Disaster monitoring and evaluating system integrating multi-source and heterogeneous remote sensing and collaborative analysis

Abstract

The invention provides a disaster monitoring and evaluating system integrating multi-source and heterogeneous remote sensing and collaborative analysis, and belongs to the technical field of remote sensing. The system comprises a data layer, an application layer and a user layer. The data layer receives and stores image data of various disasters in a folder manner; the application layer comprises a sentinel service module and an earthquake, flood and fire assessment service module; and the sentry service module monitors the image data of each folder to enter a directory, and when capturing that a new image file is created in the folder, the sentry service module calls the corresponding assessment service module to perform disaster assessment according to the task corresponding to the folder, receives the analysis feedback of each assessment service module, and sends a disaster assessment result to the user layer. By using the intelligent early warning platform, the limitation of severe weather conditions can be broken through, all-weather and all-time monitoring of disasters such as floods, earthquakes and fire disasters can be realized, passive analysis is upgraded to an intelligent early warning platform of active discovery and real-time alarm, and the efficiency and precision of emergency response are improved.

Description

Technical Field

[0001] This invention belongs to the field of remote sensing technology, specifically relating to a disaster monitoring and assessment system that integrates multi-source heterogeneous remote sensing and collaborative analysis. Background Technology

[0002] Currently, remote sensing monitoring of natural disasters generally adopts a "one disaster, one policy; one source, one application" technical model. This means that for a specific type of disaster, it relies on a specific single data source—primarily optical remote sensing—and is processed by an independent system. While these systems play a role in their respective fields, they are generally characterized by fragmented technical means and rigid data silos. If multi-hazard monitoring and assessment is desired, multiple specialized systems are required, involving professional technicians and executing complex workflows in stages: first, various types of data, such as optical and SAR data, are manually downloaded from different data centers; second, different specialized software, such as ENVI, SNAP, and ArcGIS, are used to process different disasters independently and semi-automatically; finally, the scattered analysis results are manually compiled into a report. This entire process is fragmented, passive, and inefficient, failing to meet the needs of modern comprehensive emergency management.

[0003] Moreover, existing disaster monitoring and assessment systems have significant technical shortcomings:

[0004] Fire monitoring and assessment are divided into in-flight fire detection and post-flashover assessment; among which:

[0005] Fire detection: Existing operational systems primarily rely on thermal anomaly detection algorithms from low- to medium-resolution satellites, such as the MODIS sensor aboard the US Terra / Aqua satellite and the VIIRS sensor aboard the Suomi-NPP / JPSS satellite. These algorithms identify fire points by detecting pixel brightness and temperature anomalies in the shortwave infrared and thermal infrared bands. However, these technologies have inherent drawbacks: low spatial resolution, making it difficult to detect early, small fire points; long revisit periods, hindering high-frequency continuous monitoring; and extreme sensitivity to cloud cover, with monitoring capabilities significantly reduced or even completely failing in cloudy weather.

[0006] Fire zone assessment: After a fire is extinguished, medium- to high-resolution optical imagery such as Landsat or Sentinel-2 is typically used to assess the severity of the fire by calculating the difference in normalized burn index (NBR) before and after the fire (dNBR). However, in practice, this method usually relies on remote sensing professionals adjusting thresholds in GIS software and performing extensive manual interactive mapping to correct patch boundaries and remove noise. The entire process has low automation, is time-consuming and labor-intensive, and the accuracy of the results is highly dependent on the operator's experience.

[0007] The core of flood disaster monitoring and assessment technology is to quickly and accurately obtain the inundation range; among which:

[0008] Limitations of optical image monitoring: While high-resolution optical images can clearly identify water body boundaries, their application faces an insurmountable obstacle: flood disasters are often accompanied by continuous rainy weather. Thick cloud cover prevents optical sensors from acquiring effective surface images, rendering the technology completely ineffective at the critical moment when disaster information is most needed.

[0009] Current Status of SAR Image Monitoring: Synthetic Aperture Radar (SAR) technology, due to its ability to penetrate clouds and fog, is widely recognized as an effective way to solve this problem. Existing solutions extract water bodies by comparing changes in backscattering coefficients between pre-disaster and post-disaster SAR images. However, most existing SAR flood analyses remain in a semi-automated stage, requiring manual setting of complex segmentation thresholds. The processing flow is highly specialized and difficult to achieve rapid operational response.

[0010] Earthquake hazard assessment focuses on the damage to buildings after the earthquake; among which:

[0011] Visual interpretation: In the critical 72 hours of post-earthquake emergency rescue, the most reliable and mainstream method is to wait for the acquisition of the highest resolution optical or aerial images, and then organize a large number of professional image interpretation experts to conduct a visual inspection of each building and determine the level of damage. Although this method is highly accurate, its drawback is fatal: it is extremely time-consuming and cannot avoid interference from clouds, smoke, and dust.

[0012] Bottlenecks in Heterogeneous Data Fusion: To address the aforementioned issues, the academic community has been exploring the fusion of pre-disaster optical imagery and post-disaster SAR imagery for change detection. However, because their imaging mechanisms are completely different—one being sunlight reflected from ground objects, and the other being backscattering of actively emitted radar waves from ground objects—pixel values ​​are not directly comparable. Therefore, how to conduct automated, high-precision comparative analysis remains a recognized technical challenge in the industry, and a mature and reliable operational application solution has yet to emerge.

[0013] Based on the assessment schemes for each disaster and the comprehensive monitoring and assessment schemes mentioned above, it can be seen that existing disaster remote sensing monitoring technologies generally have the following four core defects:

[0014] 1. Limited monitoring capabilities and shortcomings in all-weather technology:

[0015] Current technologies heavily rely on optical remote sensing, and their imaging principles render them completely ineffective at night and under adverse weather conditions such as clouds, rain, fog, smoke, and dust. For disasters like floods and earthquakes, which are often accompanied by severe weather, this deficiency is fatal, preventing emergency response departments from obtaining images of disaster areas at the most critical moments, creating an "information black hole" for decision-making. Therefore, current technologies have a fundamental deficiency in achieving continuous, all-weather, all-time monitoring capabilities.

[0016] 2. Data fusion is difficult, and there are technical bottlenecks in heterogeneous collaboration:

[0017] Current technologies typically use remote sensing data from different sources, such as optical and SAR images, independently, lacking effective methods for fusion analysis. Particularly for earthquake damage assessment, how to automatically and accurately compare and analyze pre-disaster optical and post-disaster SAR images to extract building damage information is a recognized technical challenge. Existing solutions cannot effectively bridge the gap created by the significant differences in imaging mechanisms between optical and radar images, resulting in a severe deficiency in the ability to collaboratively analyze heterogeneous data.

[0018] 3. The processing flow is cumbersome and lacks automation capabilities:

[0019] The workflows described in the background section, whether for fire assessment, flood detection, or earthquake damage interpretation, heavily rely on professionals switching between multiple software platforms, as well as extensive manual interaction and parameter adjustments. The entire process is semi-automated and labor-intensive, not only inefficient in handling emergency response within minute-level timeframes, but also highly dependent on the operator's subjective experience, lacking consistency and repeatability.

[0020] 4. The system architecture is isolated, with a lack of proactive services:

[0021] Most existing systems function as passive analysis tools, requiring users to manually acquire, upload, and initiate analysis tasks after a disaster occurs. This passive, reactive approach contradicts the proactive early warning model advocated by modern emergency management. Current technology lacks an integrated, proactive service platform capable of unattended operation, automatic data monitoring, and proactive alert push notifications. Summary of the Invention

[0022] In view of this, the present invention provides a disaster monitoring and assessment system that integrates multi-source heterogeneous remote sensing and collaborative analysis. It can overcome the limitations of severe weather conditions and realize all-weather, all-time monitoring of disasters such as floods, earthquakes, and fires. It upgrades from passive analysis to an intelligent early warning platform with proactive discovery and real-time alarm, thereby improving the efficiency and accuracy of emergency response.

[0023] To solve the above-mentioned technical problems, the present invention is implemented as follows.

[0024] A disaster monitoring and assessment system integrating multi-source heterogeneous remote sensing and collaborative analysis, characterized in that it includes: a data layer, an application layer, and a user layer;

[0025] The data layer receives and stores image data of various disasters in folders;

[0026] The application layer includes a sentinel service module and three assessment service modules; the assessment service modules include an earthquake assessment service module, a flood analysis service module, and a fire analysis service module.

[0027] The sentinel service module monitors the entry of image data into the directory of each folder. When it detects the creation of a new image file in a folder, the sentinel service module calls the corresponding assessment service module to perform disaster assessment according to the task corresponding to the folder, receives the analysis feedback from each assessment service module, and sends the disaster assessment results to the user layer.

[0028] The earthquake assessment service module extracts pre-disaster optical images and post-disaster SAR images from the data layer, extracts the same type of statistical features from the two types of images and calculates the differences, filters earthquake disaster patches based on the intensity of statistical feature differences, and generates an earthquake disaster assessment result file.

[0029] The flood analysis service module extracts pre-disaster and post-disaster SAR images from the data layer, extracts pre-disaster and post-disaster water body masks, uses Boolean logic difference operations on the two water body masks, determines the newly added water body areas that appear after the disaster as the flood inundation areas, and generates a flood disaster assessment result file.

[0030] The fire analysis service module extracts multispectral images from the data layer, performs fire point detection and fire zone extraction, and generates a fire assessment result file.

[0031] The user layer includes a client application used to display disaster assessment results.

[0032] Preferably, the earthquake assessment service module includes a heterogeneous image preprocessing module, a multi-feature vectorization extraction module, an adaptive threshold segmentation module, a morphological post-processing module, and an earthquake disaster assessment result output module;

[0033] The heterogeneous image preprocessing module is used to preprocess pre-disaster optical images and post-disaster SAR images.

[0034] The multi-feature vectorization extraction module is used to extract feature vectors composed of multiple statistical features from the pre-processed pre-disaster optical images and post-disaster SAR images; calculate the absolute difference between the two sets of feature vectors before and after the disaster and obtain the mean value of the vector elements to obtain the pre-disaster and post-disaster feature difference intensity corresponding to each pixel; the pre-disaster and post-disaster feature difference intensity of each pixel is composed of a single-channel change intensity map.

[0035] The adaptive threshold segmentation module is used to perform adaptive threshold segmentation on the change intensity map using the percentile method to obtain change patches that may be earthquake disaster areas.

[0036] The morphological post-processing module is used to calculate the aspect ratio of the minimum bounding rectangle of the changed patches, exclude changed patches with aspect ratios greater than a set aspect ratio threshold, and use the remaining changed patches as identified earthquake disaster areas.

[0037] The earthquake disaster assessment result output module is used to encapsulate the identified earthquake disaster areas and their attributes into a file for output.

[0038] Preferably, the heterogeneous image preprocessing module uses a contrast-limited adaptive histogram equalization method to enhance texture for pre-disaster optical images and a Lee filter to suppress speckle noise for post-disaster SAR images.

[0039] Preferably, the multi-feature vectorization extraction module extracts a feature vector composed of multiple statistical features by extracting the mean, standard deviation, gradient mean, and gradient standard deviation within the neighborhood of each pixel to form a feature vector.

[0040] Preferably, the flood analysis service module includes a water body extraction module, a water body mask time series difference module, and a flood disaster assessment result output module;

[0041] The water extraction module is used to extract water bodies from SAR images before and after the disaster in parallel, and to obtain the pre-disaster water body mask Mask_pre and the post-disaster water body mask Mask_post.

[0042] The water body mask timing difference module is used to perform Boolean logic difference operation Mask_post AND (NOTMask_pre) to obtain newly added water body areas that only appear after the disaster, i.e., the identified flood-inundated areas; where AND represents Boolean intersection operation and NOT represents Boolean negation operation.

[0043] The flood disaster assessment result output module is used to encapsulate the identified flood-inundated areas and their attributes into a file for output.

[0044] Preferably, the water extraction module extracts water in the following manner:

[0045] Threshold segmentation: Pixels in SAR images below the backscattering threshold are initially identified as water bodies to obtain the first water body mask;

[0046] Morphological closing operation: A rectangular structuring element is used to perform a morphological closing operation on the first water body mask that has been initially identified in order to fill the internal voids and obtain the second water body mask.

[0047] Connected component denoising: Connected component analysis is performed on the second water body mask to remove all patches with an area smaller than a preset pixel threshold, resulting in the third water body mask. The third water body mask obtained based on pre-disaster SAR images is called the pre-disaster water body mask Mask_pre, and the third water body mask obtained based on post-disaster SAR images is called the post-disaster water body mask Mask_post.

[0048] Preferably, the fire analysis service module includes a fire point detection module and a fire zone extraction module;

[0049] The fire detection module is used to extract single-period multispectral images from the data layer, including short-wave infrared band B12, near-infrared band B8A, and red band B4. First, based on a first combination condition where the values ​​of near-infrared band B8A and short-wave infrared band B12 are simultaneously below a first reflectivity threshold, water bodies are identified. Based on a second combination condition where the value of red band B4 is greater than a set brightness threshold, and the value of short-wave infrared band B12 is lower than an adaptive high-temperature threshold T_adaptive_hot, clouds are identified. A cloud-water mask is constructed based on the identified water bodies and clouds. Next, for areas not covered by the cloud-water mask, fire zones are identified based on a third combination condition where short-wave infrared band B12 is higher than the adaptive high-temperature threshold T_adaptive, the fire index FI = B12 / B8A is higher than the adaptive fire index threshold FI_adaptive, and the fire index FI is higher than the neighborhood threshold with a difference greater than a set difference threshold. The identified fire zones are then packaged into a file for output. The adaptive high-temperature threshold T_adaptive is determined based on the statistical characteristics of the multispectral images.

[0050] The burned area extraction module is used to extract post-disaster multispectral images from the data layer, including shortwave infrared band B12, near-infrared band B8A, red band B4, yellow band B6, and red-edge band B7; calculate the normalized rate of fire (NBR), normalized rate of vegetation (NDVI), and scorch area index (BAIS2); use the Otsu adaptive thresholding algorithm to calculate the BAIS2 threshold that distinguishes between scorched and unscorched areas; if the BAIS2 value of a pixel is greater than the BAIS2 threshold, the NBR value is lower than a preset burning characteristic threshold, and the NDVI value is lower than a preset vegetation mortality threshold, then the pixel is identified as a burned area pixel; and the identified burned areas are packaged into a file for output.

[0051] Preferably, the data layer includes folders for storing images for the earthquake assessment service module, flood analysis service module, and fire analysis service module, respectively.

[0052] A first folder is set up for the earthquake assessment service module to store pre-disaster optical images and post-disaster SAR images to be assessed; when the data in the first folder changes, the sentinel service module calls the earthquake assessment service module to perform earthquake disaster assessment.

[0053] Two folders are set up for the flood analysis service module to store SAR images from two periods. In the post-flood assessment application, when the data in the two folders changes, the sentinel service module calls the flood analysis service module to perform a flood disaster assessment. In the real-time flood monitoring application, the second folder corresponds to the historical time step, and the third folder corresponds to the current time step. When the time step changes, the latest SAR image in the third folder is stored in the second folder, and the SAR image of the new time step is received and stored in the third folder. When the data in the third folder changes, the sentinel service module calls the flood analysis service module to perform a flood disaster assessment and waits for the flood disaster assessment result file returned by the flood analysis service module. If the newly added water body area exceeds the preset value, a flood disaster alarm message is sent to the client.

[0054] The fire analysis service module is configured with a fourth and a fifth folder. The fourth folder stores multispectral images of fire detection, and the fifth folder stores multispectral images after a fire. In the post-fire assessment application, when the data in the fourth folder changes, the sentinel service module calls the fire analysis service module to detect fire points; when the data in the fifth folder changes, the sentinel service module calls the fire analysis service module to extract the burned area. In the real-time fire monitoring application, the data in the fourth folder is updated periodically. When the data in the fourth folder changes, the sentinel service module calls the fire analysis service module to detect fire points and waits for the fire assessment result file returned by the fire analysis service module. If a fire point is identified, a fire alarm message is sent to the client.

[0055] Preferably, the sentinel service module and the evaluation service module use an HTTP connection, with the absolute path of the file containing the image data to be evaluated in the folder as a parameter, and the corresponding evaluation service module is automatically invoked through an HTTP POST request;

[0056] The application layer sets up a notification API, which maintains a real-time communication channel with the client via a WebSocket long connection. After receiving the analysis feedback from the evaluation service module, the sentinel service module confirms that a new disaster has occurred, and then constructs an alarm message containing key summary information and sends it to the notification API. The notification API broadcasts the alarm message to all online clients in real time via the WebSocket long connection to achieve proactive early warning.

[0057] Preferably, the client includes an alarm management module, a 3D visualization service engine, and a statistics module;

[0058] Upon receiving an alarm message, the alarm management module triggers a visual notification to alert the user that a new disaster event has occurred; upon receiving a user instruction, it retrieves the disaster assessment result file associated with the alarm message and starts the 3D visualization service engine.

[0059] The 3D visualization service engine has a built-in style rule set that automatically matches thematic colors to the patches in the disaster area of ​​the disaster assessment result file based on the task type, and renders the patches; then it drives the 3D camera to fly and automatically focus on the disaster area.

[0060] The statistics module iterates through the information in the disaster assessment result file, automatically calculates statistical indicators including the total area of ​​the disaster area and the number of map patches, and displays them in the statistics panel on the client display interface.

[0061] Beneficial effects:

[0062] (1) This invention provides an automated and proactive disaster early warning solution. The sentinel service module monitors and automatically triggers backend analysis tasks. Upon obtaining a positive analysis result, it proactively broadcasts and alerts the frontend user interface in real time, avoiding human intervention and meeting the needs of modern comprehensive emergency management. Under the system architecture of this invention, a multi-hazard backend algorithm service module can be built-in, working collaboratively with the automated sentinel module and the intelligent visualization frontend module to form a complete technical closed loop from automatic data entry to proactive result delivery.

[0063] (2) This invention provides a building damage assessment scheme based on optical and SAR heterogeneous remote sensing images. By using a heterogeneous feature-level fusion method, changes in heterogeneous images are detected through comparison in feature space rather than pixel space, improving the collaborative analysis capability of heterogeneous data and effectively bridging the gap caused by the huge differences in imaging mechanisms between optical and radar images. In the preferred scheme, based on the characteristics of building damage, combined with geometric morphology, the changed patches are intelligently screened to accurately identify damaged buildings and improve the accuracy of emergency response.

[0064] (3) This invention provides an all-weather, automated method for extracting the flood inundation range. After extracting the water mask from the SAR images before and after the disaster, the newly added inundation area is accurately identified through Boolean logic operations based on time-series differences. Boolean logic operations have low computational cost and can support large-scale, all-weather real-time calculations, making real-time flood monitoring and early warning possible and improving the efficiency of emergency response.

[0065] (4) This invention provides an automatic fire remote sensing analysis method, which includes two key technical points: when detecting fire points, an extreme high temperature exemption rule is adopted to unconditionally identify pixels that meet a specific high temperature threshold as fire points; when mapping the fire zone, the Otsu adaptive threshold segmentation method is adopted to achieve fully automatic and highly complete fire zone extraction.

[0066] (5) Most of the thresholds of the system of the present invention are adaptively calculated based on actual image data, which does not require manual parameter adjustment, avoids excessive manual interaction, and does not require deep reliance on professional personnel to switch between multiple software platforms, thus improving the automation level of the system. Attached Figure Description

[0067] Figure 1 This is a schematic diagram of a disaster monitoring and assessment system integrating multi-source heterogeneous remote sensing and collaborative analysis, as described in an embodiment of the present invention.

[0068] Figure 2 A block diagram of the earthquake assessment service module.

[0069] Figure 3 This is an assessment result of the damaged area of ​​a building in a specific example.

[0070] Figure 4 A block diagram of the flood analysis service module.

[0071] Figure 5 This is the identification result of the flooded area in a specific example.

[0072] Figure 6 This is a flowchart illustrating the process of implementing fire detection in the fire detection module.

[0073] Figure 7 This is the identification result of the fire point region in an example.

[0074] Figure 8 This is a flowchart illustrating the process of extracting the burned area for the burned area extraction module.

[0075] Figure 9 This is the identification result of an example of an over-fire zone.

[0076] Figure 10 Workflow for the Sentinel Service module.

[0077] Figure 11 This is a block diagram of the client's components. Detailed Implementation

[0078] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0079] This invention aims to provide an integrated, intelligent, and proactive comprehensive disaster monitoring and assessment technology solution, specifically addressing the following four core technical issues:

[0080] Question 1: How to build a technical system that can integrate optical and SAR radar data to overcome the limitations of severe weather conditions and achieve all-weather, all-time monitoring capabilities for disasters such as floods and earthquakes?

[0081] Question 2: How to design an innovative feature-level change detection algorithm for heterogeneous images to overcome the technical bottleneck that makes it difficult to directly compare pre-disaster optical images and post-disaster SAR images, and to achieve rapid and automated assessment of building damage.

[0082] Question 3: How to establish an end-to-end automated processing pipeline to transform the analysis process of disasters such as fires, floods, and earthquakes from semi-manual to fully automated, thereby significantly improving the efficiency and accuracy of emergency response.

[0083] Question 4: How to design a service architecture with automated sentinel monitoring and proactive alarm functions, upgrading the system from a passive analysis tool to an intelligent early warning platform that can proactively discover and provide real-time alarms?

[0084] To solve the above problems, Figure 1 The present invention illustrates a disaster monitoring and assessment system integrating multi-source heterogeneous remote sensing and collaborative analysis, as shown in the figure. The system includes a data layer, an application layer, and a user layer.

[0085] The data layer receives and stores image data for various disasters in folders. External systems can push relevant image data to this system in real time, periodically, or upon event triggering. This system stores the data in the folder corresponding to the task within the data layer. These tasks include: earthquake assessment, flood analysis, and fire analysis.

[0086] The application layer includes a sentinel service module and three assessment service modules. The assessment service modules correspond to the tasks mentioned above, including an earthquake assessment service module, a flood analysis service module, and a fire analysis service module.

[0087] The Sentinel Service Module is the core component and bridge for realizing proactive monitoring, assessment, and alerting. It monitors the entry of image data into each folder. When a new image file is detected in a folder, the Sentinel Service Module calls the corresponding assessment service module to perform a disaster assessment based on the folder's associated task. It receives analysis feedback from each assessment service module and sends the disaster assessment results to the user layer.

[0088] Here, the Sentinel Service module can perform backend checks. If the disaster analysis result is positive, it constructs a lightweight alarm message containing key summary information and sends it to the client. If no disaster is detected, it may not issue an alarm, or may only notify the client that no disaster was detected. After seeing the alarm message, the client can confirm it on the client side, and then the Sentinel Service module will send a detailed disaster information file to the client for rendering, display, and further statistical analysis.

[0089] The earthquake assessment service module extracts pre-disaster optical images and post-disaster SAR images from the data layer, extracts the same type of statistical features from the two types of images and calculates the differences, filters earthquake hazard patches based on the intensity of statistical feature differences, and generates an earthquake hazard assessment result file.

[0090] The flood analysis service module extracts pre-disaster and post-disaster SAR images from the data layer, extracts pre-disaster and post-disaster water body masks, uses Boolean logic difference operations on the two water body masks to determine the newly added water body areas that appear after the disaster as the flood inundation areas, and generates a flood disaster assessment result file.

[0091] The fire analysis service module extracts multispectral images from the data layer, performs fire point detection and fire zone extraction, and generates fire assessment result files.

[0092] The user layer includes a client application used to display disaster assessment results.

[0093] As can be seen, this invention employs a microservice architecture with front-end and back-end separation to achieve an automated and proactive disaster early warning solution. The sentinel service module monitors and automatically triggers back-end analysis tasks. Upon obtaining positive analysis results, it proactively broadcasts and alerts to the front-end user interface in real time, avoiding human intervention and meeting the needs of modern comprehensive emergency management. Under this architecture, a multi-hazard back-end algorithm service module can be built-in, working collaboratively with the automated sentinel module and the intelligent visualization front-end module to form a complete technical closed loop from automatic data entry to proactive results delivery.

[0094] The implementation of each module is described in detail below.

[0095] (1) Earthquake assessment service module

[0096] Figure 2 The diagram shows the components of the earthquake assessment service module. As shown, it includes a heterogeneous image preprocessing module, a multi-feature vectorization extraction module, an adaptive threshold segmentation module, a morphological post-processing module, and an earthquake hazard assessment result output module.

[0097] The workflow of this earthquake assessment service module is as follows:

[0098] Step 101: The heterogeneous image preprocessing module preprocesses the pre-disaster optical images and post-disaster SAR images.

[0099] In this step, preprocessing includes: contrast-limited adaptive histogram equalization for pre-disaster optical images to enhance texture, and Lee filter for post-disaster SAR images to suppress speckle noise.

[0100] Step 102: Feature-level fusion and difference.

[0101] This step is performed by the multi-feature vectorization extraction module, which includes feature extraction and feature differencing.

[0102] Feature Extraction: The multi-feature vectorization extraction module extracts feature vectors composed of multiple statistical features from both the pre-disaster optical image and the post-disaster SAR image. In this embodiment, a four-dimensional feature vector is extracted for each pixel, which consists of four statistical features: the mean, standard deviation, gradient mean, and gradient standard deviation within the neighborhood of each pixel. This step transforms the two heterogeneous images into the same comparable feature space.

[0103] Feature Difference: The multi-feature vectorization extraction module calculates the absolute difference between the pre-disaster and post-disaster feature vectors, resulting in a four-channel feature difference map. Subsequently, the difference values ​​of the four channels are averaged and merged into a single-channel, continuous grayscale value change intensity map. The value of each element in this change intensity map represents the intensity of the pre-disaster and post-disaster feature difference, directly indicating the drastic change in the comprehensive features of the ground cover at that point.

[0104] Step 103: Variation purification and intelligent screening.

[0105] Segmentation: The adaptive threshold segmentation module is used to perform adaptive threshold segmentation on the change intensity map using the percentile method to obtain change patches that may be earthquake-prone areas. The percentile p can be selected as 80%, extracting the top 20% of pixels with the most dramatic changes from the change intensity map.

[0106] Intelligent Filtering: To achieve rapid and accurate assessment of building damage, this step employs a morphological post-processing module to calculate the aspect ratio of the minimum bounding rectangle of the segmented change patches. Since building collapse patches tend to be clump-like (aspect ratio close to 1), while changes in linear features like roads appear as long, thin strips, an aspect ratio threshold is set to exclude change patches with aspect ratios greater than this threshold. The remaining change patches are then identified as earthquake-prone areas. By automatically filtering out patches with excessively large aspect ratios, the damaged building areas are precisely identified. Figure 3 This shows an example of the assessment results for a damaged area of ​​a building.

[0107] Step 104: Output earthquake disaster results.

[0108] The earthquake disaster assessment results output module is used to encapsulate the identified earthquake disaster areas and their attributes into a file for output.

[0109] In this embodiment, the earthquake hazard assessment result output module vectorizes the finally identified changed patches, converting them into a series of polygons with geographic coordinates, and calculates the geographic area of ​​each polygon. Finally, all patches and their attributes are encapsulated into a standard GeoJSON file for output.

[0110] (2) Flood Analysis Service Module

[0111] Figure 4 The diagram shows the components of the flood analysis service module. As shown, it includes a water body extraction module, a water body mask temporal difference module, and a flood disaster assessment result output module. This module utilizes the ability of SAR imagery to penetrate clouds and rain, and through high-precision temporal difference, achieves automated, all-weather extraction of flood inundation ranges.

[0112] The workflow of this flood analysis service module is as follows:

[0113] Step 201: Input pre-disaster and post-disaster SAR images.

[0114] The system receives two SAR images covering the same area as input. These two images are typically backscattering coefficient raster data in decibels (dB) that have undergone radiometric calibration and geocoding preprocessing. VH or VV polarization data, which are most sensitive to water bodies, are selected.

[0115] Step 202: Parallel single-stage water body extraction.

[0116] To improve processing efficiency, the water extraction module executes the same water extraction sub-process in parallel for both pre-disaster and post-disaster images. This process includes three key steps:

[0117] Threshold segmentation: Utilizing the physical characteristic of calm water surfaces as specular reflectors, resulting in extremely low backscattering values ​​in SAR images, a fixed low backscattering threshold is set. The water extraction module initially identifies pixels in the SAR image below the backscattering threshold as water bodies, obtaining a first water body mask. For example, for Sentinel-1 VH polarized images that have undergone radiometric calibration and logarithmic decibel (dB) conversion, this threshold can be set to -22dB, meaning that all pixels with backscattering coefficient values ​​less than -22dB are initially identified as water bodies.

[0118] Morphological Closure Operation: To eliminate "salt and pepper noise" and voids caused by ripples, noise or small features inside the water body, the water extraction module further uses a rectangular structuring element to perform morphological closure operation on the initially identified first water body mask to fill the internal voids and obtain a second water body mask with complete and continuous water body boundaries.

[0119] Connectivity component denoising: To further eliminate isolated small patches that are not connected to the real water body due to surface shadows or sensor noise, the water extraction module performs connectivity component analysis on the second water body mask, removing all patches with an area smaller than a preset pixel threshold to obtain a clean and reliable third water body mask; the third water body mask obtained based on pre-disaster SAR imagery is the pre-disaster water body mask Mask_pre, and the third water body mask obtained based on post-disaster SAR imagery is the post-disaster water body mask Mask_post.

[0120] Step 203: Timing difference.

[0121] This is a key step in this model. After obtaining clean pre-disaster water masks (Mask_pre) and post-disaster water masks (Mask_post), the water mask temporal difference module performs precise Boolean logic difference operations to obtain newly added water areas that only appear after the disaster, i.e., the identified flood-inundated areas.

[0122] Boolean logic difference operations are as follows:

[0123] Flood-prone area mask = Mask_post AND (NOT Mask_pre)

[0124] Here, AND represents Boolean intersection operation, and NOT represents Boolean negation operation.

[0125] This calculation can identify, with extreme precision, newly created water bodies that appear only after a disaster, i.e., the extent of flooding. At the same time, it can automatically and perfectly exclude the influence of normal water bodies that existed before the disaster, such as rivers, lakes, and reservoirs, greatly improving the purity and usability of the final results. Figure 5 An example of the identification results of flood-inundated areas is shown.

[0126] Step 204: Output the flood-inundated area.

[0127] The flood disaster assessment results output module encapsulates the identified flood-inundated areas and their attributes into a file for output.

[0128] In this embodiment, the flood disaster assessment result output module vectorizes the final generated binary mask of the flood-inundated area, converting it into a series of polygons with geographic coordinates, and accurately calculates the geographic area of ​​each polygon. Finally, all flood-inundated area patches and their attributes are encapsulated into a standard GeoJSON file for output.

[0129] (3) Fire Analysis Service Module

[0130] The fire analysis service module includes a fire point detection module and a burned zone extraction module. By introducing a robust algorithm combination, this fire analysis service module elevates the automation and accuracy of traditional fire remote sensing analysis to a new level, enabling high-precision real-time fire point detection and post-disaster burned zone mapping simultaneously within a unified workflow.

[0131] Figure 6 The diagram illustrates the process of a fire detection module implementing fire detection. As shown in the figure, it includes the following steps:

[0132] Step 301: Input multispectral image.

[0133] The fire detection module extracts single-period multispectral images from the data layer, including shortwave infrared band B12, near-infrared band B8A, and red light band B4.

[0134] Step 302: Intelligent cloud water mask acquisition.

[0135] Water Mask is determined based on a first combination of conditions: both the near-infrared band B8A and the short-wave infrared band B12 values ​​are below a set low reflectivity threshold. In a preferred embodiment, the low reflectivity threshold is set to 0.05. If a pixel's B8A < 0.05 and B12 < 0.05, then that pixel is considered water and needs to be removed. Therefore, the water mask is represented as Water Mask = (B8A < 0.05) AND (B12 < 0.05).

[0136] Intelligent Cloud Mask: Traditional cloud masks rely solely on the brightness of visible light bands (such as B4), easily misidentifying equally bright fire core areas as clouds. This invention's intelligent cloud mask adds a new criterion: the B12 reflectivity must be below an adaptive high-temperature threshold T_adaptive. A pixel is only identified as a cloud if both conditions are met simultaneously: B4 is greater than the set brightness threshold and B12 is less than the adaptive high-temperature threshold T_adaptive_hot.

[0137] The adaptive high-temperature threshold T_adaptive is acquired from the image, rather than being pre-specified. In this embodiment, the median Median_b12 and the median absolute deviation MAD_b12 of the B12 band are calculated. The adaptive high-temperature threshold is calculated based on the median and the median absolute deviation, and the specific calculation formula can be: T_adaptive = Median_b12 + k T ×1.4826×MAD_b12. k T This is a scaling factor, for example, it can be 5.0.

[0138] In a preferred embodiment, if the brightness threshold is set to 0.3, then the cloud mask is represented as: CloudMask = B4 > 0.3 AND B12 <T_adaptive。

[0139] Water Mask and Cloud Mask together form a cloud and water mask, thereby keeping water and clouds out of fire detection.

[0140] Step 303: Determine the dual adaptive threshold.

[0141] To adapt to the image characteristics of different regions and seasons, this invention does not use a fixed threshold in the fire point identification step 304. Instead, it dynamically calculates two core thresholds based on the statistical characteristics of the current image itself: one is the adaptive high temperature threshold T_adaptive based on the B12 band used in step 302, and the other is the adaptive fire index threshold FI_adaptive based on the fire index FI=B12 / B8A.

[0142] Here, the adaptive fire index threshold FI_adaptive is obtained from the image, rather than being pre-specified. In this embodiment, the fire index FI = B12 / B8A is calculated for each pixel, the median of the fire index is Median_fi, and the median absolute deviation is MAD_fi. The adaptive fire index threshold is calculated based on the median and the median absolute deviation, and the specific calculation formula can be: FI_adaptive = Median_fi + k fi ×1.4826×MAD_fi. k fi This is the proportionality coefficient.

[0143] Step 304: Identify the fire point.

[0144] Identifying fire points in areas not covered by cloud / water masks. A pixel must simultaneously meet a combination of the following three conditions to be identified as a standard fire point:

[0145] a. The B12 value of the pixel is higher than the adaptive high temperature threshold T_adaptive;

[0146] b. The fire index FI=B12 / B8A of the pixel is higher than the adaptive fire index threshold FI_adaptive;

[0147] C. If the fire index FI of a pixel is significantly higher than that of its immediate surrounding background area, the fire index FI being higher than that of its neighbors and the difference being greater than a set difference threshold can be used as the judgment condition.

[0148] Step 305: Encapsulate the identified fire points into a file for output.

[0149] This step confirms, morphologically optimizes, and vectorizes all identified fire points, outputting GeoJSON of the fire point regions. Figure 7 An example of the fire point area identification results is shown.

[0150] Figure 8 The flowchart illustrates the process of fire zone extraction using the fire zone extraction module. As shown in the figure, it includes the following steps:

[0151] Step 401: Input multispectral image.

[0152] The fire zone extraction module extracts post-disaster multispectral images from the data layer, including shortwave infrared band B12, near-infrared band B8A, red light band B4, yellow light band B6, and red edge band B7.

[0153] Step 402: Calculate the multispectral index.

[0154] To enhance the surface burn signal from multiple dimensions, this step calculates three key spectral indices in parallel: Normalized Burn Index (NBR), Normalized Difference Vegetation Index (NDVI), and Burn Area Index 2 (BAIS2).

[0155] Step 403: Otsu adaptive threshold segmentation.

[0156] This is the key to achieving full automation. This invention selects the BAIS2 index image, which has the highest recognition rate for burned areas, and processes it using the Otsu adaptive thresholding algorithm. This algorithm can automatically calculate a threshold that optimally separates burned and unburned pixels based on the image's grayscale histogram, completely eliminating the subjectivity and limitations of manually setting thresholds. Based on the Otsu adaptive thresholding, burned and unburned areas in the BAIS2 index image are identified, forming a candidate overburned area mask, where 1 represents burned areas and 0 represents unburned areas.

[0157] Step 404: Multi-index collaborative constraints.

[0158] To eliminate potential spurious change points in the candidate region, such as arid vegetation and bare soil, this invention introduces NBR and NDVI as physical constraints, combined with Otsu's adaptive thresholding, to perform collaborative filtering on the candidate region. A pixel must simultaneously meet the following three conditions to be identified as a true burnt area:

[0159] a. The pixel is 1 in the candidate overfire area mask, that is, the BAIS2 value is higher than the Otsu adaptive threshold;

[0160] b. If the NBR value of a pixel is lower than the preset combustion characteristic threshold, it indicates moisture loss and structural damage;

[0161] c. If the NDVI value of a pixel is lower than the preset vegetation death threshold, it indicates that vegetation activity has been lost.

[0162] Step 405: Patch aggregation and morphological optimization.

[0163] This is crucial for improving the completeness of the results. To connect adjacent small fire spots that may be temporarily separated by roads, rivers, etc., into a complete area that reflects the overall impact of the fire, the algorithm uses an elliptical structuring element with a size of 25×25 pixels as the computational kernel for patch aggregation. This step can effectively glue together adjacent burned patches to form one or a few large, coherent fire areas. Figure 9 An example of the fire zone identification results is shown.

[0164] Step 406: Encapsulate the identified overburned areas into a file for output.

[0165] This step performs morphological cleansing, raster smoothing, and vector simplification on the aggregated mask, ultimately outputting a lightweight, boundary-smoothed overburning area vector GeoJSON.

[0166] (4) Sentinel Service Module

[0167] Figure 10 The workflow of the Sentinel Service module is illustrated. As shown in the figure, it includes the following steps:

[0168] Step 501: File monitoring and preprocessing.

[0169] See Figure 1The data layer provides separate folders for storing images for the earthquake assessment service module, flood analysis service module, and fire analysis service module. The system initiates a file system observer service as a sentinel service module to continuously monitor the file entry directories in each folder within the data layer storage space. When a new TIF format image file is created, the sentinel service module executes step 502, which, based on the folder's corresponding task, calls the corresponding assessment service module to perform a disaster assessment. To ensure the file is completely written and to avoid duplicate processing, the system performs two preprocessing steps: event deduplication and readiness waiting. For non-duplicate files, the system waits until the file is fully received before executing subsequent step 502, which calls the corresponding assessment service module.

[0170] Step 502: Automated task invocation.

[0171] The Sentinel Service Module and the Evaluation Service Module communicate via HTTP. The absolute path of the file containing the image data to be evaluated in the folder is used as a parameter, and the corresponding Evaluation Service Module is automatically invoked via an HTTP POST request.

[0172] The system of the present invention can perform assessment for earthquake disasters, post-disaster assessment and real-time monitoring and identification for flood disasters, and post-disaster assessment and real-time monitoring and identification of fire points, as well as assessment of burned areas, for fire analysis.

[0173] Specifically: A first folder is set up for the earthquake assessment service module to store pre-disaster optical images and post-disaster SAR images to be assessed; when the data in the first folder changes, the sentinel service module calls the earthquake assessment service module to perform earthquake disaster assessment.

[0174] Two folders are set up for the flood analysis service module to store SAR images from two periods. In the post-flood assessment application, when the data in the two folders changes, the sentinel service module calls the flood analysis service module to perform a flood disaster assessment. In the real-time flood monitoring application, the second folder corresponds to the historical time step, and the third folder corresponds to the current time step. When the time step changes, the latest SAR image in the third folder is stored in the second folder, and the SAR image of the new time step is received and stored in the third folder. When the data in the third folder changes, the sentinel service module calls the flood analysis service module to perform a flood disaster assessment and waits for the flood disaster assessment result file returned by the flood analysis service module. If the newly added water body area exceeds the preset value, a flood disaster alarm message is sent to the client.

[0175] The fire analysis service module is configured with a fourth and a fifth folder. The fourth folder stores multispectral images of fire detection, and the fifth folder stores multispectral images after a fire. In the post-fire assessment application, when the data in the fourth folder changes, the sentinel service module calls the fire analysis service module to detect fire points; when the data in the fifth folder changes, the sentinel service module calls the fire analysis service module to extract the burned area. In the real-time fire monitoring application, the data in the fourth folder is updated periodically. When the data in the fourth folder changes, the sentinel service module calls the fire analysis service module to detect fire points and waits for the fire assessment result file returned by the fire analysis service module. If a fire point is identified, a fire alarm message is sent to the client.

[0176] Step 503: Result processing and proactive alarm.

[0177] The Sentinel service module synchronously awaits the GeoJSON format analysis results returned by the evaluation service module API. It checks if the `features` array in the returned GeoJSON is empty. If not, the service generates a unique identifier ID for the task and saves the complete GeoJSON result file to the archive directory.

[0178] Proactive Alarm: This is a key innovation of this invention. The application layer sets up a notification API, which maintains a real-time communication channel with the client via a WebSocket long-lived connection. After receiving analysis feedback from the evaluation service module, the sentinel service module confirms a new disaster and constructs a lightweight alarm message containing key summary information, sending it to the notification API set up in the application layer. This notification API, through the maintained WebSocket long-lived connection, broadcasts the alarm message in real time to all online clients, achieving proactive early warning.

[0179] The client is responsible for presenting the detailed analysis results to the user in an intuitive and efficient manner, and for receiving proactive alerts pushed to the user. For example... Figure 11 As shown, the client includes an alarm management module, a 3D visualization service engine, and a statistics module. The client's visualization and interaction process includes the following steps:

[0180] Step 601: View initialization.

[0181] The client uses a singleton pattern to initialize a globally unique CesiumJS 3D Earth Viewer instance. It loads and merges Google satellite imagery with Chinese annotations from Tianditu (a Chinese online map platform) as a composite base map.

[0182] Step 602: Alarm Management.

[0183] The client maintains a persistent real-time communication channel with the backend notification API via WebSocket to receive proactive alarm messages pushed by the Sentinel Service module when a new disaster is detected. Upon receiving an alarm, the Alarm Management module immediately triggers a visual notification, alerting the user to the new disaster event. After the user clicks the notification, the Alarm Management module receives the user's instruction, automatically retrieves the disaster assessment result file (detailed GeoJSON result) associated with the alarm message, and starts the 3D visualization service engine to execute the subsequent rendering and presentation process.

[0184] Step 603: Dynamic thematic style rendering.

[0185] The 3D visualization service engine has a built-in set of style rules. Based on the type of the current analysis task, it automatically matches preset thematic colors to the polygons in the returned GeoJSON file, renders the polygons, and achieves professional-grade visualization without user configuration.

[0186] Step 604: Result presentation and thematic map generation.

[0187] After rendering the map features, the 3D visualization service engine drives the 3D camera to fly smoothly and automatically focus on the disaster area. At the same time, the statistics module traverses the attributes in the GeoJSON file, automatically calculates statistical indicators such as the total area of ​​the disaster area and the number of map features, and displays them in the statistics panel on the side of the interface, forming a complete feedback loop from visualization to quantification.

[0188] In summary, the solutions of the embodiments of the present invention can bring the following effects:

[0189] (1) It has achieved deep fusion and high-precision assessment of heterogeneous remote sensing data, significantly improving the level of automation and analysis accuracy:

[0190] For scenarios requiring the fusion of heterogeneous data, such as earthquake damage assessment, existing technologies face technical bottlenecks, heavily relying on inefficient and subjective manual interpretation. This invention employs a feature-level fusion method for heterogeneous images. By comparing data in the feature space rather than the pixel space, and combining this with intelligent aspect ratio filtering, it achieves, for the first time, automated and high-precision damage assessment of pre-disaster optical and post-disaster SAR images. Simultaneously, rules such as Otsu adaptive segmentation in fire and flood analysis models completely replace traditional manual threshold adjustments.

[0191] By deeply fusing and intelligently filtering heterogeneous remote sensing data, this embodiment transforms disaster assessment from labor-intensive manual interpretation to computationally intensive machine intelligence analysis. This not only reduces assessment time from days to minutes, achieving a leap in emergency response efficiency, but also ensures the scientific validity and repeatability of assessment results through objective and consistent algorithms.

[0192] (2) It has achieved a model upgrade from passive analysis to proactive early warning, greatly improving the agility of emergency response:

[0193] Existing systems are all passive analysis tools, requiring human intervention. This invention, through the design of a sentinel service module, establishes a fully automated workflow where data is analyzed immediately upon arrival and alerts are issued upon completion of analysis. This module automatically triggers analysis through file system monitoring and proactively pushes alerts to the front end via WebSocket real-time communication technology. This achieves a modern emergency response model where tasks are handled by humans, eliminating the time delay between data acquisition and manual discovery. It ensures that decision-makers are aware of disaster anomalies immediately, gaining valuable time for disaster prevention and mitigation.

[0194] (3) It has achieved integrated management and intelligent presentation of multiple disasters, significantly reducing the system's usage threshold and information barriers:

[0195] In existing technologies, disaster management systems are often isolated, resulting in fragmented information. This invention seamlessly integrates the monitoring and assessment functions for three major disasters—fire, flood, and earthquake—into a unified front-end platform. Its intelligent visualization engine, through automated layer lifecycle management and dynamic thematic style rendering, presents complex spatial analysis results to users in the most intuitive and understandable way. This system breaks down information silos, providing a unified decision-making view for comprehensive emergency command. Its highly automated visualization and statistical functions enable emergency management personnel, even those without remote sensing expertise, to easily operate and quickly interpret the analysis results, greatly expanding the breadth and depth of advanced remote sensing technology applications.

[0196] The specific embodiments described above only illustrate the design principles of the present invention. The shapes and names of the components in this description may differ and are not limited. Therefore, those skilled in the art can modify or make equivalent substitutions to the technical solutions described in the foregoing embodiments; and these modifications and substitutions do not depart from the inventive spirit and technical solutions of the present invention, and should all fall within the protection scope of the present invention.

Claims

1. A disaster monitoring and assessment system integrating multi-source heterogeneous remote sensing and collaborative analysis, characterized in that, include: Data layer, application layer, and user layer; The data layer receives and stores image data of various disasters in folders; The application layer includes a sentinel service module and three assessment service modules; the assessment service modules include an earthquake assessment service module, a flood analysis service module, and a fire analysis service module. The sentinel service module monitors the entry of image data into the directory of each folder. When it detects the creation of a new image file in a folder, the sentinel service module calls the corresponding assessment service module to perform disaster assessment according to the task corresponding to the folder, receives the analysis feedback from each assessment service module, and sends the disaster assessment results to the user layer. The earthquake assessment service module extracts pre-disaster optical images and post-disaster SAR images from the data layer, extracts the same type of statistical features from the two types of images and calculates the differences, filters earthquake disaster patches based on the intensity of statistical feature differences, and generates an earthquake disaster assessment result file. The flood analysis service module extracts pre-disaster and post-disaster SAR images from the data layer, extracts pre-disaster and post-disaster water body masks, uses Boolean logic difference operations on the two water body masks, determines the newly added water body areas that appear after the disaster as the flood inundation areas, and generates a flood disaster assessment result file. The fire analysis service module extracts multispectral images from the data layer, performs fire point detection and fire zone extraction, and generates a fire assessment result file. The user layer includes a client application used to display disaster assessment results.

2. The disaster monitoring and assessment system integrating multi-source heterogeneous remote sensing and collaborative analysis as described in claim 1, characterized in that, The earthquake assessment service module includes a heterogeneous image preprocessing module, a multi-feature vectorization extraction module, an adaptive threshold segmentation module, a morphological postprocessing module, and an earthquake disaster assessment result output module; The heterogeneous image preprocessing module is used to preprocess pre-disaster optical images and post-disaster SAR images. The multi-feature vectorization extraction module is used to extract feature vectors composed of multiple statistical features from the pre-processed pre-disaster optical images and post-disaster SAR images; calculate the absolute difference between the two sets of feature vectors before and after the disaster and obtain the mean value of the vector elements to obtain the intensity of the pre-disaster and post-disaster feature differences corresponding to each pixel. The intensity difference in pre- and post-disaster features of each pixel is used to form a single-channel intensity change map; The adaptive threshold segmentation module is used to perform adaptive threshold segmentation on the change intensity map using the percentile method to obtain change patches that may be earthquake disaster areas. The morphological post-processing module is used to calculate the aspect ratio of the minimum bounding rectangle of the changed patches, exclude changed patches with aspect ratios greater than a set aspect ratio threshold, and use the remaining changed patches as identified earthquake disaster areas. The earthquake disaster assessment result output module is used to encapsulate the identified earthquake disaster areas and their attributes into a file for output.

3. The disaster monitoring and assessment system integrating multi-source heterogeneous remote sensing and collaborative analysis as described in claim 2, characterized in that, The preprocessing of the heterogeneous image preprocessing module is as follows: for pre-disaster optical images, a contrast-limited adaptive histogram equalization method is used to enhance texture; for post-disaster SAR images, a Lee filter is used to suppress speckle noise.

4. The disaster monitoring and assessment system integrating multi-source heterogeneous remote sensing and collaborative analysis as described in claim 2, characterized in that, The multi-feature vectorization extraction module extracts a feature vector composed of multiple statistical features: extracting the mean, standard deviation, gradient mean, and gradient standard deviation within the neighborhood of each pixel to form a feature vector.

5. The disaster monitoring and assessment system integrating multi-source heterogeneous remote sensing and collaborative analysis as described in claim 1, characterized in that, The flood analysis service module includes a water body extraction module, a water body mask time series difference module, and a flood disaster assessment result output module; The water extraction module is used to extract water bodies from SAR images before and after the disaster in parallel, and to obtain the pre-disaster water body mask Mask_pre and the post-disaster water body mask Mask_post. The water body mask timing difference module is used to perform Boolean logic difference operations to obtain newly added water body areas that only appear after the disaster, i.e., the identified flood-inundated areas; the Boolean logic difference operation is as follows: Mask_post AND (NOT Mask_pre); Where AND represents the Boolean intersection operation and NOT represents the Boolean negation operation; The flood disaster assessment result output module is used to encapsulate the identified flood-inundated areas and their attributes into a file for output.

6. The disaster monitoring and assessment system integrating multi-source heterogeneous remote sensing and collaborative analysis as described in claim 5, characterized in that, The water extraction module extracts water in the following way: Threshold segmentation: Pixels in SAR images below the backscattering threshold are initially identified as water bodies to obtain the first water body mask; Morphological closing operation: A rectangular structuring element is used to perform a morphological closing operation on the first water body mask that has been initially identified in order to fill the internal voids and obtain the second water body mask. Connected component denoising: Connected component analysis is performed on the second water body mask to remove all patches with an area smaller than a preset pixel threshold, resulting in the third water body mask. The third water body mask obtained based on pre-disaster SAR images is called the pre-disaster water body mask Mask_pre, and the third water body mask obtained based on post-disaster SAR images is called the post-disaster water body mask Mask_post.

7. The disaster monitoring and assessment system integrating multi-source heterogeneous remote sensing and collaborative analysis as described in claim 1, characterized in that, The fire analysis service module includes a fire point detection module and a fire zone extraction module; The fire detection module is used to extract single-period multispectral images from the data layer, including short-wave infrared band B12, near-infrared band B8A, and red band B4. First, based on a first combination of conditions—both the near-infrared band B8A and the short-wave infrared band B12 values ​​being below a first reflectivity threshold—water bodies are identified. Based on a second combination of conditions—the red band B4 value being greater than a set brightness threshold and the short-wave infrared band B12 value being lower than an adaptive high-temperature threshold T_adaptive_hot—clouds are identified. A cloud-water mask is constructed based on the identified water bodies and clouds. Next, for areas not covered by the cloud-water mask, fire zones are identified based on a third combination of conditions: short-wave infrared band B12 being higher than the adaptive high-temperature threshold T_adaptive, the fire index FI = B12 / B8A being higher than the adaptive fire index threshold FI_adaptive, and the fire index FI being higher than the neighborhood threshold with a difference greater than a set difference threshold. The identified fire zones are then packaged into a file for output. The adaptive high-temperature threshold T_adaptive is determined based on the statistical characteristics of the multispectral images. The burned area extraction module is used to extract post-disaster multispectral images from the data layer, including shortwave infrared band B12, near-infrared band B8A, red band B4, yellow band B6, and red-edge band B7; calculate the normalized rate of fire (NBR), normalized rate of vegetation (NDVI), and scorch area index (BAIS2); use the Otsu adaptive thresholding algorithm to calculate the BAIS2 threshold that distinguishes between scorched and unscorched areas; if the BAIS2 value of a pixel is greater than the BAIS2 threshold, the NBR value is lower than a preset burning characteristic threshold, and the NDVI value is lower than a preset vegetation mortality threshold, then the pixel is identified as a burned area pixel; and the identified burned areas are packaged into a file for output.

8. The disaster monitoring and assessment system integrating multi-source heterogeneous remote sensing and collaborative analysis as described in claim 1, characterized in that, The data layer provides folders for storing images for the earthquake assessment service module, flood analysis service module, and fire analysis service module, respectively. A first folder is set up for the earthquake assessment service module to store pre-disaster optical images and post-disaster SAR images to be assessed; when the data in the first folder changes, the sentinel service module calls the earthquake assessment service module to perform earthquake disaster assessment. Two folders are set up for the flood analysis service module to store SAR images from two periods. In the post-flood assessment application, when the data in the two folders changes, the sentinel service module calls the flood analysis service module to perform flood disaster assessment. In the real-time flood monitoring application, the second folder corresponds to the historical time step, and the third folder corresponds to the current time step. When the time step changes, the latest SAR image in the third folder is stored in the second folder, and the SAR image of the new time step is received and stored in the third folder. When the data in the third folder changes, the Sentinel Service Module calls the Flood Analysis Service Module to perform a flood disaster assessment and waits for the flood disaster assessment result file returned by the Flood Analysis Service Module. If the newly added water area exceeds the preset value, a flood disaster alarm message is sent to the client. The fire analysis service module is configured with a fourth and a fifth folder. The fourth folder stores multispectral images of fire detection, and the fifth folder stores multispectral images after a fire. In the post-fire assessment application, when the data in the fourth folder changes, the sentinel service module calls the fire analysis service module to detect fire points; when the data in the fifth folder changes, the sentinel service module calls the fire analysis service module to extract the burned area. In the real-time fire monitoring application, the data in the fourth folder is updated periodically. When the data in the fourth folder changes, the sentinel service module calls the fire analysis service module to detect fire points and waits for the fire assessment result file returned by the fire analysis service module. If a fire point is identified, a fire alarm message is sent to the client.

9. The disaster monitoring and assessment system integrating multi-source heterogeneous remote sensing and collaborative analysis as described in claim 1, characterized in that, The Sentinel Service Module and the Evaluation Service Module use an HTTP connection. The absolute path of the file containing the image data to be evaluated in the folder is used as a parameter, and the corresponding evaluation service module is automatically invoked through an HTTP POST request. The application layer sets up a notification API, which maintains a real-time communication channel with the client via a WebSocket long connection. After receiving the analysis feedback from the evaluation service module, the sentinel service module confirms that a new disaster has occurred, and then constructs an alarm message containing key summary information and sends it to the notification API. The notification API broadcasts alarm messages to all online clients in real time via a WebSocket long connection, enabling proactive early warning.

10. The disaster monitoring and assessment system integrating multi-source heterogeneous remote sensing and collaborative analysis as described in claim 9, characterized in that, The client includes an alarm management module, a 3D visualization service engine, and a statistics module; Upon receiving an alarm message, the alarm management module triggers a visual notification to alert the user that a new disaster event has occurred; upon receiving a user instruction, it retrieves the disaster assessment result file associated with the alarm message and starts the 3D visualization service engine. The 3D visualization service engine has a built-in style rule set that automatically matches thematic colors to the patches in the disaster area of ​​the disaster assessment result file based on the task type, and then renders the patches. Then, the 3D camera is driven to fly and automatically focus on the disaster area; The statistics module iterates through the information in the disaster assessment result file, automatically calculates statistical indicators including the total area of ​​the disaster area and the number of map patches, and displays them in the statistics panel on the client display interface.

Images