Flood prevention and typhoon prevention monitoring and early warning method based on visual large model and augmented reality

Abstract

The invention provides a flood and typhoon prevention monitoring and early warning method based on a visual large model and augmented reality, and belongs to the technical field of flood and typhoon prevention, and the method comprises the steps: collecting multi-source data of a flood and typhoon prevention key region, and transmitting the multi-source data to a data processing center; performing analysis and anomaly detection early warning on the multi-source data by using a pre-trained visual large model to obtain a large model analysis result, and transmitting the large model analysis result to a data processing center; and the augmented reality equipment displays the data of the data processing center in a real scene in a holographic projection manner in an overlapping manner, and executes a corresponding action in the holographic projection according to the user interaction operation. The flood and typhoon prevention monitoring system has the beneficial effects that deep analysis is performed on multi-source data through a computer vision large-scale model, visual marking and real-time display of monitoring results are realized in combination with augmented reality equipment, the accuracy of flood and typhoon prevention monitoring, the timeliness of early warning, the high efficiency of information transmission and the safety and optimizability of the system are improved, the emergency disposal efficiency is improved, and the safety and reliability of flood and typhoon prevention monitoring are improved. The disaster loss is reduced to the greatest extent.

Description

Technical Field

[0001] This invention relates to the field of flood and typhoon prevention technology, and in particular to a flood and typhoon prevention monitoring and early warning method based on visual large models and augmented reality. Background Technology

[0002] Flood and typhoon prevention are crucial tasks for safeguarding people's lives and property and maintaining social stability. Traditional flood and typhoon monitoring methods rely heavily on manual patrols and simple sensor equipment. Manual patrols are inefficient, subjective, and unable to cover large areas quickly; moreover, they are highly dangerous in severe weather. Simple sensor equipment can only collect single environmental parameters and cannot provide a comprehensive assessment of complex disaster scenarios.

[0003] With the development of computer vision technology, some flood and typhoon prevention monitoring systems have incorporated image recognition technology. However, traditional image recognition models lack sufficient accuracy in complex weather conditions (such as heavy rain and dense fog) and dynamically changing scenarios (such as rapid flooding), making it difficult to accurately capture disaster characteristics and trends. Simultaneously, in the information transmission stage, existing systems lack an intuitive and efficient way to present monitoring results to on-site users, resulting in slow decision-making response and low efficiency in disaster relief. Furthermore, traditional monitoring systems suffer from poor data transmission security and lack dynamic optimization mechanisms in data management, making it difficult to meet the needs of modern flood and typhoon prevention work. Therefore, there is an urgent need for a flood and typhoon prevention monitoring and early warning method that can adapt to complex scenarios, achieve accurate monitoring and efficient information transmission, and possess data security and self-optimization capabilities. Summary of the Invention

[0004] To address the above technical problems, this invention provides a flood and typhoon prevention monitoring and early warning method based on visual large models and augmented reality.

[0005] The technical problem solved by this invention can be achieved by the following technical solutions: A flood and typhoon prevention monitoring and early warning method based on large visual models and augmented reality includes: Step S1: Collect multi-source data from key flood and typhoon prevention areas and transmit the multi-source data to the data processing center; Step S2: Analyze the multi-source data and perform anomaly detection and early warning using a pre-trained visual large model, obtain the large model analysis results, and transmit them to the data processing center. Step S3: The augmented reality device overlays the multi-source data from the data processing center and the analysis results of the large model into the real scene in the form of a holographic projection, and performs corresponding actions in the holographic projection according to the user's interactive operation.

[0006] Preferably, the multi-source data includes environmental perception data, image and video data; step S1 includes: The environmental perception data is acquired using a first data acquisition module; wherein, the first data acquisition module includes a composite sensor network composed of several sensor nodes; The image and video data are acquired using a second data acquisition module. The second data acquisition module includes several first camera units fixedly installed at preset locations in key flood and typhoon prevention areas, and a second camera unit mounted on a drone. The drone is used to autonomously plan inspection paths and conduct inspections according to the inspection paths. The second camera unit is used to acquire the image and video data during the inspection process.

[0007] Preferably, the environmental sensing data includes one or more combinations of rainfall, water level, wind speed, wind direction, water body spectrum, and vibration signals.

[0008] Preferably, the multi-source data is transmitted to the data processing center via a quantum communication link.

[0009] Preferably, the training steps of the large visual model include: Collect relevant image data for flood and typhoon prevention, including river images at different water levels, images of building damage before and after typhoons, and images of flooded areas; A deep learning model accelerated by quantum computing is constructed and trained based on the flood and typhoon prevention related image data to identify disaster-related information; wherein, the disaster-related information includes one or more combinations of water level changes, flood inundation area, building damage degree, and typhoon eye location.

[0010] Preferably, step S2 includes: Step S21A: The edge computing node processes and analyzes the image and video data in the multi-source data, extracts key features from the images, and transmits the key features to the visual big model. In step S22A, the visual big model identifies the key features to determine the boundaries of the flood-inundated area and monitor the trajectory of dangerous objects moving in the typhoon.

[0011] Preferably, step S2 further includes: Step S21B: The visual big model analyzes the environmental perception data and image and video data in the multi-source data to obtain the analysis results; Step S22B: When the environmental perception data is detected to deviate from the preset threshold range or to exhibit abnormal characteristics, a warning message of the corresponding level is generated. Step S23B: Send the warning information, related images, and analysis results to the data processing center.

[0012] Preferably, the step of determining the preset threshold range includes: Based on historical multi-source data, generative adversarial networks are used to generate monitoring data samples for non-flood and non-typhoon weather conditions. For various monitoring indicators, based on the monitoring data samples, preset threshold ranges that vary with time and season are set.

[0013] Preferably, the user interaction operation is implemented using one or more combinations of gestures, voice, and eye tracking.

[0014] Preferably, after step S3, the method further includes: Step S4: Construct a digital twin model of the flood and typhoon prevention area; Step S5: Perform virtual simulation and simulation verification based on user feedback information from the digital twin model and augmented reality device, optimize and update the large visual model, and perform information annotation and interaction optimization on the augmented reality device.

[0015] The advantages or beneficial effects of the technical solution of this invention are as follows: This invention uses a large computer vision model to perform in-depth analysis of multi-source data, and combines it with augmented reality devices to achieve intuitive labeling and real-time display of monitoring results. This improves the accuracy of flood and typhoon prevention monitoring, the timeliness of early warning, the efficiency of information transmission, and the security and optimizability of the system, thereby enhancing the efficiency of emergency response and minimizing disaster losses. Attached Figure Description

[0016] Figure 1 A flowchart illustrating the flood and typhoon monitoring and early warning method based on visual large model and augmented reality, as a preferred embodiment of the present invention; Figure 2 This is a schematic diagram of the visual large model training process in a preferred embodiment of the present invention; Figure 3 This is a flowchart illustrating the real-time edge computing analysis process in a preferred embodiment of the present invention. Figure 4 This is a schematic diagram of the process for anomaly detection and intelligent early warning generation in a preferred embodiment of the present invention; Figure 5 This is a schematic diagram of the process for determining the preset threshold range in a preferred embodiment of the present invention; Figure 6 This is a schematic diagram of the process for model optimization, information annotation, and interaction optimization in a preferred embodiment of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.

[0019] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, but this is not intended to limit the scope of the invention.

[0020] In a preferred embodiment of the present invention, based on the aforementioned problems existing in the prior art, a flood and typhoon prevention monitoring and early warning method based on large-scale visual models and augmented reality is provided. This method utilizes computer vision (CV) large-scale model technology to perform in-depth analysis of multi-source data, combined with augmented reality (AR) devices to achieve intuitive labeling and real-time display of monitoring results. Simultaneously, it leverages innovative technologies such as quantum encrypted communication, edge computing, blockchain, and digital twins to improve the accuracy of flood and typhoon prevention monitoring, the timeliness of early warning, the efficiency of information transmission, and the security and optimizability of the system, thereby enhancing the efficiency of emergency response and minimizing disaster losses.

[0021] like Figure 1 As shown, the methods for monitoring and early warning of flood and typhoon prevention include: Step S1: Collect multi-source data from key flood and typhoon prevention areas and transmit the multi-source data to the data processing center; Step S2: Analyze and detect anomalies in multi-source data using a pre-trained visual big model, obtain the big model analysis results, and transmit them to the data processing center. Step S3: The augmented reality device overlays the multi-source data and large model analysis results from the data processing center onto the real scene in the form of a holographic projection, and performs corresponding actions in the holographic projection based on user interaction.

[0022] Specifically, in this embodiment, a large computer vision model is used to perform in-depth analysis of multi-source data to achieve anomaly detection and early warning. Combined with augmented reality equipment, the monitoring results and the analysis results of the large model are superimposed and displayed in the real scene in the form of holographic projection, so as to realize intuitive labeling and real-time display of monitoring results, improve the accuracy of flood and typhoon monitoring, the timeliness of early warning, the efficiency of information transmission, and the security and optimizability of the system, thereby improving the efficiency of emergency response and minimizing disaster losses.

[0023] Furthermore, the multi-source data includes environmental perception data, image and video data; the environmental perception data is acquired by the first data acquisition module, which includes one or more combinations of rainfall, water level, wind speed, wind direction, water body spectrum, and vibration signal; the image and video data is acquired by the second data acquisition module.

[0024] In this embodiment, the first data acquisition module includes a composite sensor network composed of several sensor nodes.

[0025] Specifically, in key areas for flood and typhoon prevention, a composite sensor network integrating quantum encrypted communication technology, multimodal data acquisition and dynamic adaptive transmission scheduling technology is deployed. The composite sensor network consists of several sensor nodes, each of which is equipped with an environmental sensor. The sensor types of different sensor nodes may be the same or different.

[0026] For example, sensors include, but are not limited to, environmental sensors such as rainfall sensors, water level sensors, wind speed sensors, and wind direction sensors. In addition to traditional rainfall sensors, water level sensors, wind speed sensors, and wind direction sensors, miniature spectral sensors and vibration sensors can also be integrated. Spectral sensors can monitor the concentration of pollutants and algae growth in water in real time by analyzing water spectral data; vibration sensors are installed in critical facilities such as dams and bridges to capture minute vibration signals generated by water flow impact and geological changes in real time. These multimodal data are then fused with traditional environmental data to construct a more comprehensive disaster early warning indicator system.

[0027] Each sensor node within the network is equipped with an intelligent transmission scheduling module, incorporating a dynamic adaptive transmission scheduling algorithm. This algorithm, based on reinforcement learning principles, senses network load and data importance in real time. When network congestion is detected, it prioritizes the transmission of high-priority real-time early warning data, delaying or compressing non-urgent information such as historical data. It also automatically adjusts data acquisition and transmission frequencies based on environmental changes. For example, during stable weather, the acquisition frequency is reduced to save energy; when a disaster approaches or environmental data fluctuations intensify, the acquisition frequency is dynamically increased, balancing transmission efficiency and energy consumption while ensuring data accuracy. Furthermore, the sensor network employs a self-healing and fault-tolerant architecture, with each node possessing fault self-detection and collaborative repair capabilities. When a node detects a fault or communication link interruption, it immediately sends a distress signal to neighboring nodes. These neighboring nodes then collaborate to temporarily take over the data acquisition and transmission tasks of the faulty node. Simultaneously, the system initiates a distributed fault diagnosis program, using a Bayesian network model to quickly locate the cause of the fault and attempting self-repair through remote code updates and parameter reconfiguration, ensuring the sensor network operates stably 24 / 7.

[0028] In this embodiment, the second data acquisition module includes several first camera units fixedly installed at preset locations in key flood and typhoon prevention areas, and a second camera unit mounted on a drone. The drone is used to autonomously plan inspection paths and conduct inspections according to the inspection paths. The second camera unit is used to collect image and video data during the inspection process.

[0029] Specifically, the first camera unit uses a high-definition camera equipped with an intelligent adaptive optics system. This high-definition camera can automatically adjust its focus, aperture, and exposure time based on ambient light and weather changes to consistently acquire clear image data.

[0030] Specifically, the high-definition camera's built-in light sensor collects ambient light intensity data in real time, converts it into electrical signals, and transmits them to the built-in microprocessor. The weather sensing module uses humidity sensors, raindrop sensors, and other sensors, combined with machine learning algorithms, to identify and classify weather types, including but not limited to sunny, cloudy, rainy, and foggy weather. Based on the environmental data, the microprocessor calls adaptive algorithms for calculations. When it detects low light levels due to rain, the algorithm sends instructions to the drive circuit to automatically increase the aperture to increase light intake, extend the exposure time to capture more light, and optimize the focal length based on image sharpness feedback signals. Unlike traditional cameras that rely on manual or preset parameter adjustments, this technology can achieve millisecond-level dynamic response in complex weather environments, avoiding monitoring blind spots caused by human intervention.

[0031] Specifically, the second camera unit uses drone camera equipment with autonomous path planning capabilities. Utilizing advanced environmental perception technology, the drone senses weather conditions, terrain, and other information in real time during flight, autonomously planning the optimal inspection route to comprehensively and meticulously collect images and videos of key areas such as rivers, embankments, low-lying urban areas, and coastlines.

[0032] Specifically, the drone's onboard LiDAR acquires real-time terrain height data through 360-degree scanning, while its visual sensors use a convolutional neural network (CNN) to identify obstacle outlines and categories. During flight, the system inputs environmental perception data into the path planning module, employing an improved Dijkstra algorithm combined with reinforcement learning. The Dijkstra algorithm provides the globally optimal initial path solution, while reinforcement learning constructs a "state-action-reward" model, enabling the drone to continuously learn and adjust its path selection strategy in different environmental scenarios. For example, when encountering sudden heavy rain that reduces local visibility, the drone can quickly replan its route, prioritizing inspections of flood-prone areas. Compared to traditional fixed-route inspection drones, this technology can dynamically adjust inspection strategies based on real-time weather and terrain changes, increasing inspection coverage by over 40% and improving data collection efficiency in key areas by 60%, effectively solving problems such as missed inspections and duplicate inspections in traditional inspection methods.

[0033] Furthermore, multi-source data is transmitted to the data processing center via a quantum communication link.

[0034] Specifically, these environmental sensors integrate quantum encrypted communication technology to collect environmental data such as rainfall, water level, wind speed, wind direction, water spectrum, and vibration signals in real time at a high frequency, such as once per second. With the absolute security of quantum encrypted communication, the data is securely transmitted to the data processing center through a dedicated quantum communication link. Compared with traditional encrypted transmission methods, this completely eliminates the risk of data being eavesdropped or tampered with at the physical layer, ensuring that the data is not stolen or tampered with during transmission.

[0035] Furthermore, such as Figure 2 As shown, the training steps for a large visual model include: A1. Collect image data related to flood and typhoon prevention, including river images at different water levels, images of building damage before and after typhoons, and images of flooded areas; A2. Construct a deep learning model accelerated by quantum computing, and train the model based on flood and typhoon prevention related image data to identify disaster-related information; among which, disaster-related information includes one or more combinations of water level changes, flood inundation area, building damage degree, and typhoon eye location.

[0036] Specifically, the large-scale visual model analysis and processing includes quantum-enhanced deep learning model training. This involves collecting massive amounts of flood and typhoon-related image data, covering river images at different water levels, images of building damage before and after typhoons, and images of flooded areas. A quantum-accelerated deep learning model is constructed by introducing qubits into traditional attention-based deep learning models (Transformer), convolutional neural networks (CNNs), and other architectures. During training, quantum computing utilizes superposition and entanglement properties to simultaneously process an exponential number of training parameter combinations. Compared to the classical method of iterative optimization parameter by parameter, the powerful parallel computing capabilities of quantum computing significantly shorten model training time. For example, when processing convolutional neural network models with millions of parameters, quantum-accelerated training time can be reduced to 1 / 100th of traditional methods.

[0037] Traditional CNNs are limited by classical binary operations and have difficulty handling the nonlinear features of high-dimensional data. In contrast, the quantum convolutional neural network (QCNN) in this embodiment replaces the weight parameters in the classical convolutional layer with quantum state matrices and achieves feature extraction through quantum gate operations. The quantum convolutional kernel can capture more complex high-dimensional relationships in the data, improve the model's ability to learn complex image features, and enable it to accurately identify disaster-related information such as water level changes, flood inundation range, building damage, and typhoon eye location.

[0038] Traditional convolutional neural networks require multiple rounds of computation to be performed sequentially, while quantum computing can directly encode input data into quantum states and use quantum parallelism to perform multi-path analysis of the data. For example, in the preprocessing stage of flood control image data, quantum convolutional neural networks can simultaneously complete multiple tasks such as flood range identification and flow velocity calculation.

[0039] Furthermore, such as Figure 3 As shown, step S2 includes: In step S21A, the edge computing node processes and analyzes the image and video data from the multi-source data, extracts key features from the images, and transmits the key features to the large visual model. In step S22A, the visual big model identifies key features to determine the boundaries of the flood-inundated area and monitor the trajectories of dangerous objects moving in the typhoon.

[0040] Specifically, large-scale visual model analysis and processing also includes real-time edge computing analysis. Edge computing nodes are deployed near the data acquisition point, and the acquired real-time image and video data are first processed and analyzed at these edge computing nodes. Then, leveraging the low latency of edge computing, key features in the images are quickly extracted, and this feature data is transmitted to the large model via a high-speed network for in-depth analysis. The large model combines the feature data provided by edge computing and utilizes its powerful computing capabilities for real-time recognition and analysis. For example, advanced image segmentation algorithms, such as the improved U-Net algorithm based on deep learning, can accurately determine the boundaries of flood-inundated areas; target tracking algorithms, such as the improved Kalman filter algorithm, can monitor the trajectories of moving dangerous objects during typhoons.

[0041] Specifically, the preliminary processing steps include: using an image enhancement algorithm based on histogram equalization to enhance image contrast and make image details clearer; and using a median filtering algorithm to remove image noise.

[0042] The key feature extraction steps include: utilizing an improved scale-invariant feature transform (SIFT) algorithm, combined with the parallel processing capabilities of edge computing, to quickly extract key features such as corners and edges from the image. Compared to existing technologies that typically process data centrally in the cloud, resulting in high transmission latency and network pressure, this embodiment performs preliminary processing and key feature extraction at edge computing nodes. This fully leverages the low latency and localized processing advantages of edge computing, reducing data transmission volume and latency, and improving overall analysis efficiency.

[0043] Furthermore, such as Figure 4 As shown, step S2 further includes: Step S21B: The visual big model analyzes the environmental perception data and image and video data from the multi-source data to obtain the analysis results; Step S22B: When environmental perception data is detected to deviate from the preset threshold range or to exhibit abnormal characteristics, a warning message of the corresponding level is generated. Step S23B: Send the warning information, relevant images, and analysis results to the data processing center.

[0044] Specifically, the visual big data analysis and processing also includes anomaly detection and intelligent early warning generation. The visual big data model analyzes and monitors data in real time. When it detects that the data deviates from the normal sample range or shows abnormal features (such as tiny cracks in a dam or signs of abnormal seawater surging), it immediately generates early warning information of the corresponding level (such as yellow, orange, and red warnings) and sends the warning information, related images, and analysis results to the data processing center via quantum encrypted communication.

[0045] The large-scale model determines whether features are abnormal through threshold judgment and machine learning anomaly detection algorithms. First, a statistical distribution model of normal samples is established based on historical monitoring data (such as dam stress and water level fluctuations), and reasonable numerical threshold ranges are set. Second, an anomaly detection model is trained using algorithms such as Isolation Forest and One-Class SVM, triggering an alert when real-time data deviates significantly from the normal pattern. For example, if the width of cracks on the dam surface exceeds three times the standard deviation of the historical mean, or if the seawater flow velocity suddenly increases by 150% in a short period of time, the system will determine it as an abnormal feature.

[0046] Furthermore, such as Figure 5 As shown, the steps for determining the preset threshold range include: B1. Based on historical multi-source data, generative adversarial networks are used to generate monitoring data samples for non-flood and non-typhoon weather conditions; B2 sets preset threshold ranges that vary with time and season based on monitoring data samples for various monitoring indicators.

[0047] Specifically, based on historical monitoring data and normal scenario models, a generative adversarial network (GAN) based on self-supervised learning is used to dynamically generate monitoring data samples under normal conditions, and dynamic threshold ranges for various monitoring indicators are set, such as warning water levels and dangerous wind speeds that change with time and season.

[0048] A temporal constraint mechanism is introduced into Generative Adversarial Networks (GANs). The temporal continuity of data samples is strengthened during adversarial training. Through a multi-stage progressive training strategy, the generator gradually masters the evolution patterns of monitoring data at different time scales, solving the problem of sample discontinuity that often occurs in traditional GANs when processing time-series data.

[0049] The construction process of Generative Adversarial Networks (GANs) includes building a dual-branch generator with a spatiotemporal attention module to process spatial features and temporal series features respectively, while adding a multi-scale feature fusion layer to the discriminator to improve sensitivity to abnormal patterns. Compared to the single structure of traditional GANs, this architecture is more suitable for the multi-source heterogeneous data characteristics in flood and typhoon prevention scenarios.

[0050] Generative Adversarial Networks (GANs) employ a semi-supervised learning framework, using a small amount of labeled data to guide the generation process. Through adversarial training, the feature mapping function of the generator is dynamically adjusted, ensuring that the generated data not only conforms to historical distributions but also captures potential normal patterns caused by environmental changes. This method overcomes the limitation of traditional GANs that rely on large amounts of labeled data, making it more suitable for data-scarce problems in real-world monitoring scenarios.

[0051] Furthermore, augmented reality devices can be AR glasses integrating holographic projection technology or handheld AR terminal devices equipped with holographic interactive functions. These devices are worn by on-site users and establish an ultra-stable connection with the data processing center through a high-speed wireless network based on mobile communication technology. They receive real-time images of the real scene and early warning information and monitoring data obtained from the analysis of large visual models. Through holographic projection technology, the flood inundation boundary is presented as a red light screen, dangerous buildings are marked with 3D icons, and real-time water level values ​​are displayed in a floating window.

[0052] Augmented reality devices utilize advanced spatial computing and holographic projection technology to realistically overlay and display early warning information obtained from visual model analysis, multi-source monitoring data (such as real-time water levels and flood boundaries), and the locations of dangerous areas onto real-world scenes in the form of holographic projections. For example, flood boundaries are presented as dynamic red holographic light screens, the locations of dangerous buildings are marked with flashing 3D holographic icons, and real-time water level values ​​and warning levels are displayed in floating holographic windows.

[0053] Furthermore, user interaction is achieved through one or more combinations of gestures, voice, and eye tracking.

[0054] Specifically, users can interact with holographic information through various methods such as gestures, voice, and eye tracking. For example, they can zoom in to view detailed information about a specific area, query historical disaster data for the surrounding area, or request the system to provide the best rescue route planning to a dangerous area.

[0055] Augmented reality devices combine high-precision geographic information system (GIS) data with augmented reality path planning algorithms to generate and display clear navigation routes to dangerous areas in real time within augmented reality scenes.

[0056] The data processing center has established a distributed flood and typhoon prevention database based on blockchain technology. Environmental perception data, image and video data, large model analysis results, and early warning information collected by various sensors are securely and reliably stored on multiple nodes through blockchain's distributed storage and encryption technology, ensuring that the data is tamper-proof and traceable.

[0057] The data processing center uses big data analytics tools to deeply mine historical data and combines them with machine learning algorithms to summarize the patterns of disaster occurrence, evolution, and the characteristics of flood and typhoon prevention in different regions.

[0058] Furthermore, such as Figure 6 As shown, step S3 is followed by: Step S4: Construct a digital twin model of the flood and typhoon prevention area; Step S5 involves performing virtual simulation and simulation verification based on user feedback information from the digital twin model and augmented reality devices, optimizing and updating the large visual model, and performing information annotation and interaction optimization on the augmented reality devices.

[0059] Specifically, this embodiment employs a digital twin-driven optimization feedback mechanism to construct a digital twin model of the flood and typhoon prevention area. Real-time data drives the digital twin model to maintain synchronization with the real physical world. Based on monitoring results and feedback information from practical applications, virtual simulation and verification are performed using the digital twin model. On one hand, the results are used to optimize and update the larger model. On the other hand, combined with user feedback from augmented reality devices, virtual reality (VR) technology is used to simulate augmented reality interaction processes within the digital twin environment, further optimizing the display method and interactive functions of information annotations, and improving system usability.

[0060] The large-scale model optimization and update process includes: preprocessing newly acquired disaster image data such as denoising, enhancement, and feature extraction, and unifying the data format; simultaneously, normalizing the data from different scenarios simulated by the data twin model to ensure comparability of the two types of data in terms of scale and dimension; then, using information such as geographic coordinates and timestamps, accurately aligning the disaster image data with the simulated scenario data in spatial and temporal dimensions, and merging the two into a new dataset using feature fusion algorithms, such as weighted fusion and deep learning-based fusion networks; using the fused new dataset as input, using optimization methods such as backpropagation and genetic algorithms, comparing the model's prediction results with the features and parameters in the actual disaster images, calculating the loss function, and iteratively updating and adjusting the structural parameters and weights of the data twin model to make the model output more closely match the real disaster scenario and improve the model's ability to recognize complex scenarios.

[0061] To optimize the display and interactive functions of information annotations, the following measures can be taken: Augmented reality device feedback-driven iteration: Collect user feedback on the use of augmented reality devices, identify problems such as blurry display and interaction delay, and adjust annotation styles (such as font size and color contrast) and response logic accordingly.

[0062] VR Simulation and Scene Preview: In a digital twin environment, virtual reality (VR) technology is used to reproduce the augmented reality interaction process. By constructing simulated scenarios such as extreme weather and complex terrain, the applicability of different annotation layouts is tested.

[0063] In-depth analysis of user behavior data: Based on user behavior data, such as interaction frequency, operation duration, and number of misoperations, machine learning algorithms are used to uncover operation patterns. For example, quick trigger methods are set for high-frequency operations, and processes for long-running operations are simplified; for labeled functions with a large number of misoperations, the interaction logic and visual guidance are redesigned.

[0064] Furthermore, remote intelligent diagnostics and maintenance of the equipment can be performed regularly, using drones equipped with testing devices to inspect sensors, cameras, and other components to ensure the equipment operates normally.

[0065] In the preferred embodiments described above, the overall architecture of this invention is as follows: a quantum-encrypted communication sensor network (sensors for rainfall, water level, etc.) and intelligent adaptive image acquisition devices (high-definition cameras, drone cameras) collect multi-source data and transmit the data to the data processing center via a quantum-encrypted communication link; a large visual model trains, analyzes, and detects and warns of anomalies in the data, and transmits the results to the data processing center; augmented reality devices interact with the data processing center via a wireless network to achieve holographic information annotation and interaction; the data processing center uses a distributed blockchain database to store data and performs optimization feedback through a digital twin model, with all parts working together to achieve flood and typhoon prevention monitoring and early warning.

[0066] Application scenarios of this invention: In a flood-prone area in the middle and lower reaches of the Yangtze River, water level sensors and rainfall sensors integrating quantum encrypted communication are deployed at predetermined intervals (in meters) along the main river channel. High-definition cameras equipped with intelligent adaptive optics systems are installed at key embankment nodes. At the same time, five drones with autonomous path planning capabilities are deployed to plan refined inspection routes covering the entire area based on the region's topography, meteorological characteristics, and historical disaster data, using artificial intelligence algorithms.

[0067] A large-scale model CV analysis server cluster based on quantum computing acceleration was built at the local flood control command center, along with a distributed blockchain database and a digital twin simulation system. Thirty pairs of AR smart glasses integrating holographic projection technology were provided to on-site rescue personnel, and comprehensive debugging and integration of the equipment and systems were completed to ensure stable communication and accurate data transmission between all parts of the system.

[0068] Sensors collect environmental data at a high frequency of once per second, while cameras and drones dynamically adjust their collection frequency based on actual conditions to ensure the capture of critical information during disasters. This data is then transmitted in real-time to the data processing center via a quantum-encrypted communication link. A large-scale visual model, combined with preliminary analysis results from edge computing nodes, performs real-time, in-depth analysis of the image data. When the water level in a section of the river exceeds 90% of the dynamic warning level, a yellow alert is immediately generated, and the alert information, along with images of the water level change, is sent to AR smart glasses via quantum-encrypted communication.

[0069] On-site rescue personnel saw a prominent yellow holographic warning box appear at the corresponding river location through AR smart glasses. At the same time, the real-time water level value and the difference between the water level and the warning level were displayed in the form of a floating holographic window. The system uses augmented reality path planning algorithms to generate and display the optimal route to the area in real time in the augmented reality scene.

[0070] Guided by AR smart glasses, rescue personnel quickly arrived at the scene. Utilizing the glasses' various interactive functions, such as gesture controls and voice queries, they viewed information about potential hazards in the surrounding area and requested the command center to allocate rescue supplies such as sandbags and water pumps. The command center used a digital twin model to simulate and analyze the situation in real time, providing rescue personnel with more detailed decision-making suggestions.

[0071] After the incident is resolved, the user uploads the on-site handling results and images via AR smart glasses, and the system records the relevant data in a distributed blockchain database. The data management system conducts a comprehensive analysis and evaluation of this early warning event, using a digital twin model to simulate the effects of different handling schemes. New image data and simulated data are added to the large model training set, and model parameters are adjusted to further improve the accuracy of subsequent monitoring and early warning. Simultaneously, based on user feedback using the AR smart glasses, virtual reality technology is used in the digital twin environment to simulate and optimize information labeling and interaction processes, continuously improving the system's usability and functionality.

[0072] The advantages or beneficial effects of adopting the above technical solution are as follows: This invention employs quantum-enhanced computer vision large-scale model technology and real-time edge computing analysis to perform ultra-deep and precise analysis of multi-source data, accurately identifying various disaster characteristics. Compared with traditional methods, it significantly improves the monitoring accuracy of key information such as water level changes and flood inundation ranges, virtually eliminating missed or false alarms, and providing a solid and reliable data foundation for flood and typhoon prevention decision-making.

[0073] This invention leverages the high-speed transmission of quantum encrypted communication, the low-latency processing of edge computing, and the intelligent early warning generation mechanism. The system can quickly detect anomalies and issue early warnings the instant a disaster occurs, reducing the early warning response time to the second or even millisecond level. This greatly saves valuable time for emergency response, enabling relevant departments to take effective countermeasures at the first moment.

[0074] This invention employs a holographic augmented reality interactive approach to achieve immersive information annotation and interaction functions. It allows on-site users to feel as if they are in an intelligent command environment that blends digitalization and reality, enabling them to quickly and intuitively grasp the situation without relying on complex text reports and drawings. Combined with powerful interactive functions and intelligent navigation, the organization and execution of disaster relief operations become more efficient and orderly, significantly improving the efficiency and effectiveness of emergency response.

[0075] This invention employs quantum-encrypted communication to ensure data transmission security, and a distributed blockchain database to ensure data storage security, immutability, and traceability, providing comprehensive data security for flood and typhoon prevention work.

[0076] This invention utilizes a distributed blockchain database for data management and a digital twin-driven optimization feedback mechanism, enabling the system to continuously learn and improve from practical applications. The constantly updated model and optimized interactive functions allow it to better adapt to different regions and types of flood and typhoon prevention scenarios, maintaining high accuracy and reliability in monitoring and early warning over the long term. Furthermore, it continues to evolve with technological advancements, consistently maintaining leading performance.

[0077] The above are merely preferred embodiments of the present invention and are not intended to limit the implementation methods and protection scope of the present invention. Those skilled in the art should recognize that any equivalent substitutions and obvious changes made using the content of this specification and illustrations should be included within the protection scope of the present invention.

Claims

1. A flood and typhoon prevention monitoring and early warning method based on visual large models and augmented reality, characterized in that, include: Step S1: Collect multi-source data from key flood and typhoon prevention areas and transmit the multi-source data to the data processing center; Step S2: Analyze the multi-source data and perform anomaly detection and early warning using a pre-trained visual large model, obtain the large model analysis results, and transmit them to the data processing center. Step S3: The augmented reality device overlays the multi-source data from the data processing center and the analysis results of the large model into the real scene in the form of a holographic projection, and performs corresponding actions in the holographic projection according to the user's interactive operation.

2. The flood and typhoon monitoring and early warning method based on visual large model and augmented reality according to claim 1, characterized in that, The multi-source data includes environmental perception data, image and video data; step S1 includes: The environmental perception data is acquired using a first data acquisition module; wherein, the first data acquisition module includes a composite sensor network composed of several sensor nodes; The image and video data are acquired using a second data acquisition module. The second data acquisition module includes several first camera units fixedly installed at preset locations in key flood and typhoon prevention areas, and a second camera unit mounted on a drone. The drone is used to autonomously plan inspection paths and conduct inspections according to the inspection paths. The second camera unit is used to acquire the image and video data during the inspection process.

3. The flood and typhoon prevention monitoring and early warning method based on visual large model and augmented reality according to claim 2, characterized in that, The environmental sensing data includes one or more combinations of rainfall, water level, wind speed, wind direction, water body spectrum, and vibration signals.

4. The flood and typhoon prevention monitoring and early warning method based on visual large model and augmented reality according to claim 1, characterized in that, The multi-source data is transmitted to the data processing center via a quantum communication link.

5. The flood and typhoon prevention monitoring and early warning method based on visual large model and augmented reality according to claim 1, characterized in that, The training steps for the large visual model include: Collect relevant image data for flood and typhoon prevention, including river images at different water levels, images of building damage before and after typhoons, and images of flooded areas; A deep learning model accelerated by quantum computing is constructed and trained based on the flood and typhoon prevention related image data to identify disaster-related information; wherein, the disaster-related information includes one or more combinations of water level changes, flood inundation area, building damage degree, and typhoon eye location.

6. The flood and typhoon prevention monitoring and early warning method based on visual large model and augmented reality according to claim 1, characterized in that, Step S2 includes: Step S21A: The edge computing node processes and analyzes the image and video data in the multi-source data, extracts key features from the images, and transmits the key features to the visual big model. In step S22A, the visual big model identifies the key features to determine the boundaries of the flood-inundated area and monitor the trajectory of dangerous objects moving in the typhoon.

7. The flood and typhoon prevention monitoring and early warning method based on visual large model and augmented reality according to claim 1, characterized in that, Step S2 further includes: Step S21B: The visual big model analyzes the environmental perception data and image and video data in the multi-source data to obtain the analysis results; Step S22B: When the environmental perception data is detected to deviate from the preset threshold range or to exhibit abnormal characteristics, a warning message of the corresponding level is generated. Step S23B: Send the warning information, related images, and analysis results to the data processing center.

8. The flood and typhoon monitoring and early warning method based on visual large model and augmented reality according to claim 7, characterized in that, The steps for determining the preset threshold range include: Based on historical multi-source data, generative adversarial networks are used to generate monitoring data samples for non-flood and non-typhoon weather conditions. For various monitoring indicators, based on the monitoring data samples, preset threshold ranges that vary with time and season are set.

9. The flood and typhoon monitoring and early warning method based on visual large model and augmented reality according to claim 1, characterized in that, The user interaction is achieved through one or more combinations of gestures, voice, and eye tracking.

10. The flood and typhoon prevention monitoring and early warning method based on visual large model and augmented reality according to claim 1, characterized in that, The step S3 is followed by: Step S4: Construct a digital twin model of the flood and typhoon prevention area; Step S5: Perform virtual simulation and simulation verification based on user feedback information from the digital twin model and augmented reality device, optimize and update the large visual model, and perform information annotation and interaction optimization on the augmented reality device.

Images