Pollutant identification and early warning method and system for river patrol pollution source

Abstract

The invention relates to the technical field of river pollution identification, in particular to a pollutant identification and early warning method and system for a river patrol pollution source, and the method comprises the steps: constructing a three-dimensional monitoring network system, so as to build a three-dimensional data collection architecture; establishing a space-time reference unified framework to realize synchronous time service of different monitoring nodes; deploying an edge computing node, and implementing data preprocessing at an equipment end of the monitoring node; constructing a pollutant feature library which comprises spectral features, water quality parameter correlation features and visual morphological features of river pollutants, and establishing a water quality parameter correlation model of organic pollutants and inorganic pollutants, which comprises a plurality of feature dimensions; and a self-adaptive threshold early warning mechanism is established, a space-time composite early warning model is constructed, and accurate positioning and hazard degree grading early warning of pollution events are realized. According to the invention, through the hierarchical fusion model of technology fusion, the precision and speed of identifying the river pollutants are improved.

Description

Technical Field

[0001] The present invention relates to the field of river pollution identification, and particularly to a method and system for identifying and warning pollutants of a river patrol pollution source. Background Art

[0002] With the rapid development of urbanization and industrialization, the problem of river pollution has become increasingly severe. Due to its limitations, traditional monitoring technologies are difficult to meet the needs of precise treatment. At present, river pollution monitoring mainly relies on single means, with narrow data dimensions and insufficient timeliness, making it difficult to comprehensively capture the dynamic distribution characteristics of pollutants. For example, the coverage of fixed-point sensors is limited, and there are periodic and blind area problems in drone inspections, resulting in fragmented pollution identification. In addition, the acquisition and fusion of multi-source heterogeneous data face challenges: due to inconsistent spatio-temporal benchmarks of different devices, data misalignment is likely to occur, and traditional methods lack a dynamic calibration mechanism, making it difficult to eliminate the drift error of mobile devices, seriously affecting the accuracy of pollution source tracing and diffusion analysis.

[0003] At the data processing level, traditional architectures mostly adopt a centralized computing mode, which prolongs the transmission time of a large amount of monitoring data and is significantly interfered by environmental noise, resulting in poor real-time performance and a high misjudgment rate. Existing pollutant identification models lack robustness in complex scenarios, are difficult to synchronously process multi-dimensional information, and model iteration relies on offline training, unable to dynamically adapt to changes in the river environment. At the same time, pollution diffusion prediction is mostly based on a simplified two-dimensional model and is not deeply coupled with the three-dimensional hydrodynamic equation, with limited prediction accuracy and difficult to support precise emergency response.

[0004] Therefore, how to improve the accuracy of river pollutant identification is an urgent technical problem in this field. Summary of the Invention

[0005] The present invention improves the accuracy of river pollutant identification through a hierarchical fusion model that integrates multiple technologies.

[0006] The technical solution proposed by the present invention is: a method for identifying and warning pollutants of a river patrol pollution source, the method comprising: Combining aerial, underwater and shore-based image acquisition devices to establish a three-dimensional data acquisition architecture and construct a three-dimensional stereo monitoring network system; Obtaining the spatial coordinates of monitoring nodes according to the three-dimensional stereo monitoring network system, unifying the spatial coordinates of each monitoring node through the GNSS differential positioning system, using the Kalman filtering technology to perform real-time calibration of the coordinates of the positioned monitoring nodes, and combining the spatial coordinates and timestamp alignment algorithm of the monitoring nodes to synchronize the spatio-temporal coordinates of multi-source heterogeneous data of different monitoring nodes, and establishing a unified spatio-temporal benchmark framework; According to the obtained spatio-temporal coordinates of the monitoring nodes, performing data preprocessing on the device end of the monitoring nodes to implement the deployment of edge computing nodes; Based on the spectral characteristics, water quality parameter correlation characteristics, and visual morphological characteristics of river pollutants collected by monitoring nodes, a pollutant feature library and a water quality parameter correlation model are constructed to obtain river flow, pollutant diffusion rate, and historical pollution data; An adaptive threshold warning mechanism is established. The dynamic fuzzy comprehensive evaluation method is used to evaluate data such as river flow, pollutant diffusion rate, and historical pollution events, and a spatio-temporal composite warning model is constructed. The pollutant is located and the warning of the harm degree is classified through the spatio-temporal composite warning model.

[0007] Preferably, when the airborne image acquisition device performs image processing, a deep learning algorithm for multi-scale feature fusion is used, and the specific content is as follows: An improved U-Net++ network architecture is constructed. ResNet50 is introduced as the backbone network in the encoder part, and an attention mechanism module is added to the decoder part to establish a cross-layer feature fusion channel; a bidirectional reflectance distribution function model and an image enhancement module based on a physical model are set; a multi-task learning framework is established, and a weighted cross-entropy loss function is used to balance the class imbalance problem, so that the UAV image processing device can simultaneously complete pollutant identification, concentration inversion, and diffusion trend prediction; an online incremental learning mechanism is established, and the model collaborative evolution of each edge node is realized through the federated learning framework, and a confidence threshold is set to automatically screen high-quality samples to update the model parameters of each edge node; a virtual-real combined verification system is constructed, and a three-dimensional simulation environment of the river is created using digital twin technology, and typical pollution scenario data is injected for model verification and optimization.

[0008] Preferably, when the airborne image acquisition device performs image processing, an adaptive topology optimization method is used, and the specific process of the adaptive topology optimization method is as follows: By developing an energy-aware routing algorithm, an optimization model of node residual energy, link quality, and image transmission requirements is established. Combining the optimization model and the hybrid networking protocol, a dual-mode transmission integrating LoRaWAN low-power wide-area communication and 5G low-latency communication is established; an abnormal node self-repair mechanism is established. When a node failure is detected, combining blockchain and relay node selection algorithms, an improved particle swarm optimization method is used to quickly reconstruct the optimal communication path; mobile emergency monitoring nodes are set, and self-propelled surface robots and underwater ROVs are configured.

[0009] Preferably, the specific process of the spatio-temporal reference unified framework is as follows: Combined with satellite remote sensing data, UAV oblique photography data and lidar point cloud data, a three-dimensional river channel model is established. Through the dynamic projection matching algorithm, the spatial registration of mobile monitoring data and static maps is achieved, and the improved RANSAC algorithm is used to eliminate errors in the spatial registration process. A spatio-temporal coding system is designed to establish a two-way indexing relationship between monitoring data and river channel spatial units. Each spatial unit uses Geohash coding, and the time dimension uses the UNIX timestamp and leap second compensation mechanism. A pollutant diffusion simulation model is constructed by coupling the two-dimensional hydrodynamic model and the mass transfer equation: ; wherein, is the pollutant concentration, is the flow velocity vector, is the diffusion coefficient tensor, is the source-sink term; A 3D display platform is developed to support pollutant diffusion simulation, historical trajectory backtracking and dynamic rendering of warning areas to achieve multi-dimensional data visualization.

[0010] Preferably, the adaptive threshold warning mechanism adopts an improved fuzzy comprehensive evaluation algorithm, and the specific content of the improved fuzzy comprehensive evaluation algorithm is as follows: Construct a multi-layer evaluation index system, including a water quality parameter layer, a morphological feature layer and an environmental impact layer; Establish an adaptive weight allocation mechanism according to the multi-layer evaluation index system, and based on the historical pollution event database and real-time river channel flow data, use the entropy weight method to dynamically calculate the index weights; Combine the multi-layer evaluation index system and the adaptive weight allocation mechanism to construct a non-linear membership function library, forming trapezoidal, Gaussian and S-shaped functions for different pollutant types to generate pollution risk levels; Develop a multi-level warning trigger strategy. If the comprehensive score is greater than the highest set threshold, a red warning is launched. If the comprehensive score is less than the highest set threshold and greater than the lowest set threshold, an orange warning is launched and the emergency response system is linked; Implement a model online optimization mechanism, and use the reinforcement learning framework to dynamically adjust the fuzzy rule base according to the warning accuracy.

[0011] Preferably, the spatio-temporal composite warning model includes the criteria for pollutant treatment, and the specific criteria are as follows: An autonomous navigation surface robot platform is equipped with an obstacle perception module that integrates radar and multi-camera vision; A modular pollutant treatment unit is equipped with an automatic spreading mechanism for oil-absorbing felt, an ultrasonic demulsification device and an electrochemical degradation reactor. The treatment efficiency of the autonomous navigation surface robot platform meets the following conditions: ; wherein: is the influent pollutant concentration, is the concentration after treatment; It also includes a distributed energy system and an intelligent scheduling model for emergency supplies. The distributed energy system uses hybrid power supply. The intelligent scheduling model for emergency supplies plans the optimal interception path based on the improved Dijkstra algorithm to achieve real-time calculation of the movement trajectory of the centroid of the pollution mass.

[0012] Preferably, the deployment of the edge computing nodes is realized through a blockchain system, and the specific content of the blockchain system is as follows: Double-chain heterogeneous architecture, which stores real-time monitoring data through the chain, and the transaction chain records early warning events and disposal logs; establish a lightweight consensus mechanism, adopt the improved DPoS algorithm, elect multiple verification nodes and introduce VRF random numbers; combine the double-chain heterogeneous architecture and the lightweight consensus mechanism to build a contract template library, and complete data verification contracts, early warning trigger contracts and traceability query contracts through the contract template library; set up a cross-chain oracle system, dock with the environmental protection department database through the TLS-N two-way authentication protocol, and realize the trusted interaction of data on-chain and off-chain in the blockchain system.

[0013] Preferably, the contract template library includes the following content: Set up a multi-modal data lake to integrate structured monitoring data, unstructured image videos and semi-structured report documents; combine a knowledge extraction engine and a model based on BiLSTM-CRF to extract entity relationship triples from historical event reports; establish an incremental knowledge fusion algorithm and adopt the D-S evidence theory to synthesize different credibility values; combine the incremental knowledge fusion algorithm and the knowledge graph dynamic update strategy to establish a case reasoning mechanism, and quickly retrieve similar historical scenarios through a k-d tree.

[0014] Preferably, the specific content of the spatio-temporal composite early warning model is as follows: Use the finite volume method to solve the three-dimensional hydrodynamic equation, establish a digital twin, and set up a multi-physical field coupling simulation engine in the digital twin in combination with a hydrodynamic module, a pollutant transport module and an ecological impact assessment module; calculate the dynamic error between the measured data and the simulation result through the multi-physical field coupling simulation engine and the virtual-real data comparison algorithm, substitute the dynamic error into the model of the virtual-real combination verification system for parameter self-correction, and when the parameter self-correction value continuously exceeds the set threshold, start the Bayesian inversion algorithm to optimize the diffusion coefficient parameter, and after self-correction, substitute it into the XR training platform.

[0015] The present invention also provides a system for patrolling river pollution, and this system is used to execute a method for identifying and warning pollutants of a river pollution source.

[0016] The beneficial effects of the present invention: 1. The present invention constructs a three-dimensional monitoring network system, integrating the collaborative monitoring capabilities of drones, underwater sensor arrays, and shore-based equipment, achieving all-round perception of river pollution. The drone is equipped with hyperspectral imaging equipment, which can cruise over a large area to capture the morphological characteristics of floating pollutants on the water surface; the underwater multi-parameter sensor array is arranged in a grid pattern to collect water quality parameters at different depths in real time; the shore-based intelligent camera group realizes all-weather monitoring of key points through infrared night vision and pan-tilt control. This acquisition architecture of multi-source heterogeneous data breaks through the limitations of traditional single monitoring modes and can synchronously obtain multi-dimensional information such as the spectral characteristics, water quality parameters, and visual morphology of pollutants. Through the unified spatio-temporal reference framework, using GNSS differential positioning and timestamp alignment technology, it ensures that the spatial coordinates and time reference of all monitoring nodes are highly consistent. Combining the Kalman filtering algorithm to dynamically calibrate the drift error of mobile devices effectively eliminates the spatio-temporal misalignment problem during multi-source data fusion. This three-dimensional monitoring system not only significantly expands the coverage range but also can quickly capture the diffusion trajectory of pollution sources in case of sudden pollution events, providing comprehensive and real-time data support for subsequent accurate identification and early warning.

[0017] 2. The present invention adopts a hierarchical processing architecture, combined with edge computing and intelligent algorithm optimization, significantly improving the efficiency of data processing and the dynamic adaptability of the model. By deploying edge computing gateways near the monitoring nodes, preprocessing such as radiometric correction, geometric correction, and atmospheric correction is performed on the hyperspectral data, greatly reducing the amount of original data and eliminating environmental interference. At the same time, the improved deep learning model (such as the U-Net++ network introduced with an attention mechanism) synchronously completes pollutant identification, concentration inversion, and diffusion prediction through a multi-task learning framework, and combines the federated learning mechanism to achieve the collaborative evolution of the model among edge nodes, ensuring the robustness of the algorithm in complex scenarios. Ensure that the system maintains high precision and high reliability during long-term operation.

[0018] 3. The present invention constructs a multi-layer dynamic evaluation system and an adaptive threshold early warning mechanism, achieving precise classification and rapid response to pollution events. Based on a three-level index system of water quality parameters, morphological characteristics, and environmental impacts, the entropy weight method is used to dynamically allocate weights, and combined with a non-linear membership function to generate pollution risk levels. When the comprehensive score exceeds the threshold, the system automatically triggers multi-level early warnings and links the emergency response mechanism. For example, the autonomous navigation surface robot plans the optimal path through the fusion technology of radar and multi-view vision, and is equipped with a modular processing unit to achieve efficient interception and degradation of pollutants; the emergency material dispatching model dynamically plans the optimal interception path based on the improved Dijkstra algorithm, and combines blockchain technology to ensure the traceability of the disposal process, significantly improving the accuracy of pollutant treatment in this solution. Description of the Drawings

[0019] Figure 1Flow chart of a method and system for identifying and warning pollutants from river pollution sources according to the present invention; Figure 2 Flow chart of the pollutant identification process of a method and system for identifying and warning pollutants from river pollution sources according to the present invention. Detailed implementation manners

[0020] The following description is used to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments in the following description are only examples, and those skilled in the art can think of other obvious variations. The basic principles defined in the following description can be applied to other implementation manners, variant schemes, improvement schemes, equivalent schemes, and other technical schemes that do not depart from the spirit and scope of the present invention.

[0021] It can be understood that the term "one" should be understood as "at least one" or "one or more". That is, in one embodiment, the number of an element can be one, while in other embodiments, the number of the element can be multiple. The term "one" cannot be understood as a limitation on the number.

[0022] Such as Figure 1 And Figure 2 As shown, in this solution, multi-dimensional perception of pollutants is achieved through a three-dimensional monitoring network, the device heterogeneity error is eliminated by combining a spatio-temporal reference unified framework, edge computing nodes are deployed to complete real-time data processing, accurate identification of pollution sources is achieved based on a pollutant feature library, and finally a hierarchical response strategy is output through an adaptive threshold warning mechanism.

[0023] Construct a three-dimensional monitoring network system, including UAV image processing devices, underwater multi-parameter sensor arrays, and shore-based intelligent camera groups, to establish a three-dimensional data acquisition architecture; Establish a spatio-temporal reference unified framework. The spatial coordinates of each monitoring node are unified through a GNSS differential positioning system, the time synchronization of multi-source heterogeneous data of different monitoring nodes is achieved through a timestamp alignment algorithm, and then the Kalman filtering technology is used to perform real-time calibration on each monitoring node to achieve synchronous timekeeping of different monitoring nodes; Deploy edge computing nodes to perform data preprocessing at the device end of the monitoring node. The data preprocessing includes radiometric correction, geometric correction, and atmospheric correction of hyperspectral data to eliminate environmental interference; Construct a pollutant feature library, including spectral features, water quality parameter correlation features, and visual morphology features of river pollutants, and establish a water quality parameter correlation model of organic pollutants and inorganic pollutants including multiple feature dimensions; Establish an adaptive threshold warning mechanism, adopt a dynamic fuzzy comprehensive evaluation method, combine the river flow, pollutant diffusion rate, and historical pollution event database to construct a spatio-temporal composite warning model, and achieve accurate positioning of pollution events and hierarchical warning of the harm degree.

[0024] Construct a three-dimensional monitoring network system consisting of an unmanned aerial vehicle (UAV) image processing device, an underwater multi-parameter sensor array, and a shore-based intelligent camera. In specific implementation, the UAV is equipped with a high-resolution multispectral imager, which is responsible for conducting large-scale cruise monitoring of floating pollutants on the river surface and coastal sewage outlets (the flight altitude is controlled at 30 - 50 meters, and the coverage radius of a single flight is 5 kilometers). The underwater sensor array consists of pH sensors, dissolved oxygen sensors, conductivity sensors, etc. distributed on the riverbed, and is arranged in a grid pattern at 100-meter intervals to collect water quality parameters at different water depths (0.5m, 2m, 5m) in real time. The shore-based intelligent camera group sets up 360-degree pan-tilt cameras at key points such as bridges and sluice gates, equipped with infrared night vision functions to achieve all-weather continuous monitoring.

[0025] The three-dimensional monitoring system breaks through the limitations of traditional single monitoring modes: The UAV provides a macroscopic perspective and can quickly detect oil film pollution on the water surface (such as the rainbow-colored reflective area formed by an oil spill); the underwater sensor captures a sudden drop in dissolved oxygen (when the dissolved oxygen suddenly drops from 8 mg / L to 3 mg / L, it indicates the outbreak of organic pollution); the shore-based equipment continuously tracks fixed-point pollution sources (such as abnormal drainage at a factory sewage outlet at night). After the data of the three are fused, the system can establish a three-dimensional heat map of pollution events. For example, when the UAV detects abnormal foam on the water surface of a certain river section, and synchronously the COD value detected by the underwater sensor in this area soars from 20 mg / L to 80 mg / L, and the shore-based equipment captures a suspicious sewage vehicle at night, the pollution source can be accurately located.

[0026] Compared with the traditional manual sampling and detection method, this three-dimensional system shortens the pollutant identification response time from 24 hours to 15 minutes, and the monitoring range coverage rate is increased by 80%. Especially, the capture ability for sudden pollution events has been significantly improved. For example, in a leakage accident at a chemical plant, the system issued an early warning within 9 minutes after the accident occurred, while the traditional method had to wait for the water quality test report the next day to be verified by actual measurement.

[0027] Establish a unified framework for spatio-temporal reference: Through the GNSS differential positioning system, a centimeter-level positioning module (such as a Trimble R12 receiver) is configured for each monitoring node (UAV, buoy sensor, shore-based equipment) to establish a unified spatial coordinate system. In terms of time synchronization, the PTP precise time protocol is adopted, and through the reference clock server (StratusztC Edge) deployed in the monitoring area, it is ensured that the time error of all devices is less than 1 ms. Regarding the position drift problem of the water surface buoy sensor, the extended Kalman filter algorithm is used for dynamic calibration, and its state equation is expressed as: ; where is the state vector containing position and velocity, is the state transition matrix, is the control input matrix, is the control vector, is the process noise. The unified spatio-temporal reference framework application solves the core problem of multi-source data fusion. For example, during monitoring, the UAV captured a water pollution zone at 09:00:05.235, and at the same time, the underwater sensor recorded an abnormal COD value at 09:00:05.850. After timestamp alignment and spatial coordinate transformation, the system accurately determined that the two belonged to the same pollution event, avoiding misjudgment as two independent events.

[0028] Deploy computing nodes: Deploy edge computing gateways (using NVIDIA Jetson AGX Xavier modules) within a range of 500 meters from the monitoring device to implement three-level data processing: First, perform radiometric correction (using the MODTRAN model to eliminate the influence of atmospheric scattering), geometric correction (achieving image registration through SIFT feature matching), and atmospheric correction (inverting the surface reflectance using the 6S model) on the UAV hyperspectral data. For underwater sensor data, use sliding window filtering (window length 30 seconds) to eliminate turbulence interference, and use the Tukey outlier detection algorithm to eliminate faulty data in real time (such as a certain sensor having a continuous abnormal pH value due to biological attachment). Edge computing reduces the data processing delay to the 200ms level, speeding up the processing by 10 times compared to traditional cloud processing. Taking the preprocessing of hyperspectral data as an example, the 50GB of raw data collected by a single UAV flight is reduced to 1.2GB of effective feature data after being processed by the edge node, reducing the transmission bandwidth requirement by 95%.

[0029] Construct a pollutant feature library: Construct a pollutant feature library containing 12 major categories and 86 sub-categories: The spectral feature library collects the reflection curves of typical pollutants through an ASD FieldSpec4 ground spectrometer (for example, oily wastewater has a characteristic absorption valley in the 450 - 600nm band); the water quality parameter correlation model establishes the mapping relationship between indicators such as COD and ammonia nitrogen and pollutant types (when COD > 60mg / L and ammonia nitrogen > 5mg / L, it is determined to be domestic sewage pollution); the visual morphology library includes more than 2000 pollution sample images (such as the "green carpet" texture feature formed by the aggregation of algae). The feature library realizes multi-dimensional evidence chain verification. For example, during a certain monitoring, the UAV image showed a brown band pollution on the water surface (feature code F003). Synchronously, the dissolved oxygen in this area dropped to 2.8mg / L and the conductivity rose to 1500μS / cm. After the system compared the feature library, it accurately determined that it was a leakage of black liquor from a paper mill, rather than natural sediment suspension. During a heavy metal pollution event in 2024, the system completed the identification of the pollution type within 2 minutes through the joint analysis of spectral features (the characteristic absorption peak of copper ions at 810nm) and water quality parameters (the pH value dropped suddenly to 3.5).

[0030] The adaptive threshold warning mechanism includes an improved fuzzy comprehensive evaluation algorithm. The improved fuzzy comprehensive evaluation algorithm includes: constructing a multi-layer evaluation index system, including a water quality parameter layer, a morphological feature layer, and an environmental impact layer; establishing an adaptive weight distribution mechanism based on the multi-layer evaluation index system, and dynamically calculating the index weights using the entropy weight method based on the historical pollution event database and real-time river channel flow data; constructing a non-linear membership function library in combination with the multi-layer evaluation index system and the adaptive weight distribution mechanism, forming trapezoidal, Gaussian, and S-shaped functions for different pollutant types to generate pollution risk levels; then developing a multi-level warning trigger strategy. If the comprehensive score is greater than the highest set threshold, a red warning is initiated. If the comprehensive score is less than the highest set threshold and greater than the lowest set threshold, an orange warning is initiated, and the emergency response system is linked; implementing an online model optimization mechanism, and dynamically adjusting the fuzzy rule library according to the warning accuracy using the reinforcement learning framework.

[0031] The adaptive threshold warning mechanism: adopts a dynamic fuzzy comprehensive evaluation algorithm to construct a three-level evaluation system including a water quality parameter layer (40% weight), a morphological feature layer (35% weight), and an environmental impact layer (25% weight). The warning threshold is dynamically adjusted according to the river channel flow: the COD warning threshold in the dry season (flow < 50 m³ / s) is set to 30 mg / L, and in the flood season (flow > 200 m³ / s), it is relaxed to 50 mg / L. The formula after expanding the three-dimensional hydrodynamic-material transport coupling model is: ; where , , are the velocity components, is the diffusion coefficient tensor, is the source-sink term. The system calculates the pollution mass movement trajectory in real time, and triggers a red warning when the predicted impact range exceeds 1 square kilometer in 24 hours. The dynamic threshold mechanism significantly reduces the false alarm rate. This solution is applied to the water flow basins near and within the city. The river channel water depth is uniform or changes slowly, so the water depth can be assumed to be constant and can be indirectly reflected by the flow rate. This pollution diffusion model couples the two-dimensional hydrodynamic equation and is applied to the emergency scenario of sudden pollution. It gives priority to the planar trajectory of pollution diffusion. Therefore, under the premise of such uniform water depth, without considering the water depth, a simplified two-dimensional model can be used to quickly solve the problem without complex three-dimensional calculations, meeting the timeliness requirements. Since the water depth is set to be constant, then , and the simplified pollution diffusion model coupling the two-dimensional hydrodynamic equation according to the formula after expanding the three-dimensional hydrodynamic-material transport coupling model is 。During a certain monitoring, the instantaneous COD value of a certain river section reached 45 mg / L. However, based on the then flow rate of 200 m³ / s (threshold 50 mg / L) and the prediction results of the diffusion model, the system determined it as a non-accident discharge that could be self-purified in the short term, avoiding false triggering of the emergency response. Another example is in the leakage incident of a chemical plant. When the COD value reached 55 mg / L, the system immediately initiated the emergency procedure, responding 30 minutes earlier than the traditional fixed threshold method. Compared with the traditional monitoring method, the accuracy of pollutant identification in this solution was increased by 42%, the early warning response speed was increased by 53%, the pollution source location error was reduced by 78%, and the comprehensive efficiency of the system was increased by 45%.

[0032] After dispatching drones to collect images of river pollutants, the drone image processing link is carried out, and a deep network architecture based on U-Net++ is constructed. The encoder part uses ResNet50 as the backbone network to solve the problem of gradient disappearance in deep networks through residual connections. For example, when analyzing the image of oil film pollution on the water surface, the deep convolutional layers of ResNet50 (such as the 4th and 5th layers) can effectively extract the fine texture features of the oil film edge (such as the reflectivity change caused by the thickness difference). The decoder part introduces a cross-layer attention mechanism module (Cross-Layer Attention Module), which dynamically fuses multi-scale features by calculating the correlation weights between the shallow feature maps (such as edge information) and the deep feature maps (such as semantic information). When dealing with the interference of water surface reflection, this mechanism can suppress the noise in the highlighted area and highlight the feature response of the real pollution area. The improved U-Net++ architecture forms a synergy with the physical enhancement module. For example, in a certain oil spill accident, the original image taken by the drone had serious water surface reflection due to direct sunlight. The improved network increased the feature weight of the oil film area to 0.85 (the weight of the reflective area decreased to 0.12) through the attention mechanism, and corrected the specular reflection component in cooperation with the bidirectional reflectance distribution function (BRDF) model, finally increasing the oil film recognition accuracy from 72% of the traditional U-Net to 94%.

[0033] The multi-task learning framework can split the UAV image processing into a three-task parallel processing framework: the main branch conducts pollutant semantic segmentation (such as identifying categories like algae and oil films), the secondary branch inversely calculates the pollutant concentration through a regression network (the prediction error of COD value < 5mg / L), and the auxiliary branch predicts the diffusion range in the next 30 minutes based on LSTM. A dynamic weight adjustment strategy is adopted. At the initial stage of training, a higher weight is given to the segmentation task (the initial weight is 0.6), and as the model converges, the weight of the concentration prediction task is gradually increased (the final weight is 0.4). For the algae outbreak scenario, the system synchronously outputs the segmentation map of the algae coverage area, the chlorophyll a concentration heat map (accuracy ±0.2μg / L), and the diffusion direction vector map (angle error < 5°). The multi-task framework and federated learning form a closed-loop optimization. For example, in the cyanobacteria monitoring of a certain basin, the new aggregation form of algae (filamentous distribution) discovered by edge node A is uploaded to the cloud model library through federated learning. When edge node B encounters the same features, the confidence of the model for the segmentation task is increased from 0.65 to 0.92, and the concentration inversion error is reduced by 40%.

[0034] Federated learning and incremental learning mechanism: Build a distributed federated learning framework, and each edge node (such as UAV base stations deployed upstream and downstream) regularly uploads encrypted model parameters to the cloud aggregation server. Adopt a dynamic confidence threshold screening mechanism: When the segmentation confidence of a certain sample > 0.9 and the concentration prediction error < 8%, it is determined as a high-quality sample for incremental update of the local model. In a benzene series substance leakage incident in a certain chemical plant, the pollutant spectral characteristics captured by the upstream node (the characteristic absorption peak of the benzene ring at 250nm) are shared with the downstream node through federated learning, enabling the downstream system to identify the same type of pollutant 20 minutes in advance.

[0035] Establish a three-dimensional simulation environment for the river channel through a digital twin verification system, integrating a hydrodynamic model (solving the Navier-Stokes equation using the finite volume method), a pollutant transport model (solving the convection-diffusion equation), and an ecological impact assessment module. For example, in a certain emergency drill, the digital twin system simulates a 10-ton diesel leakage scenario: The hydrodynamic module calculates the flow velocity distribution (2.1m / s in the main river channel, 0.3m / s on the shore); the pollutant transport module predicts that the pollution front will reach the downstream water intake after 30 minutes; the ecological module assesses that the fish mortality rate will reach 35%; based on the simulation results, the system automatically generates an optimal disposal plan: Set up an oil boom at 3 kilometers and simultaneously start the oil absorbent robot, thereby improving the efficiency of treating river channel pollutants.

[0036] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions, and operations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in the flowchart or block diagram may represent a module, a segment of a program, or a part of code that contains one or more executable instructions for implementing a specified logical function. It should also be noted that, in some alternative implementations, the functions marked in the blocks may occur in a different order than that marked in the accompanying drawings. For example, two consecutive blocks shown may actually be executed substantially in parallel, and they may sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, and the combinations of blocks in the block diagram and / or flowchart, can be implemented by a dedicated hardware-based system that performs the specified functions or operations, or can be implemented by a combination of dedicated hardware and computer instructions.

[0037] Energy-Aware Routing and Dual-Mode Communication: In the river channel monitoring network, an energy-aware routing algorithm is set up. By real-time monitoring of three core parameters, namely the remaining battery power of nodes, link quality, and image transmission requirements, a dynamic routing optimization model is constructed. When the remaining energy of a certain node is lower than the threshold, the algorithm automatically removes it from the data forwarding path and selects a detour path to ensure network connectivity. At the same time, the hybrid networking protocol integrates LoRaWAN (transmission distance 10 km, rate 50 kbps) and 5G (rate 1 Gbps, latency < 10 ms) dual-mode communication: Conventional water quality data (such as pH value, dissolved oxygen) is transmitted with low power consumption through LoRaWAN, while high-definition images of sudden pollution events are switched to the 5G channel. For example, in a monitoring case of a cross-river bridge, the traditional single communication mode caused the transmission of high-definition images to take up to 15 minutes. Through dual-mode dynamic switching, after the system detects the characteristics of oil pollution images, it immediately enables the 5G channel to transmit a 2GB high-definition image back to the command center within 8 seconds, while keeping the LoRaWAN channel continuously transmitting water quality parameters. The energy-aware algorithm extends the node battery life from 3 months to 8 months, and the network survival period is increased by 167%.

[0038] Abnormal Node Self-Repair Mechanism: To achieve dynamic reconstruction and intelligent optimization, when a monitoring device node fails due to flood impact or equipment failure, the pollutant identification and early warning method activates the collaborative repair mechanism of blockchain verification and particle swarm optimization (PSO). First, the distributed ledger of blockchain records the historical data of nodes (such as the stability score of a certain buoy sensor in the past 30 days) to quickly identify the location of the faulty node. Subsequently, the improved PSO algorithm takes the maximization of network connectivity as the objective function and calculates the optimal relay path within 0.5 seconds. For example, during a certain rainstorm, 3 underwater sensors went offline, and the system automatically selected 2 buoys with the best link quality among the remaining nodes as relays, and the network throughput after reconstruction recovered to 92% of that before the failure.

[0039] The mobile emergency monitoring network can achieve rapid response and autonomous networking. By deploying self-propelled surface robots (with adjustable speed of 0 - 5 m / s) and underwater ROVs (with a maximum diving depth of 50 m), and equipping them with multi-parameter sensors and mechanical grasping devices. When a sudden pollution event occurs, the emergency management center automatically dispatches 3 surface robots and 2 ROVs to form a mobile monitoring network. During execution, the surface robots sample along the edge of the pollution zone in a zigzag path (taking a water sample every 200 meters), and the ROV dives to a depth of 5 meters underwater to monitor the sedimentation of pollutants. All mobile nodes communicate through self-organized MESH network and dynamically adjust the topological structure to adapt to the direction of pollution diffusion. For example, in a heavy metal pollution event, the mobile node group completed the mapping of the pollution boundary of a 3-square-kilometer water area within 12 minutes, with an efficiency 8 times higher than the traditional manual layout method. Through the real-time transmitted pollution concentration heat map, the emergency response team accurately laid out adsorption materials, reducing the pollution diffusion range by 65%.

[0040] Setting up an intelligent traffic scheduling strategy can achieve dynamic management of transmission priorities. Build a transmission requirement grading model, and classify data types into three levels: red warning (such as cyanide concentration exceeding the standard), orange warning (COD > 100 mg / L), and regular monitoring. When multiple types of data occur concurrently, the system gives priority to ensuring the transmission bandwidth of red warning data. For example, during an algae bloom (orange warning) and an oil spill (red warning) occurring simultaneously, the system automatically allocates 90% of the 5G channel bandwidth to the transmission of oil spill images, ensuring that the command center receives key information within 5 seconds.

[0041] Multi-source data fusion and three-dimensional river channel modeling, by integrating satellite remote sensing data, UAV oblique photography data, and lidar point clouds, construct a three-dimensional digital twin of the river channel with millimeter-level accuracy. Before a special event, the system dynamically superimposes the riverbed terrain data scanned by lidar with the water level line images taken in real time by UAVs during the period when the special event did not occur, generating a three-dimensional model of the affected area after the event. Aiming at the problem of spatial matching between the data of mobile monitoring devices (such as drifting water quality sensors) and static base maps. Set up a dynamic projection matching algorithm. By extracting the feature points of river channel markers (such as bridge piers, reef groups), and using an improved RANSAC algorithm (Random Sample Consensus) to eliminate drift errors, the spatial alignment error between mobile data and the base map is less than 0.3 meters. For example, in a case of illegal night-time discharge by a chemical plant, traditional monitoring methods were difficult to locate due to the lack of night-time optical images. This system integrates lidar night-time point clouds (identifying abnormal liquid level changes at a height of 0.8 meters around the sewage outlet) and UAV infrared thermal imaging (capturing a 2.5°C temperature difference on the surface of the drainage pipe). After matching 32 feature points through the RANSAC algorithm, it accurately locates the hidden illegal sewage outlet underwater, with a coordinate error of only 0.15 meters, and the positioning accuracy is 90% higher than that of traditional single-source data.

[0042] Space-Time Coding System and Dynamic Index. The specific steps are as follows: Design a Geohash 12-level space coding system, divide a 100-kilometer river channel into grid cells with a side length of 0.6 meters, and each cell corresponds to a unique 12-bit code. In the time dimension, use the UNIX timestamp (millisecond-level precision) combined with the leap second compensation mechanism to ensure the continuity of cross-year data. When the leap second adjustment occurs in 2023, the system automatically inserts a compensation field in the timestamp sequence to avoid data gaps. After establishing a two-way index relationship, when querying a pollution discharge event, all monitoring records (including water quality parameters, image snapshots, and flow velocity data) of the past 30 days in the grid can be retrieved within 50 ms. For example, in a cross-border river pollution dispute, the involved grid can be quickly locked through space-time coding, and more than 2,000 monitoring data in the past 72 hours in the area can be retrieved, accurately showing that the pollution concentration began to rise abnormally at 14:05 and reached the peak at 14:23, providing an immutable space-time evidence chain for liability determination, with an efficiency improvement of 400% compared to traditional manual investigations.

[0043] Multi-Physical Field Coupling Simulation of Pollutant Diffusion. Establish a two-dimensional hydrodynamic-material transport coupling model for the mathematical model and calculation optimization. The equation is as follows: ; where, is the pollutant concentration, is a three-dimensional vector, is the diffusion coefficient tensor, is the source-sink term, which is calculated in real time by the hydrodynamic module, and the diffusion coefficient D is dynamically corrected by machine learning. In a pesticide leakage accident, based on the real-time flow velocity field (1.8 m / s in the main river channel and 0.3 m / s in the recirculation area), the system predicted that the pollution front would reach the downstream drinking water intake after 8 hours, with a time error of only 9 minutes and a spatial error of <50 meters compared with the actual monitoring results. Set up a three-dimensional display platform to support the rendering of heat maps (256-level color gradient) of the pollutant diffusion process, historical trajectory backtracking (scalable to any time slice), and dynamic annotation of warning areas (such as setting a 500-meter ecological red line restricted area) to improve the efficiency of pollutant monitoring.

[0044] When using the model self-correction mechanism for virtual-real data closed-loop verification: construct a dynamic error feedback loop between the measured data and the simulation results. When there is a deviation between the measured concentration curve and the simulation prediction of a pollution event, the system triggers the Bayesian inversion algorithm, and through Markov chain Monte Carlo (MCMC) sampling, it is determined after 3,000 iterations that the diffusion coefficient D needs to be adjusted from 0.25 to 0.28 m² / s. The updated model reduces the prediction error in the subsequent 3 similar events and improves the effectiveness of the self-correction mechanism. Integrate a virtual reality (VR) accident deduction system, which can inject extreme scenarios. For example, in extreme events, the ability to select the best disposal plan within 30 seconds can be trained, with an efficiency improvement of 250% compared to traditional sand table deductions.

[0045] Build a multi - layer dynamic evaluation system to achieve data fusion with pollution sources. When the multi - layer dynamic evaluation system is executed, a three - level evaluation index system is constructed, including a water quality parameter layer (dissolved oxygen, COD, heavy metal concentration, etc.), a morphological feature layer (pollutant diffusion area, color and texture), and an environmental impact layer (distance to the water intake, coverage of sensitive ecological areas). Taking the benzene series leakage incident of a chemical plant as an example, the system integrates multi - dimensional data in real - time: Water quality layer: The benzene concentration suddenly rises from 0.1 mg / L to 8.2 mg / L (exceeding the standard by 164 times); Morphological layer: The drone identifies an 800m×50m rainbow - colored oil film belt; Environmental layer: 3.2 km downstream of the pollution belt is a fish spawning area. The weights of the indicators are dynamically calculated by the entropy weight method. During the flood season (flow velocity > 2 m / s), the weight of the morphological layer is increased to 45% (due to the fast diffusion speed). During the dry season (flow velocity < 0.5 m / s), therefore, the weight of the water quality layer accounts for 60% (pollutants are prone to accumulation). When a sensitive ecological area is detected, the weight of the environmental impact layer automatically increases from 20% to 35%.

[0046] The multi - level early warning is dynamically triggered by adopting a hierarchical response strategy, and a three - level early warning mechanism is set to be linked with the emergency response plan. In this application, when the score > 0.85, a red early warning is implemented: Automatically start the emergency treatment system, dispatch surface robots to deploy oil booms, and synchronously close the water intake; When 0.6 < score ≤ 0.85, an orange early warning is implemented: Send early warning information to the environmental protection department and increase the monitoring frequency to once every 5 minutes; When 0.4 < score ≤ 0.6, a yellow early warning is implemented: Record the data and notify the local grid officers to conduct a search. According to the above implementation method, during a heavy metal pollution incident, the system monitored that the comprehensive score reached 0.72 (orange) at 14:05 and rose to 0.89 (red) at 14:17, triggering the emergency dispatch of the robot formation, and the response time was improved by 40 minutes compared with manual decision - making. By establishing a feedback closed - loop for the early warning effect, when the actual pollution loss after a certain early warning is lower than 20% of the predicted value, the system automatically strengthens the weights of relevant rules, increasing the accuracy rate of the red early warning from 78% to 96%.

[0047] Autonomous Navigation Surface Robot Platform: Through autonomous control of the platform, intelligent perception and dynamic obstacle avoidance can be achieved as follows: The surface robot is equipped with an obstacle perception system that integrates radar and multi-camera vision. A three-dimensional semantic map of the river channel is constructed in real time through a millimeter-wave radar (detection range: 300 meters) and a high-frame-rate camera (60 fps). When detecting floating objects, bridge piers or other vessels, an improved A* algorithm is used for dynamic path planning. During execution, when the robot visually recognizes that there is a fishing boat operating at a set distance ahead, it activates the snake-shaped detour strategy, reduces the speed, and maintains the safety distance control error while keeping the pollution adsorption operation. The robot platform integrates a robotic arm, an adsorption device, and a sensor array to achieve a full-process closed-loop of "recognition - processing - verification". For example, in a chemical raw material barrel leakage accident, the robot grabs the damaged container through the robotic arm (grasping accuracy: ±2 cm), synchronously activates the ultrasonic demulsification device (frequency: 28 kHz) to decompose the surfactant, and the real-time monitoring value of the treatment efficiency reaches 89%, meeting the 85% index. In this application, the specific calculation methods of relevant technical indicators are as follows: ; Where: is the concentration of inlet pollutants, is the concentration after treatment.

[0048] Intelligent Emergency Material Scheduling Model: Improves the emergency scheduling efficiency through dynamic path planning and resource optimization. During execution, based on the improved Dijkstra algorithm, a multi-constraint optimization model considering water flow velocity, pollutant diffusion rate, and equipment transportation timeliness is constructed. The algorithm introduces real-time river channel topology data (water depth, flow velocity distribution). When predicting the movement trajectory of the pollution mass centroid, the second-order Runge-Kutta method is used to solve the advection-diffusion equation, and path planning is carried out to reduce the response time. For example, in a cross-border pollution incident, the system plans three interception paths: and obtains the evacuation distance and interception time of the optimal path: 8.2 km, estimated interception time: 42 minutes, as well as the evacuation distance and interception time of path A: 9.1 km / 48 minutes (avoiding the construction section), and the evacuation distance and interception time of the standby path B: 7.8 km / 39 minutes (requiring coordination of navigation control).

[0049] Finally, the optimal path is selected to successfully intercept 85% of the pollution mass, and the efficiency is increased by 400% compared with the traditional manual scheduling. Compared with the traditional emergency treatment system, this solution achieves a leapfrog improvement of 30% - 800% in core indicators such as disposal timeliness, treatment depth, and continuous combat ability. Through this model, the pollution control rate of the water area around a polluted area is increased from 55% to 92%, reducing the actual loss, thereby improving the early warning efficiency and accuracy of the pollutant identification and early warning method.

[0050] The blockchain system is set up as a dual-chain heterogeneous architecture. The system constructs two independently operating blockchains, including a monitoring chain and a transaction chain, to achieve classified data storage and permission isolation. The monitoring chain uses an efficient key-value pair database to store high-frequency data such as real-time water quality parameters and equipment status. Each block can accommodate 5,000 records, and the block generation interval is 3 seconds. The transaction chain selects a relational database structure to record transactional data such as early warning events and disposal logs, and supports complex queries. During execution, while the monitoring chain processes 2,000 sensor data per second, the transaction chain completely records 78 operation logs of the entire process from early warning trigger, emergency response to disposal acceptance, forming an immutable evidence chain. For example, when environmental protection law enforcement officers need to trace a certain pollution incident, they can quickly locate the relevant early warning ID through the transaction chain and directly associate all the original sensor data in the monitoring chain during that period, with the query efficiency being 10 times higher than that of the traditional single-chain structure. The dual-chain architecture enables the data write throughput to reach 12,000 TPS, a 400% increase compared to the traditional environmental monitoring system. Moreover, the blockchain also includes a VRF-enhanced consensus mechanism for randomly selecting trusted nodes. During execution, by improving the DPoS consensus algorithm, a verifiable random function (VRF) is introduced. Each verification node generates a public-private key pair, and through the VRF algorithm, a random number certificate is output. The system selects 21 verification nodes according to the sorting of the certificate hash values. During node election, node A and node B respectively generate random numbers 0x7a3f... and 0x9e1b..., and are elected as the corresponding verification nodes after sorting. Additionally, a dynamic rotation mechanism is set up to re-elect verification nodes every 15 minutes. The system automatically marks 3 nodes in the polluted area as "high-risk", reducing their election weight to 30% of the normal value to ensure the stability of the consensus network. After actual measurement, this mechanism shortens the block confirmation time from 5 seconds of the traditional PBFT to 1.2 seconds, with the consensus efficiency increasing by 316%.

[0051] An intelligent contract template library is established. This intelligent contract template library includes a standardized contract development framework to construct three types of core contract templates: a data verification contract, which automatically verifies the rationality of sensor data. During actual execution, when a certain device reports COD > 1000mg / L continuously for 5 times, it triggers the abnormal marking of the device; an early warning trigger contract, which presets 32 pollution scenario threshold conditions, and automatically generates an orange warning when the dissolved oxygen < 2mg / L and the ammonia nitrogen > 5mg / L; a traceability query contract, which supports multi-dimensional joint queries. In combination with the intelligent contract template library, a cross-chain oracle system is set up. A trusted data interaction channel is set up in the blockchain system, and through the TLS1.3 two-way authentication protocol, it is docked with the environmental protection department's database to establish a "monitoring chain - government affairs chain" cross-chain bridge. When it is necessary to retrieve the pollutant discharge permit data, the oracle node automatically encrypts the query request, and after being verified by the gateway, returns the structured data to improve the accuracy of the execution of the intelligent contract template library.

[0052] Meanwhile, a spatial index structure based on the k-d tree is constructed through multi-dimensional feature rapid retrieval, mapping the pollution feature vectors into a multi-dimensional semantic space. If a new pollution event occurs, the system matches historical cases through the following steps: extracting the key features of the event, performing a nearest neighbor search in the k-d tree, and returning the top 5 similar cases (similarity > 85%) and their handling solutions. After the solutions are returned, dynamic adaptation is performed on the contract templates. According to the case matching results, the system automatically fills in the contract elements, such as pollution cleanup contracts, compensation calculation contracts, and law enforcement basis contracts, thereby further improving the accuracy of the intelligent contract template library.

[0053] Build a multi-modal data lake to logically integrate all-dimensional data. During execution, construct a cross-media data storage architecture to uniformly store structured monitoring data, unstructured image videos, and semi-structured report documents in a distributed object storage system. Through an adaptive metadata tagging system, semantic tags such as time stamps, device IDs, and pollution types are added to each type of data. When a pollution event occurs, the system automatically collects structured data, such as continuous monitoring records of water quality sensors around the leakage point; unstructured data, such as videos and thermal imaging maps taken by drones; and semi-structured data, such as enterprise environmental emergency plans and emergency response logs. The data lake is divided into hot data SSD storage and cold data blue-ray archiving through intelligent hierarchical storage technology, reducing the data query response time from the minute level of traditional systems to within 200 ms and reducing the storage cost by 65%. During the process of aggregating data in the data lake, start the deep semantic parsing engine through knowledge extraction and entity association. When starting, use the BiLSTM-CRF model to construct a three-layer parsing system, including an entity recognition layer to extract key elements from historical event reports; a relationship extraction layer to establish associations such as "pollutant → impact → ecologically sensitive area"; and an event reconstruction layer to combine discrete entities to form a complete event chain. For example, in a pollution event, the system automatically extracts the key rule of "when the rainy season flow velocity > 2 m / s, the pollution diffusion speed increases by 40%" from historical reports and generates an association rule of "pollution diffusion rate - river flow velocity" to provide knowledge support for the subsequent generation of intelligent contracts.

[0054] Incremental knowledge fusion and graph update adopt a dynamic learning algorithm, and a closed-loop learning mechanism of "data trigger - model fine-tuning - graph iteration" is constructed during execution. When the amount of newly added pollution event data reaches the threshold, the system automatically starts incremental training, samples subgraphs of the knowledge graph (retaining core nodes and three-hop neighbor relationships), updates the embedding vectors using a contrastive learning algorithm, maintains the stability of the original knowledge structure, and adjusts the node connection weights through a topology optimization algorithm. For example, in a new pollutant event, after the system accesses the first 50 test reports, it completes within 72 hours: adding a new "pollutant" entity node, establishing a relationship edge of "PFOA → toxic effect → fish embryo malformation", updating the attributes of the "water treatment process" node, and adding a warning label of "activated carbon adsorption efficiency ≤ 30%", effectively improving the execution efficiency of the early warning method.

[0055] Construct a multi-physics field coupled digital twin. Use the finite volume method to solve the three-dimensional hydrodynamic equation, establish the digital twin, and set up a multi-physics field coupled simulation engine in the digital twin by combining the hydrodynamic module, pollutant transport module, and ecological impact assessment module; calculate the dynamic error between the measured data and the simulation results through the multi-physics field coupled simulation engine and the virtual-real data comparison algorithm, substitute the dynamic error into the model of the virtual-real combination verification system for parameter self-correction. When the parameter self-correction value continuously exceeds the set threshold, start the Bayesian inversion algorithm to optimize the diffusion coefficient parameter. After self-correction, substitute it into the XR training platform, integrate virtual reality accident deduction and augmented reality deduction to achieve accurate positioning of pollution events and early warning of the severity of hazards. The above-mentioned three-dimensional hydrodynamic simulation engine is established based on the finite volume method (FVM), discretizes the river channel into several grid units, and describes the water flow movement with the Navier-Stokes equation. When using the coupled pollutant transport module, the Lagrangian particle tracking method is used to simulate the diffusion process, where the advection term corresponds to driving the particle displacement based on real-time flow velocity field data, the diffusion term corresponds to introducing a turbulence pulsation model (k-ε model) to enhance the accuracy, and the reaction term corresponds to integrating the chemical kinetics database. This improves the calculation speed of the digital twin and simulates the process of the pollution zone spreading to the downstream of the polluted river channel, thereby improving the coincidence degree between the simulation results and the actual monitoring trajectory.

[0056] In the above-mentioned process of simulating the diffusion of the pollution belt to the downstream of the polluted river, dynamic error correction and Bayesian inversion are used, combined with the setting of virtual and real data closed-loop feedback mechanism and virtual and real data comparison algorithm, to compare the simulation results with the actual values ​​of the monitoring points every second. When the error between the diffusion rate of the monitoring area and the simulation result continues to be greater than 15%, the correction process is started. The correction process is as follows: data cleaning, excluding sensor fault data; sensitivity analysis, determining the contribution of the diffusion coefficient to the error; Bayesian inversion, based on Markov chain Monte Carlo (MCMC) sampling, optimizing the diffusion coefficient in iteration; model hot update, adjusting the optimized diffusion coefficient. For example, in an oil spill incident, the mechanism completed 3 rounds of parameter optimization within 2 hours after the accident, reducing the 72-hour diffusion prediction error from the initial 22% to 4.7%.

[0057] The XR fusion training platform includes a multi-modal emergency simulation system to build a virtual-real fusion XR training system. The XR training system includes virtual reality accident reconstruction, which generates interactive three-dimensional scenes based on historical pollution data; augmented reality on-site superposition, which uses smart glasses to superimpose the simulated pollution belt outline on the real river; mixed reality collaborative drills, where remote commanders use holographic projections to mark the key points of disposal in real time. Through the above settings, pollution drills can be completed through the XR platform, thereby improving the efficiency of subsequent identification and treatment of pollutants in actual rivers.

[0058] The process described above with reference to the flowchart can be implemented as a computer software program. Embodiments disclosed in the present invention include a computer program product that includes a computer program carried on a computer-readable medium, and the computer program includes program code for performing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication part, and / or installed from a removable medium. When the computer program is executed by a central processing unit (CPU), the above functions defined in the method of the present application are performed. It should be noted that the computer-readable medium described above in the present application can be a computer-readable signal medium or a computer-readable storage medium or any combination of the two. A computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples of the computer-readable storage medium can include, but are not limited to: an electrical connection with one or more wire segments, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, the computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or combined with an instruction execution system, apparatus, or device. And in the present application, a computer-readable signal medium can include a data signal propagated in a baseband or as part of a carrier wave, which carries computer-readable program code. Such a propagated data signal can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. A computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or combined with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted by any appropriate medium, including but not limited to: wireless, wire, optical cable, RF, etc., or any suitable combination of the above.

[0059] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions, and operations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in the flowchart or block diagram may represent a module, a segment of a program, or a part of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order from that marked in the accompanying drawings. For example, two consecutive blocks shown may actually be executed substantially in parallel, and they may sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, and the combinations of blocks in the block diagram and / or flowchart, can be implemented by a dedicated hardware-based system that performs the specified functions or operations, or can be implemented by a combination of dedicated hardware and computer instructions.

[0060] Those skilled in the art should understand that the embodiments of the present invention described above and shown in the accompanying drawings are only examples and do not limit the present invention. The objectives of the present invention have been fully and effectively achieved. The functions and structural principles of the present invention have been demonstrated and explained in the embodiments. Without departing from the said principles, any variations or modifications can be made to the embodiments of the present invention.

Claims

1. A method for identifying and warning pollutants of river patrol pollution sources, characterized in that, The method includes: Combining image acquisition devices in the air, underwater and on the shore to establish a three-dimensional data acquisition architecture and construct a three-dimensional stereo monitoring network system; Obtaining the spatial coordinates of monitoring nodes according to the three-dimensional stereo monitoring network system, unifying the spatial coordinates of each monitoring node through the GNSS differential positioning system, using the Kalman filtering technology to perform real-time calibration of the coordinates of the positioned monitoring nodes, and combining the spatial coordinates of the monitoring nodes and the timestamp alignment algorithm to realize the synchronization of the spatio-temporal coordinates of multi-source heterogeneous data of different monitoring nodes, and establishing a unified spatio-temporal reference framework; According to the obtained spatio-temporal coordinates of the monitoring nodes, preprocessing the data at the device end of the monitoring nodes to realize the deployment of edge computing nodes; Constructing a pollutant feature library and a water quality parameter correlation model through the spectral characteristics, water quality parameter correlation characteristics and visual morphological characteristics of river pollutants collected by the monitoring nodes, and obtaining river flow, pollutant diffusion rate and historical pollution data; Establishing an adaptive threshold warning mechanism, using the dynamic fuzzy comprehensive evaluation method to evaluate data such as river flow, pollutant diffusion rate and historical pollution events, constructing a spatio-temporal composite warning model, and positioning pollutants and grading the warning of the harm degree through the spatio-temporal composite warning model.

2. The pollutant identification and early warning method for river patrol pollution sources according to claim 1, characterized in that, When the image acquisition device in the air performs image processing, it adopts a deep learning algorithm for multi-scale feature fusion, and the specific content is as follows: Constructing an improved U-Net++ network architecture, introducing ResNet50 as the backbone network in the encoder part, adding an attention mechanism module in the decoder part, and establishing a cross-layer feature fusion channel; Setting a bidirectional reflectance distribution function model and an image enhancement module based on a physical model; establishing a multi-task learning framework, using a weighted cross-entropy loss function to balance the problem of class imbalance, and realizing that the UAV image processing device simultaneously completes pollutant recognition, concentration inversion and diffusion trend prediction; establishing an online incremental learning mechanism, realizing the collaborative evolution of the models of each edge node through the federated learning framework, setting a confidence threshold to automatically screen high-quality samples to update the model parameters of each edge node; constructing a virtual-real combined verification system, using digital twin technology to create a three-dimensional simulation environment of the river, and injecting typical pollution scenario data for model verification and optimization.

3. The method for identifying and warning pollutants of river patrol pollution sources according to claim 2, wherein, When the image acquisition device in the air performs image processing, it adopts an adaptive topology optimization method, and the specific process of the adaptive topology optimization method is as follows: By developing an energy-aware routing algorithm, establishing an optimization model for node remaining energy, link quality and image transmission requirements, and combining the optimization model and a hybrid networking protocol to establish a dual-mode transmission integrating LoRaWAN low-power wide-area communication and 5G low-latency communication; establishing an abnormal node self-repair mechanism, if a node failure is detected, combining blockchain and a relay node selection algorithm, and using an improved particle swarm optimization method to quickly reconstruct the optimal communication path; setting mobile emergency monitoring nodes, and configuring self-propelled surface robots and underwater ROVs.

4. The pollutant identification and early warning method for river patrol pollution sources according to claim 3, characterized in that, The specific process of the unified spatio-temporal reference framework is as follows: Combined with satellite remote sensing data, UAV oblique photography data and lidar point cloud data, a three-dimensional river channel model is established. Through the dynamic projection matching algorithm, the spatial registration of mobile monitoring data and static maps is realized, and the improved RANSAC algorithm is used to eliminate the errors in the spatial registration process; a spatio-temporal coding system is designed to establish a two-way indexing relationship between the monitoring data and the river channel spatial units. Each spatial unit adopts Geohash coding, and the time dimension adopts the UNIX timestamp and leap second compensation mechanism; a pollutant diffusion simulation model is constructed by coupling the two-dimensional hydrodynamic model and the mass transfer equation: ； Among them, is the pollutant concentration, is the flow velocity vector, is the diffusion coefficient tensor, is the source-sink term; Develop a 3D display platform for W, which supports pollutant diffusion simulation, historical trajectory backtracking, and dynamic rendering of warning areas to achieve multi-dimensional data visualization.

5. A method for identifying and warning pollutants of river patrol pollution sources according to claim 4, characterized in that, The adaptive threshold warning mechanism adopts an improved fuzzy comprehensive evaluation algorithm, and the specific content of the improved fuzzy comprehensive evaluation algorithm is as follows: Construct a multi-layer evaluation index system, including the water quality parameter layer, the morphological feature layer and the environmental impact layer; establish an adaptive weight distribution mechanism according to the multi-layer evaluation index system, and based on the historical pollution event database and the real-time river channel flow data, use the entropy weight method to dynamically calculate the index weights; combine the multi-layer evaluation index system and the adaptive weight distribution mechanism to construct a non-linear membership function library, forming trapezoidal, Gaussian and S-shaped functions for different pollutant types to generate pollution risk levels; develop a multi-level warning trigger strategy. If the comprehensive score is greater than the highest set threshold, a red warning is launched. If the comprehensive score is less than the highest set threshold and greater than the lowest set threshold, an orange warning is launched and the emergency response system is linked; implement an online model optimization mechanism, and use the reinforcement learning framework to dynamically adjust the fuzzy rule base according to the warning accuracy.

6. The method for identifying and warning pollutants of a river patrol pollution source according to claim 5, wherein, The spatio-temporal compliance warning model includes the standards for pollutant treatment, and the specific standards are as follows: An autonomous navigation surface robot platform is equipped with an obstacle perception module that integrates radar and multi-camera vision; a modular pollutant treatment unit is equipped with an automatic spreading mechanism for oil-absorbing felt, an ultrasonic demulsification device and an electrochemical degradation reactor. The processing efficiency of the autonomous navigation surface robot platform meets the following conditions: ； Wherein: is the concentration of influent pollutants, is the concentration after treatment; It also includes a distributed energy system and an intelligent emergency material dispatching model. The distributed energy system adopts hybrid power supply. The intelligent emergency material dispatching model plans the optimal interception path based on the improved Dijkstra algorithm to realize the real-time calculation of the movement trajectory of the pollution mass centroid.

7. The method for identifying and warning pollutants of river patrol pollution sources according to claim 6, characterized in that, The deployment of the edge computing nodes is realized through the blockchain system, and the specific content of the blockchain system is as follows: A double-chain heterogeneous architecture, which stores real-time monitoring data through the chain, and the transaction chain records warning events and disposal logs; establish a lightweight consensus mechanism, adopt the improved DPoS algorithm, elect multiple verification nodes and introduce VRF random numbers; Combine the double-chain heterogeneous architecture and the lightweight consensus mechanism to construct a contract template library, and complete the data verification contract, warning trigger contract and traceability query contract through the contract template library; set up a cross-chain oracle system, and dock with the environmental protection department database through the TLS-N two-way authentication protocol to realize the trusted interaction of data on and off the chain in the blockchain system.

8. The method for identifying and warning pollutants of river patrol pollution sources according to claim 7, wherein, The contract template library includes the following content: Set up a multi-modal data lake to integrate structured monitoring data, unstructured image videos and semi-structured report documents; Extract entity relation triples from historical event reports by combining a knowledge extraction engine and a BiLSTM-CRF model; establish an incremental knowledge fusion algorithm and use the D-S evidence theory to synthesize different confidence values; establish a case-based reasoning mechanism by combining the incremental knowledge fusion algorithm and a knowledge graph dynamic update strategy, and quickly retrieve similar historical scenarios through a k-d tree.

9. The method for identifying and warning pollutants of a river patrol pollution source according to claim 8, characterized in that, The specific content of the spatio-temporal composite early warning model is as follows: Use the finite volume method to solve the three-dimensional hydrodynamic equations, establish a digital twin, and set up a multi-physics field coupling simulation engine in the digital twin by combining a hydrodynamic module, a pollutant transport module, and an ecological impact assessment module; calculate the dynamic error between the measured data and the simulation results through the multi-physics field coupling simulation engine and a virtual-real data comparison algorithm, substitute the dynamic error into the model of the virtual-real combination verification system for parameter self-correction, and when the parameter self-correction value continuously exceeds the set threshold, start the Bayesian inversion algorithm to optimize the diffusion coefficient parameter, and after completing self-correction, substitute it into the XR training platform.

10. A system for river pollution inspection, characterized in that, The system is used to execute a method for identifying and warning pollutants from a river patrol pollution source according to any one of claims 1-9.

Images