area is calculated c , which is specifically expressed as: Among them, F c (l, v) represents the feature map of the c-th channel, γ1 represents the reference weight, and γ2 represents the weight adjustment coefficient.

6. The judgment method for the four chaos problems of water conservancy rivers and lakes according to claim 1, characterized in that: In S3, performing regional segmentation and feature dimensionality reduction on the feature data set specifically includes: Based on the geographic grid coding rule of the target water area, divide the high-resolution image into block units of a preset size through a dynamic block algorithm; Perform sparse representation on each block unit to generate an optimized sparse basis for discrete cosine transform; Construct an adaptive measurement matrix based on real-time environmental perception data; Generate four types of lightweight feature packets through compressive sampling.

7. A method for judging the four chaos problems of water conservancy rivers and lakes according to claim 1, characterized in that: In step S4, the multi-dimensional decision-making map includes the risk levels, disposal priorities, associated geographic grid codes, and quarterly timestamps corresponding to the problem types of illegal occupation, illegal mining, illegal piling, and illegal construction.

8. The research and judgment method for the four chaos problems of water conservancy rivers and lakes according to claim 1, characterized in that: In step S5, the dynamic visualization method for the four illegal problems includes the flood discharge capacity influence coefficient matrix, grid-based risk probability, and two-dimensional gradient color scale.

9. A judgment system for the four chaos problems of water conservancy rivers and lakes, characterized in that, A method for judging the four illegal problems in water conservancy rivers and lakes applicable to any one of claims 1-8, comprising: Water area data acquisition module: Obtain real-time images of the target water area through a drone equipped with a multi-spectral sensor, and simultaneously receive satellite remote sensing data, geographic information system vector data, and real-time environmental perception data including light intensity, vegetation cover index, and hydrological parameters, and construct a multi-source heterogeneous data set of the target water area; Dynamic environmental feature extraction module: Extract dynamic environmental features from the multi-source heterogeneous data set to generate a feature data set of the target water area; Four illegal feature segmentation module: Pre-deploy a lightweight recognition engine at the edge computing node, perform region segmentation and feature dimensionality reduction on the feature data set, and generate four types of lightweight feature packets; Spatio-temporal analysis module: Perform spatio-temporal correlation analysis on the four types of lightweight feature packets and the historical governance database to generate a multi-dimensional decision-making map of the target water area; Visualization warning module: Drive a visualization warning platform based on the multi-dimensional decision-making map of the target water area to realize the dynamic visualization of the four illegal problems in the target water area.

10. An electronic device, comprising a system central processing unit, a user information terminal, and a system operation database, characterized in that, The system operation database is used to store one or more computer program instructions, wherein the one or more computer program instructions are executed by the system central processing unit to implement the method according to any one of claims 1 to 8.

Images