A coastal disaster-bearing body identification method, device, equipment, medium and product

Abstract

The present application relates to the technical field of marine disaster warning, and discloses a coastal disaster-bearing body identification method, device, equipment, medium and product, which comprises the following steps: standardizing multi-source original data of a target coastal area to obtain standardized multi-source original data; based on a basic disaster-bearing body investigation task, a first visual model is used to classify disaster-bearing bodies according to the standardized multi-source original data, and a preliminary disaster-bearing body identification result is generated; based on a disaster damage assessment task, a second visual model is used to measure differences and quantify damage levels according to the standardized multi-source original data, and a preliminary disaster damage result is generated; the preliminary disaster-bearing body identification result and the preliminary disaster damage result are optimized by spatial constraint; and disaster-bearing body distribution data and disaster damage assessment data are integrated to obtain a coastal disaster-bearing body identification result. The present application solves the problems of insufficient scene adaptability and business adaptability, and provides data support for disaster prevention and mitigation decision-making.

Description

Technical Field

[0001] This invention relates to the field of marine disaster early warning technology, specifically to a method, device, equipment, medium, and product for identifying coastal disaster-bearing bodies. Background Technology

[0002] Coastal disaster-bearing bodies refer to various natural and man-made objects located in coastal areas that are susceptible to marine disasters. These objects may suffer varying degrees of damage when marine disasters (such as storm surges, waves, and tsunamis) occur. They are the key targets for marine disaster risk assessment and disaster prevention and mitigation. Therefore, accurate identification of coastal disaster-bearing bodies is a core prerequisite for marine disaster risk assessment and emergency decision-making.

[0003] Current methods for identifying disaster-bearing bodies along the coast lack adaptability to various scenarios and business operations, making it difficult for the identification results to directly support disaster prevention and mitigation decisions. Summary of the Invention

[0004] This invention provides a method, device, equipment, medium, and product for identifying coastal disaster-bearing bodies, in order to solve the problem that current methods for identifying coastal disaster-bearing bodies lack adaptability to various scenarios and business operations, making it difficult for the identification results to directly support disaster prevention and mitigation decisions.

[0005] In a first aspect, the present invention provides a method for identifying coastal disaster-bearing bodies, the method comprising:

[0006] Multi-source raw data of the target coastal area were collected and standardized to obtain standardized multi-source raw data. The standardized multi-source raw data includes standardized remote sensing image data, UAV video data and spatial auxiliary data. Obtain the target business demand distribution task, and based on the basic disaster-bearing body investigation task in the target business demand distribution task, classify the disaster-bearing bodies using the first vision model according to the standardized multi-source raw data, and generate preliminary disaster-bearing body identification results. Based on the disaster damage assessment task in the target business demand-based task, the second vision model is used to measure the difference and quantify the damage level according to the standardized multi-source raw data to generate preliminary disaster damage results. Spatial constraint optimization was performed on the preliminary disaster-bearing body identification results and preliminary disaster damage results to obtain disaster-bearing body distribution data and disaster damage assessment data; By integrating the data on the distribution of disaster-bearing bodies and the disaster damage assessment data, the results of coastal disaster-bearing body identification are obtained.

[0007] This invention provides a method for identifying coastal disaster-bearing bodies. It targets data sources such as remote sensing images and UAV videos of the target coastal area, and distributes tasks based on target business needs. A first visual model is used for disaster-bearing body classification, achieving automated classification and identification. A second visual model is used for difference measurement and damage level quantification, enabling comparative analysis of post-disaster damage to disaster-bearing bodies. Furthermore, spatial constraints are optimized for the preliminary disaster-bearing body identification results and preliminary disaster damage results, improving the spatial accuracy and business adaptability of coastal disaster-bearing body identification results. This solves the problem of insufficient scenario and business adaptability, providing accurate data support for disaster prevention and mitigation decision-making in coastal counties and districts.

[0008] Secondly, the present invention provides a coastal disaster-bearing body identification device, the device comprising: The data preprocessing module is used to collect multi-source raw data of the target coastal area, and to standardize the multi-source raw data to obtain standardized multi-source raw data. The standardized multi-source raw data includes standardized remote sensing image data, UAV video data, and spatial auxiliary data. The classification module is used to obtain the target business demand diversion task. Based on the basic disaster-bearing body investigation task in the target business demand diversion task, the module uses the first vision model to classify the disaster-bearing bodies according to the standardized multi-source raw data and generate preliminary disaster-bearing body identification results. The quantification module is used to perform disaster damage assessment tasks in the task distribution based on target business needs. Based on the standardized multi-source raw data, it uses a second vision model to measure differences and quantify damage levels to generate preliminary disaster damage results. The auxiliary optimization module is used to perform spatial constraint optimization on the preliminary disaster-bearing body identification results and preliminary disaster damage results to obtain disaster-bearing body distribution data and disaster damage assessment data. The integration module is used to integrate disaster-bearing body distribution data and disaster damage assessment data to obtain coastal disaster-bearing body identification results.

[0009] Thirdly, the present invention provides an electronic device, comprising: a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the coastal disaster-bearing body identification method of the first aspect or any corresponding embodiment described above.

[0010] Fourthly, the present invention provides a computer-readable storage medium storing computer instructions for causing a computer to execute the coastal disaster-bearing body identification method of the first aspect or any corresponding embodiment described above.

[0011] Fifthly, the present invention provides a computer program product, including computer instructions for causing a computer to execute the coastal disaster-bearing body identification method described in the first aspect or any corresponding embodiment thereof. Attached Figure Description

[0012] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0013] Figure 1 This is a schematic diagram of an application scenario according to an embodiment of the present invention; Figure 2 This is a flowchart illustrating the integrated identification method for coastal disaster-bearing bodies based on a large artificial intelligence model according to an embodiment of the present invention. Figure 3 This is a schematic diagram of the first process of a method for identifying coastal disaster-bearing bodies according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the second process of a coastal disaster-bearing body identification method according to an embodiment of the present invention; Figure 5 This is a schematic diagram of the third process of a coastal disaster-bearing body identification method according to an embodiment of the present invention; Figure 6 This is a schematic diagram of the fourth process of a coastal disaster-bearing body identification method according to an embodiment of the present invention; Figure 7 This is a structural block diagram of a coastal disaster-bearing body identification device according to an embodiment of the present invention; Figure 8 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0014] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, 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.

[0015] It is understood that before using the technical solutions disclosed in the various embodiments of the present invention, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in the present invention and their authorization should be obtained in accordance with relevant laws and regulations through appropriate means.

[0016] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0017] As an optional application scenario of this invention, such as Figure 1 As shown, the coastal disaster-bearing body identification system may include at least one terminal device and at least one server. Figure 1 The system is illustrated in the example, which includes a computer 101, a mobile terminal 102, and a server 103, and the terminal devices such as the computer 101 and the mobile terminal 102 are connected to the server 103 through a network 110.

[0018] Specifically, the terminal device can be a smartphone, tablet, laptop, PDA, desktop computer, game console, smart TV, smart wearable device, in-vehicle terminal, VR (Virtual Reality) device, AR (Augmented Reality) device, etc. Server 103 can be a standalone physical server, a server cluster, a distributed system, or a cloud server providing cloud services. Network 110 can be a wired or wireless network, examples of which include, but are not limited to, the Internet, corporate intranet, local area network, wide area network, mobile communication network, and combinations thereof.

[0019] Accurate identification of coastal disaster-bearing bodies has significant practical and strategic value: From an emergency management perspective, precise information on the distribution of disaster-bearing bodies can quickly delineate high-risk areas before marine disasters occur, providing data support for risk warning and pre-deployment of emergency resources; during the disaster response phase, accurate comparison of the damage to disaster-bearing bodies before and after a disaster can clarify the scope and extent of the disaster's impact, assisting in prioritizing emergency rescue efforts; from a long-term risk management perspective, continuous identification and dynamic updating of disaster-bearing bodies can build a comprehensive coastal disaster risk management database, providing a scientific basis for disaster risk assessment, prevention and control system construction, and post-disaster reconstruction planning; furthermore, automated and high-precision identification methods can significantly reduce investigation costs and improve work efficiency, promoting the transformation of coastal disaster risk management from experience-driven to data-driven, and enhancing the scientific rigor and timeliness of disaster prevention and mitigation decisions.

[0020] Currently, methods for identifying disaster-bearing bodies along the coast mainly fall into three categories: First, the traditional manual field survey method, which involves staff conducting on-site surveys and recording information about disaster-bearing bodies. This method relies heavily on human experience for accuracy but is extremely inefficient and costly, making it unsuitable for large-scale, dynamic surveys. Second, the semi-automatic identification method, which combines visual interpretation of remote sensing images with simple image processing techniques (such as threshold segmentation and edge detection). While this improves efficiency to some extent, it still requires significant manual intervention and has limited accuracy in identifying disaster-bearing bodies against complex backgrounds. Third, the artificial intelligence identification method, which uses deep learning models (such as CNN and Transformer) to achieve automated identification. Among these, ViT (Vision Transformer, a deep learning model that successfully migrates the Transformer architecture from natural language processing to computer vision) and Swin... Large-scale models such as Transformer (a type of visual model based on Transformer) have demonstrated advantages in remote sensing imagery and video target recognition due to their strong feature extraction capabilities. However, these models are mostly designed for general scenarios and lack specific adaptability to complex coastal marine environments (tides, fog, backlight) and various niche disaster-bearing bodies (such as drilling platforms). A mature end-to-end solution has not yet been developed. From an application perspective, the results of coastal disaster-bearing body identification mostly focus on a single disaster-bearing body type or a small area, lacking an integrated identification system that covers multiple types of disaster-bearing bodies and adapts to the needs of grassroots disaster prevention and mitigation operations.

[0021] Based on the current status of disaster-bearing body identification, and considering the characteristics of coastal areas and operational needs, the current identification of disaster-bearing bodies faces four main core problems: First, significant interference from complex backgrounds. Coastal remote sensing images / videos contain interference factors such as tidal fluctuations, cloud cover, and sea surface reflections, resulting in blurred boundaries between disaster-bearing bodies and the background, increasing the difficulty of identification. Second, uneven difficulty in identifying different types of disaster-bearing bodies. The boundaries of areal disaster-bearing bodies such as salt fields and aquaculture ponds are irregular, and the feature ratio of small target disaster-bearing bodies such as drilling platforms is low, making related models prone to missed detections and false detections. Third, insufficient accuracy in identifying post-disaster damage. After a disaster, disaster-bearing bodies may exhibit varying degrees of damage, such as minute cracks and local collapses. Comparison models struggle to comprehensively capture multi-scale damage features and cannot accurately quantify damage levels. Fourth, insufficient operational adaptability. Related identification methods focus primarily on improving model accuracy and do not fully integrate with operational scenarios such as GIS (Geographic Information System) spatial constraints and grassroots decision-making data needs, making it difficult for identification results to directly support disaster prevention and mitigation decisions.

[0022] Therefore, current methods for identifying coastal disaster-bearing bodies suffer from bottlenecks such as low efficiency of manual surveying, reliance on manual intervention for semi-automatic identification, and insufficient adaptability of general AI models to complex coastal scenarios (tides, clouds and fog) and niche disaster-bearing bodies.

[0023] This invention provides a method for identifying coastal disaster-bearing bodies. It employs ViT-L / 14 (a large-scale visual transformer model), Video Swin Transformer (a deep learning framework designed for video understanding tasks), and ViT-Siamese network, integrating GIS spatial constraints and multi-scale feature fusion technology. Through targeted improvements such as deformable convolution and adaptive correction to the marine background, it achieves automated identification of various types of disaster-bearing bodies, including salt fields and chemical plants, within a 50km radius along the coast, and accurate quantification of the damage to coastal dikes 3-5km after a disaster. This method constructs an end-to-end workflow of "data input - model recognition - spatial optimization - output," capable of outputting structured data and visualized results, significantly improving identification efficiency and accuracy (multi-category identification accuracy ≥90%). It provides precise data support for disaster prevention and mitigation decision-making in coastal counties and districts, and offers an efficient and accurate technical paradigm for identifying coastal disaster-bearing bodies, contributing to the digital transformation of disaster risk management.

[0024] Among them, such as Figure 2 As shown, the overall architecture of the coastal disaster-bearing body identification system includes a data acquisition layer, a data preprocessing layer, a task allocation and core processing layer, a GIS-assisted optimization layer, a result fusion and visualization layer, and an output layer. The data preprocessing layer includes a data enhancement module, a GIS data parsing module, and a spatial mask generation module. The task allocation and core processing layer includes a task allocation module, a core identification subsystem, and a disaster comparison subsystem. The core identification subsystem includes a feature extraction unit and a classification decision unit. The disaster comparison subsystem includes a shared feature extraction unit and a difference measurement and quantification unit. The GIS-assisted optimization layer includes a two-way feedback module, a spatial range filtering module, and a topological relationship verification module. The result fusion and visualization layer includes a structured data generation module, a visualization generation module, and a report generation module. The overall architecture of the coastal disaster-bearing body identification system can be integrated into the aforementioned terminal equipment.

[0025] The coastal disaster-bearing body identification system follows an end-to-end processing flow of "data input → preprocessing → core processing → optimization and screening → fusion output," achieving full-process automation based on business needs: First, the data acquisition layer aggregates raw data from multiple sources, and the data preprocessing layer completes standardization and preliminary preparation (including spatial mask generation); the task allocation module distributes data to the corresponding core subsystems for feature extraction and decision-making based on business needs such as "basic disaster-bearing body investigation" or "post-disaster damage comparison"; the core processing results undergo spatial constraints and verification through the GIS auxiliary layer, and the preprocessing stage is optimized through two-way feedback; finally, the result fusion and visualization layer generates structured data, visualization results, and professional reports, providing output to support disaster prevention and mitigation decision-making.

[0026] The coastal disaster-bearing body identification system adopts a hierarchical architecture of "layered serial connection + module parallelism," with each layer vertically connected and modules at the same level collaborating horizontally: Vertical hierarchical connection: data acquisition layer → data preprocessing layer → task allocation and core processing layer → GIS-assisted optimization layer → result fusion and visualization layer → output layer. Data flows seamlessly between layers through standardized data interfaces, with the output of one layer becoming the input of the next. Horizontal module collaboration: Within the data preprocessing layer, the GIS data parsing module and the image / video preprocessing module work in parallel, ultimately converging to the spatial mask generation module; Within the core processing layer, the basic identification and disaster comparison subsystems are deployed in parallel and invoked as needed; Within the result fusion layer, the structured data, visualization, and report generation modules output in parallel, improving efficiency.

[0027] The entire data flow in the coastal disaster-bearing body identification system forms a closed-loop path of "one-way main line + two-way feedback". The core data flow is as follows: Main line data flow: multi-source raw data (remote sensing / video / GIS) → GIS data parsing to generate spatial range information → image / video preprocessing + data augmentation to obtain standardized data → task splitting under spatial mask constraints → core subsystem extracts features and completes classification / difference quantification → GIS spatial screening and topology verification → fusion to generate structured data / visualization results / professional reports → output application; Two-way feedback flow: the topology verification results of the GIS auxiliary optimization layer (such as preprocessing clipping deviation, insufficient mask accuracy) are fed back to the data preprocessing layer to optimize the clipping range and spatial mask generation parameters, achieving iterative improvement of accuracy throughout the entire process.

[0028] The core logic of the coastal disaster-bearing body identification system revolves around "business demand-driven + deep integration of AI and GIS + accurate and efficient identification." Key logical connections are as follows: Demand-Task Mapping Logic: Business demands (basic survey / disaster assessment) directly drive task allocation. The basic survey calls the core identification subsystem to classify multiple categories of disaster-bearing bodies, while the disaster assessment calls the disaster comparison subsystem to quantify pre- and post-disaster differences and determine damage levels. Spatial Constraint Throughout Logic: From mask generation in the preprocessing stage to range constraints in the core processing, and finally to topological verification of the final results, GIS spatial information is consistently used to ensure that the identification results conform to the "50km coastal..." The business boundary requirement is "3-5km disaster impact zone"; collaborative optimization logic: the AI ​​model is responsible for feature extraction and accurate identification, while GIS is responsible for spatial positioning and range selection. The two complement each other through two-way feedback, while the data enhancement module improves the generalization ability of the AI ​​model to ensure the recognition accuracy in complex coastal scenarios (tidal / cloud / fog / motion blur); decision support logic: the final output of structured data, visualization results and professional reports directly corresponds to the core needs of coastal counties and districts in disaster-bearing body distribution survey and disaster damage assessment, providing accurate data support for key attention area delineation, emergency rescue priority division and post-disaster reconstruction planning.

[0029] In summary, the overall system architecture of the coastal disaster-bearing body identification system follows a workflow logic of "data input → processing → optimization → output," dividing the system into six core layers. Each layer has clearly defined responsibilities and collaborates effectively, breaking through the separation model of model recognition + manual post-processing. It deeply embeds business requirements (basic investigation / disaster assessment) into each layer of the system, constructing a fully automated architecture encompassing data input, preprocessing, task allocation, core identification, spatial optimization, and result output. The task allocation module dynamically binds business requirements to the core model, the GIS-assisted optimization layer implements spatial constraints and two-way feedback, and the result fusion layer directly outputs structured data and decision reports adapted to the GIS system. This completes the entire chain of transformation from raw data to decision support without manual intervention, adapting to the actual business scenarios of grassroots disaster prevention and mitigation. The integrity and business adaptability of the technical architecture are innovative.

[0030] According to an embodiment of the present invention, a method for identifying coastal disaster-bearing bodies is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0031] This embodiment provides a method for identifying coastal disaster-bearing bodies, which can be used in the aforementioned terminal equipment. Figure 3 This is a flowchart of a coastal disaster-bearing body identification method according to an embodiment of the present invention, such as... Figure 3 As shown, the process includes the following steps: Step S301: Collect multi-source raw data of the target coastal area, and perform standardization processing on the multi-source raw data to obtain standardized multi-source raw data; wherein, the standardized multi-source raw data includes standardized remote sensing image data, UAV video data and spatial auxiliary data.

[0032] Specifically, remote sensing images (GF-2 / ZY-3, etc.) and UAV videos (with GPS positioning) of the target coastal area that meet the accuracy requirements are collected; spatial auxiliary data of the target coastal area is obtained, and the spatial auxiliary data adopts GIS auxiliary data, which includes coastal line vectors, county and district administrative boundaries, elevation, tide data, etc.

[0033] Furthermore, the data formats and basic information (such as image resolution, video frame rate, and geographic coordinate integrity) of remote sensing images, UAV videos, and spatial auxiliary data are initially verified. Based on the verification results, the above data are then optimized to obtain a multi-source raw dataset (including remote sensing images, UAV videos, and spatial auxiliary data).

[0034] Furthermore, the raw data is standardized to generate input data adapted to the core model, while spatial constraint preparation is completed. The main steps include: remote sensing image / UAV video preprocessing (cropping, normalization, atmospheric correction, frame extraction and registration); data augmentation (improving sample diversity through flipping, rotation, etc.); GIS-aided data parsing and spatial mask generation (delineating a 50km coastal area and a 3-5km disaster impact area mask), thereby obtaining standardized remote sensing image / UAV video data, augmented training data, and parsed GIS basic data (i.e., standardized spatial auxiliary data).

[0035] The enhanced training data includes: enhanced remote sensing image samples: image data of coastal remote sensing images (GF-2, ZY-3, etc.) cropped to 224×224 pixels, after random flipping, 0-90° rotation, 0.8-1.2x scaling, and Gaussian noise addition; enhanced UAV video frame samples: video frame data of UAV videos after extracting one frame every three frames and performing inter-frame registration, and then performing the same enhancement strategy as above; and accompanying annotation files: label files corresponding one-to-one with the enhanced samples, and converting the file format from VOC (an audio file format) to COCO (CommonObjects in Context, a standard dataset format widely used in computer vision tasks). The annotation content includes the category and location information of six types of disaster-bearing bodies (salt fields, aquaculture ponds, chemical plants, oil refineries, drilling platforms, and dams) and background. Rare categories (such as drilling platforms) have been additionally sampled.

[0036] Furthermore, the parsed GIS basic data includes: Format-standardized geographic data: a standardized geographic dataset after format parsing and unification of the original SHP (a geographic information system vector data format) and GeoJSON (a format for encoding various geographic data structures) coastal vector data, county and district administrative boundary data, elevation data, and tide gauge data; Spatial constraint mask data: two types of spatial masks generated based on the parsed geographic data, namely, a basic identification mask within a 50-kilometer radius along the coast and a contrast mask for the disaster-affected area within a 3-5 kilometer radius of the coastline, used to limit the geographic range for model recognition.

[0037] Furthermore, considering only geographical constraints without combining multi-source geographical data such as elevation and tide level, the identification results cannot screen disaster-bearing bodies that are vulnerable to marine disasters, making it difficult to directly support disaster prevention and mitigation decisions, thus resulting in insufficient business adaptability.

[0038] Furthermore, since the GIS and AI fusion only uses a single geographic boundary constraint and does not integrate business-related data such as elevation and tide levels, the identification results contain a large number of low-risk disaster-bearing bodies and have a lot of redundant information. Therefore, a dual constraint mechanism of geographic range + disaster sensitivity is adopted to integrate multi-source geographic data such as coastal line vector, elevation (low-lying areas <5m), and tide level (historical highest tide level); generate accurate spatial masks to guide the model to focus only on the identification of disaster-bearing bodies in high-risk areas, reduce invalid calculations, and improve the matching degree between the identification results and disaster risk management needs, solving the problems of poor business adaptability and a lot of redundant information. The above steps are applied to the spatial mask generation module of the data preprocessing layer and the spatial range filtering module of the GIS-assisted optimization layer to delineate the effective identification area of ​​the 50km coastal range / 3-5km disaster-affected area.

[0039] Furthermore, a spatial mask with dual constraints is generated by integrating geographical range, elevation, and tidal data. The spatial mask is represented as follows: (1) In the above formula, This is a spatial mask, where 1 represents a valid identification area and 0 represents an invalid area. Geographic coordinates; The preset geographical range is (50km coastal area / 3-5km disaster impact area). This is an area with elevation constraints. , for Elevation of the location Elevation threshold; The area affected by tide levels. , for The tide level at that location, This is the highest climax in history.

[0040] Step S302: Obtain the target business demand diversion task. Based on the basic disaster-bearing body investigation task in the target business demand diversion task, classify the disaster-bearing bodies using the first vision model according to the standardized multi-source raw data, and generate preliminary disaster-bearing body identification results.

[0041] Specifically, such as Figure 2 As shown, the target business requirement diversion tasks include basic disaster-bearing body investigation tasks and disaster damage assessment tasks. Based on the basic disaster-bearing body investigation tasks and disaster damage assessment tasks, the standardized multi-source raw data is diverted to the core identification subsystem (corresponding to the basic disaster-bearing body investigation task) and the disaster comparison subsystem (corresponding to the disaster damage assessment task).

[0042] Furthermore, considering the data source characteristics (remote sensing images / UAV videos), task requirements (classification and identification + post-disaster comparison) and accuracy requirements for coastal disaster-bearing body identification, the core focus is on fine-tuning a pre-trained large model, taking into account both identification efficiency and generalization ability. The pre-trained large model consists of ViT-L / 14 and Vision Transformer-Large pre-models.

[0043] Furthermore, the first-vision model can use the ViT-L / 14+Video Swin Transformer fusion model to extract features and complete the classification of disaster-bearing bodies. For remote sensing images, the ViT-L / 14 model is used, and for UAV videos, the VideoSwin Transformer model is used.

[0044] Step S303: Based on the disaster damage assessment task in the target business demand diversion task, according to the standardized multi-source raw data, the second vision model is used to measure the difference and quantify the damage level to generate preliminary disaster damage results.

[0045] Specifically, the second vision model can adopt the ViT+Siamese network model, which extracts shared features through the ViT+Siamese network to complete the difference measurement and damage level quantification.

[0046] Step S304: Spatial constraint optimization is performed on the preliminary disaster-bearing body identification results and preliminary disaster damage results to obtain disaster-bearing body distribution data and disaster damage assessment data.

[0047] Specifically, spatial constraints are used to optimize the identification results, eliminate invalid data, and simultaneously achieve bidirectional feedback iteration with the preprocessing layer.

[0048] Furthermore, the preliminary disaster-bearing body identification results and preliminary disaster damage results are processed in the GIS-assisted optimization layer through three core steps: spatial range filtering, topological relationship verification, and bidirectional feedback iteration. The final output is a valid result that conforms to business boundaries and spatial logic, which in turn feeds back into the optimization preprocessing stage. After processing by the GIS-assisted optimization layer, two types of results are output: first, optimized valid identification results, including disaster-bearing body distribution data and damage assessment data that conform to range constraints and spatial logic; second, feedback optimization parameters, which are used to iteratively improve the processing quality of the data preprocessing layer.

[0049] Step S305: Integrate the disaster-bearing body distribution data and disaster damage assessment data to obtain the coastal disaster-bearing body identification results.

[0050] Specifically, the optimized and effective identification results (including disaster-bearing body distribution data and disaster damage assessment data) are integrated to generate structured data and visualization results, and a professional report is prepared.

[0051] This embodiment provides a coastal disaster-bearing body identification method. Leveraging the strong feature extraction and generalization capabilities of large-scale artificial intelligence models, it achieves automated classification and identification of designated disaster-bearing bodies within a 50-kilometer coastal area, targeting data sources such as remote sensing images and UAV videos. Simultaneously, it supports comparative analysis of damage to disaster-bearing bodies (primarily dikes) within a 3-5 kilometer coastal zone in the same region after disasters (such as strong winds). This aims to address the insufficient scenario and business adaptability of existing coastal disaster-bearing body identification methods, providing accurate data support for disaster prevention and mitigation decision-making in coastal counties. Furthermore, considering the unique characteristics of coastal scenarios, it innovatively constructs a multi-dimensional optimization algorithm system—background adaptation, target adaptation, difference adaptation, and spatial adaptation—to address the model's insufficient adaptability from four core dimensions: environment, target, difference, and space. The multi-category identification accuracy is ≥90%, and the post-disaster damage level quantification error is ≤5%.

[0052] This embodiment provides a method for identifying coastal disaster-bearing bodies, which can be used in the aforementioned terminal equipment. Figure 4 This is a flowchart of a coastal disaster-bearing body identification method according to an embodiment of the present invention, such as... Figure 4 As shown, the process includes the following steps: Step S401: Collect multi-source raw data of the target coastal area, and perform standardization processing on the multi-source raw data to obtain standardized multi-source raw data; wherein, the standardized multi-source raw data includes standardized remote sensing image data, UAV video data, and spatial auxiliary data. For details, please refer to... Figure 3 Step S301 of the illustrated embodiment will not be described again here.

[0053] Step S402: Obtain the target business demand diversion task. Based on the basic disaster-bearing body investigation task in the target business demand diversion task, classify the disaster-bearing bodies using the first vision model according to the standardized multi-source raw data, and generate preliminary disaster-bearing body identification results.

[0054] Specifically, the input data preprocessing layer outputs standardized remote sensing image data and UAV video data, while simultaneously loading a spatial mask covering a 50-kilometer coastal area to define the geographical boundaries for model recognition.

[0055] In some optional implementations, step S402 above includes: Step S4021: Extract features from the standardized remote sensing image data to obtain image features.

[0056] Specifically, the ViT-L / 14 model was used to extract features from the standardized remote sensing image data.

[0057] In some optional implementations, step S4021 above includes: Step a1: Perform dual-scale feature coding on the standardized remote sensing image data to obtain dual-scale coded features.

[0058] Specifically, the problem of uneven difficulty in identifying different types of disaster-bearing bodies is addressed. Small target disaster-bearing bodies such as drilling platforms have a low proportion of features, while areal disaster-bearing bodies such as salt fields / aquaculture ponds have irregular boundaries. Traditional single-patch size encoding cannot take both types of targets into account, and is prone to missed detections and false detections. In this embodiment, parallel encoding of 16×16 and 32×32 dual-scale patches (image blocks) is performed on the standardized remote sensing image data to transform the remote sensing image into sequential features, focusing on strengthening the feature representation of small target disaster-bearing bodies such as drilling platforms.

[0059] Furthermore, the criteria for distinguishing between small targets and disaster-bearing bodies with irregular boundaries are based on their spatial scale, morphological characteristics, and pixel ratio in remote sensing images, combined with the differences between actual geographical attributes and marine background.

[0060] Furthermore, the core characteristics of small-target disaster-bearing bodies are low pixel ratio, small physical scale, and point-like or small compact structure, such as drilling platforms. The specific identification dimensions of small-target disaster-bearing bodies are as follows: 1) Spatial scale and pixel ratio: In remote sensing images with a resolution of ≥0.5m, the target pixel ratio is much lower than that of isometric disaster-bearing bodies such as salt fields and aquaculture ponds. Their pixel coverage is small, and single-scale patch encoding is prone to feature loss.

[0061] 2) The target in terms of morphology and structure is point-like or small compact area, with a simple structure and no extended boundary. Against the background of the ocean, there is a certain contrast between the target and the background, but the feature area is concentrated and small, and is easily obscured by interference factors such as tides and clouds.

[0062] 3) Geographical attribute characteristics: The targets are mostly offshore or near-shore structures with specific geographic coordinates. The targets are usually far from the land and located in the sea area. Combining GIS land and sea distribution data can help in the identification.

[0063] Furthermore, the core characteristics of irregular boundary disaster-bearing bodies are planar extension, no fixed geometric shape at the boundary, and blurred or tortuous edges. Typical examples include salt fields and aquaculture ponds. The specific identification dimensions of irregular boundary disaster-bearing bodies are as follows: 1) Boundary morphology features: targets have irregular geometric outlines, with blurred, tortuous edges or changes with terrain and tides; for example, the internal structure of the grid in salt fields is regular, but the external boundary is irregular with the shoreline; the water boundary of aquaculture ponds is affected by the rise and fall of tides, and there is a dynamic blur zone.

[0064] 2) Spatial distribution and texture features: The target extends in a planar manner, with identifiable texture patterns inside, but the overall boundary has no uniform shape; for example, the grid-like texture of salt fields and the water reflection texture of aquaculture ponds can be used as internal features, while the external boundary has no fixed reference.

[0065] 3) The background correlation feature boundary has low distinguishability from the ocean background, and the edge pixels are easily confused with interference factors such as tides and clouds; the pixel gradient change of the target boundary is small, and the fixed sampling method of the relevant convolution kernel is difficult to accurately capture edge features.

[0066] Furthermore, for the aforementioned small target disaster-bearing bodies and irregular boundary disaster-bearing bodies, a dual-scale parallel coding of 16×16 (focusing on small target details) and 32×32 (focusing on planar target outlines) is adopted.

[0067] Step a2 involves adaptively assigning weights to the dual-scale encoded features to obtain the fused encoded features.

[0068] Specifically, for small targets and irregularly boundaryed disaster-bearing bodies, features are extracted and fused using 16×16 and 32×32 dual-scale patch encoding. The calculation formula for the fused features is shown below: (2) (3) (4) In the above formula, The fused feature map Feature maps encoded as 16×16 patches (emphasizing details of small targets). Feature map encoded for 32×32 patch (focusing on the overall outline of planar targets). and It is an adaptive weight (dynamically adjusted according to the type of disaster-bearing body). This represents the pixel percentage of small disaster-bearing targets in remote sensing images. This represents the percentage of pixels representing isomorphic disaster-bearing bodies in remote sensing images.

[0069] Step a3: Use deformable convolution to optimize the fused encoded features to obtain optimized image features.

[0070] Specifically, the feature extraction unit of the core identification subsystem (ViT-L / 14) works in collaboration with the deformable convolution module to adapt to the feature extraction needs of various types of disaster-bearing bodies in remote sensing images; among them, the edge features of irregular boundary disaster-bearing bodies such as salt fields and aquaculture ponds are captured by adaptively adjusting the sampling position of the convolution kernel through deformable convolution.

[0071] Furthermore, by using dual-scale patch encoding and deformable convolution, target adaptation is achieved, which simultaneously adapts to small targets and irregular boundary disaster-bearing bodies.

[0072] Furthermore, the fused feature map In the deformable convolution module, three core steps are used to process the data: dimension adaptation, adaptive sampling adjustment, and feature aggregation enhancement. The final output is an optimized feature map, which works in conjunction with dual-scale patch encoding to accurately capture the irregular boundaries and small target features of coastal disaster-bearing bodies.

[0073] Furthermore, the fused feature map is processed using a 1×1 convolution kernel. Channel compression / expansion is performed to unify the number of feature channels to the input specification of deformable convolutional modules (e.g., 512 channels), while preserving the core semantic information of the features; the fused feature map is then processed through interpolation / downsampling. The spatial resolution is adjusted to match the size of the dual-scale patch encoding, for example, adjusted to 224×224 / 448×448, to adapt to 16×16 / 32×32. The process involves: Patch partitioning; based on the dimension-adapted feature map, a lightweight convolutional branch predicts the sampling offset field, which contains the x / y offset values ​​of each convolutional kernel sampling point; considering the characteristics of coastal disaster-bearing bodies, for irregular boundary areas such as salt fields and aquaculture ponds, the offset is adjusted towards the boundary contour direction to make the convolutional kernel sampling points fit the boundary; for small target areas such as drilling platforms, the offset is clustered towards the target center to increase the sampling density of small target areas; deformable convolutional kernels adjust the sampling point positions according to the predicted offset to obtain an adaptively sampled feature map, in which the irregular boundary features of the disaster-bearing body are completely preserved and the signal-to-noise ratio of small target features is significantly improved; the adaptively sampled features are aggregated with the features encoded by dual-scale patches, fusing local detail features (small targets) and global spatial features (irregular boundaries) to form a more robust optimized feature map, resulting in optimized image features.

[0074] Step a4: Perform adaptive correction of the ocean background on the optimized image features to obtain the image features.

[0075] Specifically, in response to the problem of significant interference from complex backgrounds, namely the blurring of the boundary between the disaster-bearing body and the background caused by tidal fluctuations, cloud cover, and sea surface reflection in coastal remote sensing images, and the inability of traditional models to effectively suppress marine background noise, resulting in a decrease in recognition accuracy, this embodiment uses adaptive correction of marine background to suppress background noise such as tides and clouds, thereby enhancing the distinction between the disaster-bearing body and the background.

[0076] Furthermore, the feature extraction unit of the core identification subsystem (ViT-L / 14) performs adaptive correction of the marine background on the optimized image features. Specifically, it is used in the feature extraction stage of remote sensing images such as GF-2 / ZY-3, and completes background noise suppression and disaster-bearing body feature enhancement after dual-scale feature encoding and deformable convolution.

[0077] Furthermore, through adaptive correction of the marine background, complex disturbances such as tides and clouds are dynamically suppressed, achieving background adaptation.

[0078] Furthermore, to address atmospheric scattering interference in coastal remote sensing images, brightness distortion caused by clouds, fog, and backlighting is eliminated using an atmospheric correction factor. The formula for calculating the atmospheric correction factor is shown below: (5) In the above formula, This is the atmospheric correction factor, with a value ranging from 0.1 to 0.9; This represents the average brightness value of the ocean region in the remotely sensed image. This is the dark channel value of the image (the ideal dark pixel brightness with no reflected light). Reference brightness values ​​for clear coastal areas (obtained from standard radiometric calibration data of GF-2 / ZY-3 satellite imagery).

[0079] Furthermore, ocean background noise is dynamically suppressed through a spatial attention mechanism to enhance the characteristics of the disaster-bearing body. The calculation formula for the corrected characteristics is shown below: (6) (7) In the above formula, This is the corrected feature map (i.e., image features). This is the original feature map after dual-scale feature encoding in the ViT-L / 14 model. This is a spatial attention mask, with values ​​ranging from 0 to 1. The closer the value is to 1, the higher the probability that the location is a disaster-prone area. The spatial coordinates of the original feature map. It is the Sigmoid activation function. The size of the original feature map. The standard deviation of brightness in the ocean region.

[0080] In the above optional embodiments, since the ViT model uses fixed-size (e.g., 16×16) patch encoding, small targets are easily ignored, and the feature extraction of targets with irregular boundaries is incomplete. Furthermore, it does not consider the dynamic weight allocation for different types of disaster-bearing bodies. Therefore, a dual-scale parallel encoding method using 16×16 (emphasizing small target details) and 32×32 (emphasizing areal target outlines) is adopted to simultaneously cover the feature requirements of both small targets and areal targets with irregular boundaries. An adaptive weight allocation mechanism is designed to dynamically adjust the fusion weights based on the pixel proportions of small targets and areal targets in the image, avoiding the "one-sidedness" of single-scale encoding and improving multi-scale performance. The accuracy of identifying disaster-bearing bodies is improved. In addition, since the ViT model only relies on the original feature extraction and does not have a dedicated correction mechanism for the marine background, it mostly uses fixed threshold denoising or global atmospheric correction, which cannot dynamically adapt to the differences in clouds, fog and reflectivity in different regions. Therefore, through the dual mechanism of atmospheric correction factor and spatial attention mask, the correction factor is dynamically calculated by statistically analyzing the brightness of the marine area to eliminate atmospheric scattering interference. An attention mask is generated based on brightness differences to accurately distinguish disaster-bearing bodies from the marine background, dynamically suppress background noise, and strengthen the feature representation of targets with blurred boundaries, thus solving the pain points of inaccurate denoising and easy feature loss.

[0081] Step S4022: Extract features from the standardized UAV video data to obtain video frame features.

[0082] Specifically, the Video Swin Transformer model is used to extract features from the standardized UAV video data. In particular, to address the problem of uneven difficulty in identifying various types of disaster-bearing bodies due to complex background interference in video stream scenarios, namely, UAV flight jitter and airflow interference causing motion blur in video frames, the traditional fixed spatiotemporal window cannot adapt to dynamic blur changes, resulting in missed or false detections of linear / area-shaped disaster-bearing bodies (dams, aquaculture ponds) and low processing efficiency, this optional embodiment performs dynamic spatiotemporal window adjustment to achieve feature extraction from UAV video data.

[0083] Furthermore, the spatiotemporal feature extraction unit of the core recognition subsystem extracts features from the standardized UAV video data to meet the real-time recognition requirements of UAV video (>15fps, with GPS positioning).

[0084] In some optional implementations, step S4022 above includes: Step b1: Perform inter-frame difference calculation on the standardized UAV video data to obtain the degree of motion blur.

[0085] Specifically, the motion blur level of the drone video is calculated using inter-frame difference. The formula for calculating the motion blur level is as follows: (8) In the above formula, This represents the degree of motion blur, with a value ranging from 0 to 255. A higher value indicates more severe blurring. Where is the size of a single frame in the drone video, and T is the length of the video frame sequence. and For the first Frame and the Frame of video image, This is the row index of the video frame pixels. The column index for the video frame pixels. and Together they determine the specific pixel positions within a single frame of image.

[0086] Step b2: Adaptively adjust the spatiotemporal window based on the degree of motion blur to obtain a dynamic spatiotemporal window.

[0087] Specifically, the spatiotemporal window is adaptively adjusted according to the degree of motion blur, and the size of the dynamic spatiotemporal window can be expressed as: (9) In the above formula, The size of the spatiotemporal window (the first two dimensions are the spatial window, and the third dimension is the temporal window). and As an empirical threshold, , .

[0088] Furthermore, the window step size of the dynamic spatiotemporal window It can be represented as: (10) Step b3: Use a dynamic spatiotemporal window to sample the standardized UAV video data to obtain video frame features.

[0089] Specifically, the dynamic window size and stride are feature sampling parameters in the spatiotemporal attention mechanism of the Video Swin Transformer model, acting on the feature extraction stage within the model. The core function of the window stride is to control the sampling interval of spatiotemporal features, adapting in conjunction with the window size to the degree of motion blur. When the motion blur is low, the drone footage is stable, and the features of the disaster-bearing body are clear. In this case, reducing the window size and setting a small stride focuses on local details, improving recognition accuracy. Conversely, when the motion blur is high, drone shaking causes blurry images and discontinuous disaster-bearing body features. In this case, increasing the window size and setting a large stride expands the feature sampling range, ensuring the continuity of disaster-bearing body features and avoiding missed detections.

[0090] In the above optional embodiments, since the Video Swin Transformer uses a fixed-size spatiotemporal window, the continuity of features is broken when the motion blur is severe, and the detail extraction is insufficient when the blur is small. Furthermore, the dynamic adaptability of the video stream is not considered. Therefore, based on the motion blur quantization method of inter-frame difference, the quality of video frames is dynamically evaluated, and a linkage adjustment mechanism of blur degree - window size = step size is designed. When the blur is small, the window is reduced to improve the detail accuracy, and when the blur is large, the window is increased to ensure the continuity of features. Then, through adaptive optimization of the window step size, the video processing frame rate is increased from 15fps to more than 25fps while improving the recognition accuracy, thus solving the pain point of not being able to balance accuracy and efficiency.

[0091] Step S4023: Deep semantic feature extraction is performed on image features and video frame features to obtain global spatial correlation semantic features of the disaster-bearing body.

[0092] Specifically, the optimized features (i.e., image features and video frame features) are input into the Transformer Encoder layer, and the global spatial association semantic features of the disaster-bearing body are extracted through a multi-head attention mechanism and a feedforward network.

[0093] Step S4024: Perform weighted cross-entropy classification on the global spatial association semantic features of the disaster-bearing body to obtain preliminary disaster-bearing body identification results; wherein, the preliminary disaster-bearing body identification results include disaster-bearing body category, disaster-bearing body location and confidence level.

[0094] Specifically, the global spatial association semantic features of the disaster-bearing body are processed by a fully connected layer and a Softmax activation function to classify 6 types of disaster-bearing bodies plus the background class, and output preliminary disaster-bearing body identification results, including disaster-bearing body category, disaster-bearing body location and confidence level.

[0095] Step S403: Based on the disaster damage assessment task in the target business demand-based task allocation, and using the standardized multi-source raw data, a second visual model is employed to measure differences and quantify damage levels, generating preliminary disaster damage results. For details, please refer to [link to details]. Figure 3 Step S303 of the illustrated embodiment will not be described again here.

[0096] Step S404 involves spatial constraint optimization of the preliminary disaster-bearing body identification results and preliminary disaster damage results to obtain disaster-bearing body distribution data and disaster damage assessment data. For details, please refer to [link to relevant documentation]. Figure 3 Step S304 of the illustrated embodiment will not be described again here.

[0097] Step S405 integrates the disaster-bearing body distribution data and disaster damage assessment data to obtain the coastal disaster-bearing body identification results. For details, please refer to [link to relevant documentation]. Figure 3 Step S305 of the illustrated embodiment will not be described again here.

[0098] Since the ViT series models are not specifically adapted for complex coastal backgrounds (tides, clouds, fog, sea surface reflections), they rely only on general pre-training data and lack a mechanism to suppress marine background noise. The related deformable convolutions are only applied to general target recognition and do not combine dual-scale patch encoding to solve the dual pain points of small coastal disaster-bearing bodies (low feature ratio of small targets + irregular boundaries of areal disaster-bearing bodies). Therefore, this embodiment provides a coastal disaster-bearing body recognition method, which proposes a three-pronged improvement scheme of dual-scale patch encoding + deformable convolution + marine background adaptive correction. The three are deeply integrated into the feature extraction link of ViT-L / 14. The model parameters are optimized through a pre-training task specific to coastal scenes, and it can simultaneously adapt to the recognition needs of small targets and disaster-bearing bodies with irregular boundaries, significantly improving the recognition accuracy in complex marine environments.

[0099] Furthermore, because Video Swin Transformer uses a fixed-size spatiotemporal window for feature extraction, it does not consider the dynamic changes in motion blur caused by flight jitter and airflow interference in coastal scenes. At the same time, it lacks a target trajectory constraint mechanism, which is prone to missed or false detections of linear / area-shaped disaster-bearing bodies (such as dikes and aquaculture ponds) due to camera shake. Therefore, the coastal disaster-bearing body identification method provided in this embodiment is based on a dynamic spatiotemporal window adjustment mechanism of inter-frame difference, combined with target trajectory constraints of Kalman filtering, to achieve a dynamic balance between the accuracy and efficiency of video stream processing. In the field of coastal disaster-bearing body video recognition, this dynamic adaptation logic can increase the frame rate of drone video recognition from 15fps to more than 25fps, while reducing the missed detection rate of small targets by 40%.

[0100] This embodiment provides a method for identifying coastal disaster-bearing bodies, which can be used in the aforementioned terminal equipment. Figure 5 This is a flowchart of a coastal disaster-bearing body identification method according to an embodiment of the present invention, such as... Figure 5 As shown, the process includes the following steps: Step S501: Collect multi-source raw data of the target coastal area, and perform standardization processing on the multi-source raw data to obtain standardized multi-source raw data; wherein, the standardized multi-source raw data includes standardized remote sensing image data, UAV video data, and spatial auxiliary data. For details, please refer to... Figure 4 Step S401 of the illustrated embodiment will not be described again here.

[0101] Step S502: Obtain the target business requirement allocation task. Based on the basic disaster-bearing body investigation task within the target business requirement allocation task, and using the standardized multi-source raw data, classify the disaster-bearing bodies using a first-vision model to generate preliminary disaster-bearing body identification results. For details, please refer to [link to relevant documentation]. Figure 4 Step S402 of the illustrated embodiment will not be described again here.

[0102] Step S503: Based on the disaster damage assessment task in the target business demand diversion task, according to the standardized multi-source raw data, the second vision model is used to measure the difference and quantify the damage level to generate preliminary disaster damage results.

[0103] Specifically, the ViT+ Siamese network achieves feature alignment and difference capture of pre- and post-disaster images by sharing a feature extraction backbone.

[0104] In some optional implementations, step S503 above includes: Step S5031: Based on the standardized multi-source raw data, extract the pre-disaster standardized image data and post-disaster standardized image data of the same region.

[0105] Specifically, standardized pre-disaster and post-disaster image data of the same area are extracted from the standardized remote sensing images to ensure that the shooting range and resolution of the two are consistent, and a spatial mask of 3-5 kilometers of disaster-affected area is loaded.

[0106] Step S5032: Extract features from the pre-disaster standardized image data and the post-disaster standardized image data respectively to obtain the basic feature vectors of the pre-disaster images and the post-disaster images.

[0107] Specifically, pre-disaster standardized image data and post-disaster standardized image data are input into two branches of the Siamese network, respectively. The two branches share the same set of ViT-L / 14 pre-trained weights and extract the basic feature vectors of the images to ensure the consistency of feature extraction.

[0108] Step S5033: Perform multi-scale feature weighted fusion on the basic feature vectors of the pre-disaster image and the post-disaster image respectively to obtain the fused pre-disaster feature vector and the fused post-disaster feature vector.

[0109] Specifically, addressing the issue of insufficient accuracy in identifying post-disaster damage—that is, the relevant models only use single-scale features and cannot comprehensively capture multi-level damage from "minor cracks to partial collapse to overall destruction," and the problem that fixed-weight fusion cannot adapt to the differences in damage patterns of different disaster types (strong winds, heavy rain)—this optional embodiment extracts shallow texture features (corresponding to minor cracks), mid-level structural features (corresponding to partial collapse), and deep semantic features (corresponding to overall integrity) from the basic features output by the two branches through a multi-scale feature fusion module, and performs weighted fusion. The fused pre-disaster and post-disaster feature vectors are further processed by the feature encoding layer of a Siamese network to achieve accurate alignment of pre-disaster and post-disaster features, providing a basis for subsequent difference measurement.

[0110] Furthermore, the difference measurement and quantification unit applied to the disaster comparison subsystem is the core formula for calculating the difference between pre-disaster and post-disaster images, and is suitable for the damage comparison needs of disaster-bearing bodies such as embankments in the 3-5km disaster-affected area.

[0111] Furthermore, multi-scale feature weighting fusion is performed on the basic feature vectors of pre-disaster images and post-disaster images respectively, which can be expressed as: (11) (12) In the above formula, and This refers to the multi-scale features resulting from the fusion of pre-disaster and post-disaster features (i.e., the fused pre-disaster feature vector and the fused post-disaster feature vector). Scale weights ( , , (determined through cross-validation) and The first two days before and after the disaster, respectively. Scale features, among which, Time indicates shallow texture features. Time indicates the characteristics of the middle layer structure. Time represents deep semantic features, the first and second times before and after the disaster. Scale features are multi-granular features extracted from different layers of the Transformer Encoder through the ViT-L / 14 backbone shared by the Siamese network.

[0112] Step S5034: Calculate the fusion difference degree based on the fused pre-disaster feature vector and the fused post-disaster feature vector.

[0113] Specifically, the formula for calculating the fusion difference is as follows: (13) (14) In the above formula, To determine the degree of difference, the value ranges from 0 to 1, with a larger value indicating more severe damage. For cosine similarity, It is a structural similarity index. Image / image patch of the same area before the disaster. For post-disaster images / image patches of the same area, The weights are dynamic (ranging from 0.4 to 0.8). Standard deviation of damage characteristics corresponding to disaster type (e.g., strong wind disaster) Rainstorm disaster ), This represents the maximum value of the standard deviation of the damage characteristics.

[0114] Among them, by matching disaster types with a damage feature database, calculation In the disaster comparison subsystem of task allocation and core processing layer, the model calls a preset damage feature library based on the verified disaster type. Different disaster types correspond to specific damage features: for example, strong wind disasters correspond to macroscopic features such as structural collapse and large crack widths, while rainstorm disasters correspond to microscopic texture features such as dam leakage and surface erosion. Based on the corresponding features, the model extracts multi-scale feature difference values ​​from pre-disaster and post-disaster images and calculates the standard deviation of damage features under that disaster type. This is used for subsequent differential measurement and damage level calibration.

[0115] Furthermore, since the relevant difference measurement model uses a single deep semantic feature, it can only identify explicit damage (such as collapse) and cannot capture implicit damage (such as fine cracks). Moreover, the fusion weights of cosine similarity and SSIM are fixed, resulting in poor adaptability. Therefore, by constructing a multi-scale feature fusion system of shallow texture + mid-level structure + deep semantics, corresponding to different levels of damage features, we can achieve full coverage of all types of damage. Through a dynamic weight adjustment mechanism, the fusion weights (α∈0.4~0.8) are adaptively optimized according to the standard deviation of the damage features corresponding to the disaster type, thereby improving the adaptability to different disaster types and solving the problems of incomplete damage identification and poor adaptability.

[0116] Furthermore, by fusing multi-scale features and dynamically adjusting weights, it adapts to different degrees of damage and disaster types, achieving differential adaptiveness.

[0117] Step S5035: Map the fusion difference degree to a damage level score, and determine the preliminary disaster damage result based on the damage level score; wherein, the preliminary disaster damage result includes the damaged area and the disaster damage level.

[0118] Specifically, addressing the extended problem of insufficient accuracy in identifying damage after a disaster, where relevant models can only output binary classification results of damaged / undamaged, unable to accurately quantify damage levels and support business needs such as emergency rescue priority division, this optional embodiment completes the quantitative output of damage levels after difference measurement, directly providing data support for disaster damage assessment reports, and is applied to the damage level quantification unit of the disaster comparison subsystem.

[0119] Furthermore, the fusion difference is expressed using the Sigmoid function. Mapped to a damage rating of 0-1: (15) In the above formula, The injury level is scored, with 0 indicating no injury and 1 indicating severe injury. The slope parameter (controls the steepness of the scoring curve); For threshold parameters (difference greater than) (determined to be damaged) This is the Sigmoid activation function.

[0120] Furthermore, the criteria for classifying damage levels are as follows: Minor damage (tiny cracks); Moderate damage (local breakage); Severe damage (collapse / complete destruction).

[0121] Furthermore, since the relevant models lack a damage level calibration mechanism, there is no clear mapping relationship between the degree of difference and the actual degree of damage, and the output results cannot be directly used for business decision-making. Therefore, a calibration model is constructed based on actual damaged samples (slight cracks, moderate damage, and severe collapse). The degree of difference is mapped to a quantitative score of 0 to 1 through the Sigmoid function. Through scientific level classification thresholds, the damage level can be accurately defined, solving the pain point of being able to only qualitatively identify damage but not quantitatively, and making the identification results directly adapt to the business needs of disaster assessment.

[0122] Step S504 involves spatial constraint optimization of the preliminary disaster-bearing body identification results and preliminary disaster damage results to obtain disaster-bearing body distribution data and disaster damage assessment data. For details, please refer to [link to relevant documentation]. Figure 4 Step S404 of the illustrated embodiment will not be described again here.

[0123] Step S505 integrates the disaster-bearing body distribution data and disaster damage assessment data to obtain the coastal disaster-bearing body identification results. For details, please refer to [link to relevant documentation]. Figure 4 Step S405 of the illustrated embodiment will not be described again here.

[0124] Because relevant disaster comparison models only use single-scale features (such as deep semantic features) for difference measurement, they cannot comprehensively capture multi-level damage features such as minute cracks, local collapses, and overall destruction. Moreover, the difference calculation uses fixed weight fusion (such as α=0.6), which does not consider the differences in damage patterns caused by different disaster types (strong winds, rainstorms, storm surges), and lacks a quantitative calibration mechanism for damage levels, it can only output a binary classification result of damaged / undamaged. Therefore, this embodiment provides a coastal disaster-bearing body identification method, which constructs a three-stage disaster comparison model of multi-scale feature fusion + dynamic weight difference measurement + damage level quantitative calibration. It covers all levels of damage through weighted fusion of shallow, middle and deep features, dynamically adjusts the fusion weights of cosine similarity and SSIM to adapt to different disaster types, and finally outputs a quantitative damage level of 0 to 1 through model calibration. This dual-modal fusion and quantification scheme solves the core pain points of incomplete model identification, poor adaptability, and inability to quantify.

[0125] This embodiment provides a method for identifying coastal disaster-bearing bodies, which can be used in the aforementioned terminal equipment. Figure 6 This is a flowchart of a coastal disaster-bearing body identification method according to an embodiment of the present invention, such as... Figure 6 As shown, the process includes the following steps: Step S601: Collect multi-source raw data of the target coastal area, and standardize the multi-source raw data to obtain standardized multi-source raw data; wherein, the standardized multi-source raw data includes standardized remote sensing image data, UAV video data, and spatial auxiliary data. For details, please refer to... Figure 5 Step S501 of the illustrated embodiment will not be described again here.

[0126] Step S602: Obtain the target business requirement allocation task. Based on the basic disaster-bearing body investigation task within the target business requirement allocation task, and using the standardized multi-source raw data, classify the disaster-bearing bodies using a first-vision model to generate preliminary disaster-bearing body identification results. For details, please refer to [link to relevant documentation]. Figure 5 Step S502 of the illustrated embodiment will not be described again here.

[0127] Step S603: Based on the disaster damage assessment task in the target business demand-based task allocation, and using the standardized multi-source raw data, a second visual model is employed to measure differences and quantify damage levels, generating preliminary disaster damage results. For details, please refer to [link to details]. Figure 5 Step S503 of the illustrated embodiment will not be described again here.

[0128] Step S604: Spatial constraint optimization is performed on the preliminary disaster-bearing body identification results and preliminary disaster damage results to obtain disaster-bearing body distribution data and disaster damage assessment data.

[0129] Specifically, step S604 includes: Step S6041: Using the standardized spatial auxiliary data, the spatial range of the preliminary disaster-bearing body identification results and the preliminary disaster damage results is filtered to obtain candidate disaster-bearing body data.

[0130] Specifically, by utilizing the geographical range constraints in the spatial constraint mask data, valid identification results within a 50-kilometer coastal area are selected, and interfering targets exceeding the boundary are eliminated. That is, identification results exceeding the boundary are filtered according to the preset geographical range. In the spatial range filtering module of the GIS-assisted optimization layer, the spatial constraint mask data can be obtained using the above step S304.

[0131] Furthermore, using the preliminary disaster-bearing body identification results (including the coordinates of the disaster-bearing body's center point, category, etc.) and the coastal line vector and disaster-affected area data after GIS parsing as input, data filtering is carried out based on the geographical range constraint formula d(P,L)≤R: First, the geographical coordinates P of the center point of each disaster-bearing body are extracted, and the shortest straight-line distance from it to the coastal line vector L is calculated; then, according to the business type, the constraint radius R is matched (50km for basic identification and 3-5km for disaster comparison), and disaster-bearing body data whose distance meets the threshold is retained, while invalid identification results that exceed the geographical range are directly eliminated, resulting in a disaster-bearing body candidate dataset with compliant range.

[0132] Step S6042: Based on the standardized spatial auxiliary data, perform topological relationship verification on the candidate disaster-bearing body data to obtain disaster-bearing body distribution data and disaster damage assessment data.

[0133] Specifically, topology verification includes eliminating results with location conflicts and attribute contradictions, such as drilling platforms in land areas.

[0134] Furthermore, using the candidate dataset of disaster-bearing bodies after spatial range filtering and GIS multi-source geographic data (coastal line, county and district administrative boundaries, elevation, tide vector, etc.) as input, spatial topological overlay analysis and attribute consistency verification are carried out: First, the spatial coordinates of the disaster-bearing bodies are topologically matched with the GIS geographic feature layer to verify the rationality of the location of the disaster-bearing bodies and the geographic scene (e.g., drilling platforms need to be located in the sea area, and dikes need to be distributed along the coastline), and results with location conflicts are eliminated; then, a second verification is carried out by combining the disaster-bearing body category and geographic attributes to eliminate invalid data with attribute contradictions (e.g., low-lying non-coastal areas are labeled as aquaculture ponds); finally, optimized and effective identification results with compliant spatial location and attributes are obtained, and preprocessing deviations found during the verification process (e.g., clipping range errors) are extracted to form feedback parameters.

[0135] Furthermore, addressing the closed-loop optimization problem of insufficient business adaptability and recognition accuracy—namely, the unidirectional input relationship between GIS and AI recognition, where deviations in the clipping range and mask accuracy during the preprocessing stage cannot be corrected by subsequent recognition results, making it difficult to iteratively improve the accuracy throughout the entire process—the bidirectional feedback module of the GIS-assisted optimization layer feeds back the topology verification results to the preprocessing layer, optimizes the clipping range and mask accuracy, and outputs the optimized effective recognition results and feedback optimization parameters. The bidirectional feedback module is the core module connecting the data preprocessing layer and the GIS-assisted optimization layer, enabling iterative improvement of accuracy throughout the entire process.

[0136] Furthermore, the relevant system is a unidirectional pipeline architecture with fixed preprocessing parameters. There is no feedback linkage between the recognition result and the preprocessing stage, and problems such as cropping deviation and insufficient mask accuracy cannot be dynamically corrected. Therefore, to solve the above problems, a closed-loop optimization mechanism of preprocessing-recognition-verification-feedback is constructed. With the GIS topology verification result as a constraint, the cropping range deviation loss and mask accuracy loss are defined. By minimizing the total loss function, the preprocessing parameters (cropping range coordinates, mask generation threshold) are optimized to achieve dynamic iteration of accuracy throughout the process, solving the pain points of fixed parameters and inability to continuously optimize accuracy.

[0137] Furthermore, using the GIS topology verification results as constraints, the preprocessing clipping range and mask accuracy are optimized: (16) (17) (18) In the above formula, To optimize parameters (including cropping range coordinates and mask generation threshold). For the total loss function, To account for the loss due to the deviation in the cutting range, To compensate for the loss of mask accuracy, and To lose weight, , , To identify the target quantity, For the first The scope of the target The first one after topological relationship verification The true range of the target For generating a true spatial mask for GIS, The size of the space mask.

[0138] in, The corresponding identification targets are six types of core coastal disaster-bearing bodies. The acquisition of the target range needs to be achieved through the entire process of model feature extraction + boundary positioning + GIS spatial constraints. The six types of core coastal disaster-bearing bodies include areal disaster-bearing bodies: salt fields and aquaculture ponds; point-like / compact areal disaster-bearing bodies: chemical plants, oil refineries and drilling platforms; and linear disaster-bearing bodies: dikes.

[0139] Further, by solving the above formula (16), the optimal cropping range coordinates and the optimal mask generation threshold are obtained. The optimal cropping range coordinates and the optimal mask generation threshold are output to the data preprocessing layer to reprocess the input multi-source raw data.

[0140] Furthermore, by combining GIS two-way feedback with multi-source geographic data fusion, a precise match between spatial constraints and business needs is achieved, thus realizing spatial self-adaptation.

[0141] Step S605: Integrate the disaster-bearing body distribution data and disaster damage assessment data to obtain the coastal disaster-bearing body identification results.

[0142] Specifically, information such as the type, location, scale, and damage level of disaster-bearing bodies is collected to generate structured data; based on the optimized and effective identification results, visualization results are produced, including distribution heat maps, pre- and post-disaster comparison maps, and vector layers; professional reports are prepared: a disaster-bearing body investigation report and a disaster damage assessment report are written; in summary, a structured dataset, visualization result files, and a draft of the professional report are obtained; among them, the disaster-bearing body investigation report focuses on the distribution investigation of multiple types of disaster-bearing bodies within a 50-kilometer radius of the coast, and the core data is divided into three categories: basic information, spatial distribution, and accuracy verification; the disaster damage assessment report focuses on the damage assessment of disaster-bearing bodies in the disaster-affected area 3-5 kilometers from the coast, and the core data is divided into three categories: damage information, pre- and post-disaster comparison, and decision support.

[0143] Furthermore, the output layer outputs the final deliverables according to business needs, ensuring the usability and adaptability of the deliverables. Specific steps include: exporting structured data in CSV / JSON format for GIS system import; displaying visualized deliverables (providing interactive heatmaps and vector layer displays); outputting professional reports (outputting the final report to support disaster prevention and mitigation decision-making); and finally outputting importable structured data, interactive visualized deliverables, and the final professional decision-making report.

[0144] Since the integration of GIS and AI recognition is mostly a one-way input mode (GIS only provides spatial range information to AI, and the AI ​​recognition results do not feed back to GIS for optimization), and GIS only provides geographic boundary constraints without integrating multi-source geographic data such as elevation and tide level for priority screening of disaster-bearing bodies, the matching degree between the recognition results and disaster assessment needs is insufficient. Therefore, this embodiment provides a coastal disaster-bearing body identification method that constructs a two-way feedback closed loop between GIS and AI recognition. The GIS topology verification results are fed back to the AI ​​preprocessing parameter optimization. At the same time, elevation and tide level data are integrated to generate a spatial mask with dual constraints of geographic range and disaster sensitivity. This integration mechanism is novel in the field of coastal disaster-bearing body identification and can significantly improve the spatial accuracy and business adaptability of the identification results.

[0145] Furthermore, the coastal disaster-bearing body identification method provided in this embodiment breaks through the limitation of models that prioritize accuracy over application, directly linking the identification results with disaster prevention and mitigation decision-making needs: the basic identification results output structured information including category, location, scale, and disaster sensitivity, supporting the delineation of high-risk areas; the disaster comparison results output quantitative damage levels + visual comparison maps, supporting the prioritization of rescue efforts; and the integration of elevation and tide data to mark disaster-bearing bodies in vulnerable areas supports the pre-deployment of emergency resources. This application model promotes the transformation of disaster risk management from experience-driven to data-driven, shortening emergency decision-making response time by 30%, and providing a new technological paradigm for disaster prevention and mitigation in coastal counties and districts.

[0146] This embodiment also provides a coastal disaster-bearing body identification device, which is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0147] This embodiment provides a coastal disaster-bearing body identification device, such as... Figure 7 As shown, it includes: The data preprocessing module 701 is used to collect multi-source raw data of the target coastal area, and to perform standardization processing on the multi-source raw data to obtain standardized multi-source raw data; wherein, the standardized multi-source raw data includes standardized remote sensing image data, UAV video data and spatial auxiliary data. Classification module 702 is used to obtain target business demand diversion tasks, and based on the basic disaster-bearing body investigation task in the target business demand diversion tasks, it uses the first vision model to classify disaster-bearing bodies according to the standardized multi-source raw data and generates preliminary disaster-bearing body identification results. The quantification module 703 is used for disaster damage assessment tasks in the task diversion based on target business needs. Based on the standardized multi-source raw data, it uses a second visual model to measure differences and quantify damage levels to generate preliminary disaster damage results. The auxiliary optimization module 704 is used to perform spatial constraint optimization on the preliminary disaster-bearing body identification results and preliminary disaster damage results to obtain disaster-bearing body distribution data and disaster damage assessment data. The integration module 705 is used to integrate the disaster-bearing body distribution data and disaster damage assessment data to obtain the coastal disaster-bearing body identification results.

[0148] In some alternative implementations, the classification module 702 includes: The first feature extraction unit is used to extract features from the standardized remote sensing image data to obtain image features; The second feature extraction unit is used to extract features from the standardized UAV video data to obtain video frame features. The third feature extraction unit is used to perform deep semantic feature extraction on image features and video frame features to obtain the global spatial correlation semantic features of the disaster-bearing body. The classification unit is used to perform weighted cross-entropy classification on the global spatial association semantic features of the disaster-bearing body to obtain preliminary disaster-bearing body identification results; among which, the preliminary disaster-bearing body identification results include disaster-bearing body category, disaster-bearing body location and confidence level.

[0149] In some optional implementations, the first feature extraction unit includes: The coding subunit is used to perform dual-scale feature coding on the standardized remote sensing image data to obtain dual-scale coded features. The allocation subunit is used to adaptively allocate weights to the dual-scale encoded features to obtain the fused encoded features. The optimization subunit is used to optimize the fused encoded features using deformable convolution to obtain optimized image features. The correction subunit is used to perform adaptive correction of the marine background on the optimized image features to obtain the image features.

[0150] In some optional implementations, the second feature extraction unit includes: The computational subunit is used to perform inter-frame difference calculation on the standardized UAV video data to obtain the degree of motion blur; The adjustment subunit is used to adaptively adjust the spatiotemporal window based on the degree of motion blur, thus obtaining a dynamic spatiotemporal window; The sampling subunit is used to sample the standardized UAV video data using a dynamic spatiotemporal window to obtain video frame features.

[0151] In some alternative implementations, the quantization module 703 includes: The fourth feature extraction unit is used to extract pre-disaster standardized image data and post-disaster standardized image data of the same region based on the standardized multi-source raw data. The fifth feature extraction unit is used to extract features from pre-disaster standardized image data and post-disaster standardized image data respectively, to obtain the basic feature vectors of pre-disaster images and post-disaster images. The weighted fusion unit is used to perform multi-scale feature weighted fusion on the basic feature vectors of pre-disaster images and post-disaster images respectively, to obtain the fused pre-disaster feature vector and the fused post-disaster feature vector. The computing unit is used to calculate the fusion difference degree based on the fused pre-disaster feature vector and the fused post-disaster feature vector; The scoring unit is used to map the fusion difference degree to a damage level score and determine the preliminary disaster damage result based on the damage level score; wherein, the preliminary disaster damage result includes the damaged area and the disaster damage level.

[0152] In some alternative implementations, the auxiliary optimization module 704 includes: The spatial range filtering unit is used to filter the spatial range of the preliminary disaster-bearing body identification results and the preliminary disaster damage results using standardized spatial auxiliary data to obtain candidate disaster-bearing body data. The topology verification unit is used to verify the topology of candidate disaster-bearing bodies based on standardized spatial auxiliary data, so as to obtain disaster-bearing body distribution data and disaster damage assessment data.

[0153] The coastal disaster-bearing body identification device provided in this embodiment of the invention can execute the coastal disaster-bearing body identification method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the method. Further functional descriptions of the above modules and units are the same as those in the corresponding embodiments, and will not be repeated here.

[0154] Figure 8 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention.

[0155] The following is a detailed reference. Figure 8 This diagram illustrates a suitable structural schematic for implementing an electronic device according to embodiments of the present invention. The electronic device may include a processor (e.g., a central processing unit, graphics processor, etc.) 801, which can perform various appropriate actions and processes based on a program stored in read-only memory (ROM) 802 or a program loaded from memory 808 into random access memory (RAM) 803. The RAM 803 also stores various programs and data required for the operation of the electronic device. The processor 801, ROM 802, and RAM 803 are interconnected via a bus 804. An input / output (I / O) interface 805 is also connected to the bus 804.

[0156] Typically, the following devices can be connected to I / O interface 805: input devices 806 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 807 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; memory devices 808 including, for example, magnetic tapes, hard disks, etc.; and communication devices 809. Communication device 809 allows electronic devices to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 8Electronic devices with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown, and more or fewer devices may be implemented or have instead.

[0157] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 809, or installed from a memory 808, or installed from a ROM 802. When the computer program is executed by a processor 801, it performs the functions defined in the coastal disaster-bearing body identification method of the embodiments of the present invention.

[0158] Figure 8 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0159] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, it implements the coastal disaster-bearing body identification method shown in the above embodiments.

[0160] A portion of this invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the invention through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0161] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A method for identifying coastal disaster-bearing bodies, characterized in that, The method includes: Collect multi-source raw data of the target coastal area, and perform standardization processing on the multi-source raw data to obtain standardized multi-source raw data; wherein, the standardized multi-source raw data includes standardized remote sensing image data, UAV video data and spatial auxiliary data; Obtain the target business demand diversion task, and based on the basic disaster-bearing body investigation task in the target business demand diversion task, classify the disaster-bearing bodies using the first visual model according to the standardized multi-source raw data, and generate preliminary disaster-bearing body identification results. Based on the disaster damage assessment task in the target business demand diversion task, according to the standardized multi-source raw data, the second visual model is used to measure the difference and quantify the damage level to generate preliminary disaster damage results. Spatial constraint optimization is performed on the preliminary disaster-bearing body identification results and the preliminary disaster damage results to obtain disaster-bearing body distribution data and disaster damage assessment data; The disaster-bearing body distribution data and the disaster damage assessment data are integrated to obtain the coastal disaster-bearing body identification results.

2. The method according to claim 1, characterized in that, The basic disaster-bearing body investigation task in the target business demand-based traffic diversion task, based on the standardized multi-source raw data, uses a first visual model to classify disaster-bearing bodies and generate preliminary disaster-bearing body identification results, including: Feature extraction is performed on the standardized remote sensing image data to obtain image features; Feature extraction is performed on the standardized UAV video data to obtain video frame features; Deep semantic feature extraction is performed on the image features and video frame features to obtain the global spatial correlation semantic features of the disaster-bearing body; The global spatial association semantic features of the disaster-bearing body are subjected to weighted cross-entropy classification to obtain the preliminary disaster-bearing body identification result; wherein, the preliminary disaster-bearing body identification result includes the disaster-bearing body category, disaster-bearing body location and confidence level.

3. The method according to claim 2, characterized in that, The process of extracting features from standardized remote sensing image data to obtain image features includes: The standardized remote sensing image data is subjected to dual-scale feature encoding to obtain dual-scale encoded features; Adaptive weight allocation is performed on the dual-scale encoded features to obtain the fused encoded features; The fused encoded features are optimized using deformable convolution to obtain optimized image features. The optimized image features are subjected to adaptive correction to the marine background to obtain the image features.

4. The method according to claim 2, characterized in that, The step of extracting features from the standardized UAV video data to obtain video frame features includes: The motion blur level is obtained by performing inter-frame difference calculation on the standardized UAV video data; The spatiotemporal window is adaptively adjusted based on the degree of motion blur to obtain a dynamic spatiotemporal window; The standardized UAV video data is sampled using the dynamic spatiotemporal window to obtain the video frame features.

5. The method according to claim 1, characterized in that, The disaster damage assessment task in the target business demand-based traffic diversion task, based on the standardized multi-source raw data, uses a second visual model to measure differences and quantify damage levels, generating preliminary disaster damage results, including: Based on the standardized multi-source raw data, pre-disaster standardized image data and post-disaster standardized image data of the same region are extracted. Feature extraction is performed on the pre-disaster standardized image data and the post-disaster standardized image data respectively to obtain the basic feature vectors of the pre-disaster image and the post-disaster image. Multi-scale feature weighted fusion is performed on the basic feature vectors of the pre-disaster image and the basic feature vectors of the post-disaster image to obtain the fused pre-disaster feature vector and the fused post-disaster feature vector. The fusion difference degree is calculated based on the fused pre-disaster feature vector and the fused post-disaster feature vector; The fusion difference is mapped to a damage level score, and the preliminary disaster damage result is determined based on the damage level score; wherein the preliminary disaster damage result includes the damaged area and the disaster damage level.

6. The method according to claim 1, characterized in that, The spatial constraint optimization of the preliminary disaster-bearing body identification results and the preliminary disaster damage results yields disaster-bearing body distribution data and disaster damage assessment data, including: The spatial range of the preliminary disaster-bearing body identification results and the preliminary disaster damage results is filtered using standardized spatial auxiliary data to obtain candidate disaster-bearing body data. Based on the standardized spatial auxiliary data, the topological relationship of the candidate disaster-bearing body data is verified to obtain the disaster-bearing body distribution data and the disaster damage assessment data.

7. A coastal disaster-bearing body identification device, characterized in that, The device includes: The data preprocessing module is used to collect multi-source raw data of the target coastal area, and to standardize the multi-source raw data to obtain standardized multi-source raw data; wherein, the standardized multi-source raw data includes standardized remote sensing image data, UAV video data and spatial auxiliary data. The classification module is used to obtain the target business demand diversion task, and based on the basic disaster-bearing body investigation task in the target business demand diversion task, it uses the first visual model to classify the disaster-bearing bodies according to the standardized multi-source raw data, and generates preliminary disaster-bearing body identification results. The quantification module is used to perform the disaster damage assessment task in the task diverted based on the target business needs. Based on the standardized multi-source raw data, it uses a second visual model to measure the difference and quantify the damage level to generate preliminary disaster damage results. An auxiliary optimization module is used to perform spatial constraint optimization on the preliminary disaster-bearing body identification results and the preliminary disaster damage results to obtain disaster-bearing body distribution data and disaster damage assessment data. The integration module is used to integrate the disaster-bearing body distribution data and the disaster damage assessment data to obtain the coastal disaster-bearing body identification results.

8. An electronic device, characterized in that, include: A memory and a processor are interconnected, the memory stores computer instructions, and the processor executes the coastal disaster-bearing body identification method according to any one of claims 1 to 6 by executing the computer instructions.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to execute the coastal disaster-bearing body identification method according to any one of claims 1 to 6.

10. A computer program product, characterized in that, It includes computer instructions for causing a computer to execute the coastal disaster-bearing body identification method according to any one of claims 1 to 6.

Images