Disaster area surveying and mapping and deep learning analysis system based on cooperation of low-altitude unmanned aerial vehicle and satellite remote sensing

Abstract

The invention discloses a disaster area surveying and mapping and deep learning analysis system based on cooperation of a low-altitude unmanned aerial vehicle and satellite remote sensing, and belongs to the field of surveying and mapping and deep learning, and the system comprises the steps: carrying out the preliminary cruise scanning of a suspected disaster area through a low-altitude unmanned aerial vehicle platform, and obtaining high-resolution local image data from the suspected disaster area; marking abnormal features identified in the image to generate preliminary disaster distribution information; according to the real-time change information, in combination with time sequence image data of satellite remote sensing, calculating the dynamic evolution trend of the disaster hotspot by applying a long-short term memory network time sequence analysis algorithm, and determining the evolution direction and speed; and constructing a disaster diffusion prediction model based on data driving according to the evolution direction and speed data, adjusting an unmanned aerial vehicle cruising route and a satellite monitoring key area according to a possible diffusion path identified in a prediction path distribution diagram, and continuously tracking key nodes on the path to obtain updated monitoring data.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of surveying and mapping and deep learning, and particularly relates to a disaster area surveying and deep learning analysis system based on low-altitude unmanned aerial vehicle and satellite remote sensing cooperation. BACKGROUND

[0002] In the field of disaster emergency management, the surveying and analysis method based on low-altitude unmanned aerial vehicle and satellite remote sensing cooperation has crucial strategic value. This field provides indispensable support for disaster warning, response, and recovery by integrating various technical means, especially showing great potential in quickly locating disaster areas and accurately assessing disaster conditions. However, current research and application still face many challenges, and innovation breakthroughs are urgently needed to improve the efficiency and accuracy of disaster management.

[0003] Existing methods have significant limitations in disaster monitoring and response, mainly in the lack of ability to identify and track dynamic changes in disaster hotspots. Many solutions often rely on a single data source or fixed monitoring periods, making it difficult to adapt to the complexity and suddenness of disasters, especially in multi-source information fusion and real-time analysis. This limitation leads to the inability to capture the dynamic evolution trend of disaster hotspots in a timely manner, thereby affecting the accuracy and timeliness of emergency decision-making. SUMMARY

[0004] To solve the above technical problems, the present application provides a disaster area surveying and deep learning analysis system based on low-altitude unmanned aerial vehicle and satellite remote sensing cooperation, comprising:

[0005] A preliminary monitoring module for preliminary scanning of suspected disaster areas by low-altitude unmanned aerial vehicles, obtaining local image data and marking abnormal features, and generating preliminary disaster distribution information;

[0006] A wide-area monitoring module for continuously observing a larger area using satellite remote sensing based on preliminary disaster distribution information, obtaining long-time sequence image data, and determining the overall scope of the disaster;

[0007] A data fusion module for fusing satellite remote sensing data and unmanned aerial vehicle data to generate a comprehensive disaster distribution view;

[0008] A hotspot division module for dividing disaster hotspot areas based on the comprehensive view, extracting key feature point data, and determining potential high-risk areas;

[0009] A risk triggering module for triggering unmanned aerial vehicles to dynamically cruise high-risk areas and obtain real-time change information if the key feature point data exceeds a preset threshold;

[0010] An evolution analysis module for analyzing the dynamic evolution trend of disaster hotspots by combining real-time change information and time series data.

[0011] a diffusion prediction module, configured to construct a disaster diffusion prediction model according to the evolution trend, and generate a possible diffusion path;

[0012] a path adjustment module, configured to adjust the monitoring path of the unmanned aerial vehicle and the satellite according to the diffusion path, and continuously track the key node;

[0013] a prediction optimization module, configured to optimize the disaster diffusion prediction result by using the updated monitoring data, and update the comprehensive disaster distribution view.

[0014] Compared with the prior art, the present application has the following advantages and technical effects:

[0015] The application discloses a disaster monitoring and prediction method based on multi-source data fusion, which obtains high-resolution image data of a disaster area through a low-altitude unmanned aerial vehicle and a satellite remote sensing system, generates a comprehensive disaster distribution view by using a multi-source data fusion technology, divides a disaster hotspot area by applying a support vector machine and a spatial clustering method, analyzes a disaster dynamic evolution trend in combination with a long short-term memory network, constructs a data-driven disaster diffusion prediction model, generates a possible diffusion path through Monte Carlo simulation, and dynamically adjusts a monitoring strategy according to a prediction result, and continuously optimizes the prediction result. BRIEF DESCRIPTION OF DRAWINGS

[0016] The accompanying drawings, which form a part of this application, are intended to provide further understanding of the application and are incorporated herein for a purpose of explanations and are in no way limiting on the present application. In the drawings:

[0017] Figure 1 It is a system structure schematic diagram of the embodiment of the present application. DETAILED DESCRIPTION

[0018] It should be noted that the embodiments and the features in the embodiments in the present application can be combined with each other without conflict. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0019] It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a group of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described herein can be executed in an order different from that shown herein.

[0020] Embodiment one

[0021] As Figure 1As shown, the embodiment provides a disaster area mapping and deep learning analysis system based on low-altitude unmanned aerial vehicle and satellite remote sensing cooperation, comprising:

[0022] A preliminary monitoring module is configured to perform preliminary scanning on a suspected disaster area by a low-altitude unmanned aerial vehicle, acquire local image data and mark abnormal features, and generate preliminary disaster distribution information;

[0023] A wide-area monitoring module is configured to perform continuous observation on a larger range of areas by satellite remote sensing according to the preliminary disaster distribution information, acquire long-time sequence image data, and determine the overall range of the disaster;

[0024] A data fusion module is configured to fuse satellite remote sensing data and unmanned aerial vehicle data to generate a comprehensive disaster distribution view;

[0025] A hotspot division module is configured to divide disaster hotspot areas based on the comprehensive view, extract key feature point data, and determine potential high-risk areas;

[0026] A risk triggering module is configured to trigger the unmanned aerial vehicle to perform dynamic cruising on the high-risk area to acquire real-time change information if the key feature point data exceeds a preset threshold;

[0027] An evolution analysis module is configured to analyze the dynamic evolution trend of the disaster hotspot in combination with the real-time change information and time sequence data;

[0028] A diffusion prediction module is configured to construct a disaster diffusion prediction model according to the evolution trend to generate possible diffusion paths;

[0029] A path adjustment module is configured to adjust the monitoring paths of the unmanned aerial vehicle and the satellite according to the diffusion paths to continuously track the key nodes;

[0030] A prediction optimization module is configured to optimize the disaster diffusion prediction results by using updated monitoring data and update the comprehensive disaster distribution view.

[0031] Further, the preliminary monitoring module comprises:

[0032] An unmanned aerial vehicle scanning unit is configured to acquire high-resolution local image data by a low-altitude unmanned aerial vehicle carrying scanning technology and store the data as an initial image set;

[0033] An image preprocessing unit is configured to perform denoising and enhancement processing on the initial image set to obtain a clear image set;

[0034] A feature recognition unit is configured to recognize abnormal features in the clear image set by using a pre-established convolutional neural network model and output an abnormal feature distribution map;

[0035] a deep analysis unit configured to, if a significant abnormal area is detected in the abnormal feature distribution map, perform deep analysis on image data of the area to determine a specific category of the abnormal feature;

[0036] a mapping generation unit configured to generate a preliminary mapping result of disaster distribution in combination with the specific category of the abnormal feature and the distribution map;

[0037] a path adjustment unit configured to, according to a key area of the disaster distribution, adjust a cruising path of the low-altitude unmanned aerial vehicle to collect more detailed image data and obtain a supplementary image set;

[0038] an information updating unit configured to update the abnormal feature distribution map and the mapping result of disaster distribution according to the supplementary image set to generate final disaster distribution information.

[0039] The low-altitude unmanned aerial vehicle is used to preliminarily cruise the disaster area by a scanning technology to obtain high-resolution local image data and store the data as an initial image set. According to the initial image set, an image preprocessing technology is used to perform denoising and enhancement processing on the local image to obtain a clear image set. For the clear image set, a pre-established convolutional neural network model is used to identify abnormal features therein to output an abnormal feature distribution map. If a significant abnormal area is detected in the abnormal feature distribution map, deep analysis is performed on image data of the area to determine a specific category of the abnormal feature. Through the specific category of the abnormal feature and the distribution map, a preliminary mapping result of disaster distribution is generated in combination with regional coverage information. After the preliminary mapping result is obtained, for a key area of the disaster distribution, a cruising path of the low-altitude unmanned aerial vehicle is adjusted to collect more detailed image data and obtain a supplementary image set. According to the supplementary image set, the abnormal feature distribution map and the mapping result of disaster distribution are updated to generate final disaster distribution information.

[0040] When conducting preliminary cruise scanning of suspected disaster areas through low-altitude UAV platforms, UAV equipment equipped with high-precision optical cameras and infrared sensors can be used to conduct grid coverage scanning of target areas under the condition of an altitude of about 100 meters and a flight speed of 5 meters per second, with a scanning grid unit set to 50 meters x 50 meters, ensuring a coverage rate of more than 95% and obtaining high-definition image data with a resolution of 0.1 meters per pixel. Subsequently, deep learning algorithms such as YOLOv5 are used to detect targets in the image data and identify abnormal features in the image, such as cracks, water accumulation, or collapsed structures, with a detection threshold set to 0.8 to reduce false positive rates. Areas with crack widths greater than 0.5 meters in the detection results are marked as high-risk points, and areas with water accumulation exceeding 10 square meters are marked as secondary high-risk points. Next, for the marked abnormal features, spatial analysis is conducted in combination with geographic information systems (GIS), and the abnormal points are superimposed with terrain data to calculate the disaster impact radius, for example, taking the crack point as the center and marking the area within a radius of 50 meters as a potential impact area. Image gray-scale analysis algorithms are also used to evaluate water depth, assuming a linear relationship between gray value and depth, with a depth threshold set to 0.3 meters, and areas exceeding this value have their priority increased. Finally, when generating preliminary disaster distribution information, heat map visualization technology is used to mark high-risk points in red and secondary high-risk points in orange, generating a disaster distribution map with a resolution of 1 meter per pixel, and spatial clustering algorithms (such as DBSCAN) are used to analyze abnormal point density, with a clustering radius set to 20 meters and a minimum number of points set to 5, to identify disaster concentration areas and output structured data files containing coordinates, risk levels, and impact ranges.

[0041] Further, the wide-area monitoring module comprises:

[0042] A satellite data acquisition unit is configured to acquire data of a target area through a satellite remote sensing system, obtain original image data covering disaster distribution, and preliminarily mark abnormal features and feature positions to obtain an initial image set;

[0043] A data preprocessing unit is configured to perform denoising and correction processing on the initial image set to generate a first corrected image set, and perform data integration in the time dimension for long-time sequence characteristics;

[0044] A range determination unit is configured to extract disaster impact features within the region coverage through coverage analysis of a wide-angle view to obtain a disaster impact distribution layer;

[0045] A region expansion unit is configured to, if the range of disaster impact in the distribution layer exceeds a preset region coverage boundary, call the satellite remote sensing system to perform supplementary data acquisition on the expanded region, acquire image data of the new region, and generate an expanded image set;

[0046] The boundary division unit is configured to fuse the extended image set and the first image set, extract features of the fused image data by using a convolutional neural network model, and determine the overall range of disaster impact.

[0047] The trend analysis unit is configured to extract a change trend of disaster distribution by comparing the final range layer and long-time sequence image data, and determine a dynamic evolution feature of disaster impact in a time dimension.

[0048] The monitoring plan generation unit is configured to generate a targeted monitoring plan according to the dynamic evolution feature of the change trend, acquire subsequent image data, and determine a real-time update state of disaster distribution.

[0049] The satellite remote sensing system is used to collect data of a target region, acquire original image data covering disaster distribution, preliminarily label abnormal features and feature positions, and obtain an initial image set containing disaster distribution information. According to the initial image set, an image preprocessing technique is used to denoise and correct the image data, to generate a corrected first image set. The data in a time dimension is integrated according to the long-time sequence characteristic, to determine the continuity of the image data. If the continuity of the first image set in the time dimension meets a preset threshold, disaster impact features within a regional coverage are extracted by coverage analysis of a wide-angle view, to obtain a distribution layer of disaster impact. If the range of disaster impact in the distribution layer exceeds a preset regional coverage boundary, the satellite remote sensing system is called to collect supplementary data of an extended region, to acquire image data of the new region, and generate an extended image set. According to the fusion of the extended image set and the first image set, a convolutional neural network model is used to extract features of the fused image data, to divide the boundary of the overall range of disaster impact, and determine a final range layer of disaster impact. By comparing the final range layer and long-time sequence image data, a change trend of disaster distribution is extracted, and a dynamic evolution feature of disaster impact in a time dimension is determined. According to the dynamic evolution feature of the change trend, a targeted monitoring plan is generated according to the requirement of continuous monitoring, subsequent image data is acquired, and a real-time update state of disaster distribution is determined.

[0050] Based on the marked abnormal feature position in the preliminary disaster distribution information, first, the marked point is spatially positioned by a geographic information system (GIS), assuming that the abnormal feature position is a latitude and longitude coordinate (116.3, 39.9), and taking the point as the center, an initial observation range with a radius of 50 kilometers is set, an interface of a satellite remote sensing system is called, and visible light and infrared image data with a resolution of 1 meter in the region are automatically obtained, the data collection frequency is set to be once every 6 hours, and the continuous observation time is 7 days, forming a time series image data set. Subsequently, using satellite data with a wide field of view, combined with image processing algorithms such as NDVI (normalized vegetation index) calculation, the image data is preliminarily extracted for disaster impact range, the specific calculation formula is NDVI = (NIR-Red) / (NIR+Red), wherein NIR is the near-infrared band value, Red is the red band value, and the region with an NDVI threshold less than 0.2 is set as a possible disaster area, a disaster impact heat map is generated, and through spatial overlay analysis, the heat map is combined with terrain data to exclude non-disaster related areas with an elevation greater than 1000 meters, and the disaster impact range is preliminarily determined to be about 300 square kilometers. Then, for long time series image data, time series analysis algorithms such as periodic change detection based on Fourier transform are used to extract regions with a brightness value change rate exceeding 30% as the core area of disaster dynamic expansion, combined with spatial clustering algorithms (DBSCAN, radius parameter set to 2 kilometers, minimum point number set to 5), the disaster boundary is further refined, and finally the overall disaster impact range is output as 350 square kilometers with an accuracy error controlled within 5%. To ensure the reliability of the results, the system automatically calls a weather data interface to obtain rainfall (assuming an average rainfall of 50 mm) and wind speed (average wind speed of 5 m / s) data during the observation period, and performs correlation analysis with the disaster range changes to verify whether the disaster expansion is consistent with the weather conditions, and if the correlation coefficient is greater than 0.8, the result is confirmed to be reliable.

[0051] Further, the data fusion module comprises:

[0052] a data standardization unit for performing preliminary format conversion and metadata extraction on image data collected by satellite remote sensing and unmanned aerial vehicles to obtain a standardized image data set;

[0053] a spatial registration unit for performing spatial registration processing on the overall image data and the local image data by a geographic information system, performing projection transformation for coordinate system differences, and determining a unified geographic reference frame;

[0054] a bias adjustment unit for finely adjusting the overall image and the local image by a feature point matching method to obtain spatially consistent fused image data if there is a positional bias in the image data after spatial registration;

[0055] a time alignment unit configured to perform time alignment on images of different sources by using timestamp information, and perform time interpolation correction on the image data if the timestamp difference exceeds a preset threshold;

[0056] a pixel fusion unit configured to perform pixel-level fusion on the overall image and the local image by using a weighted average method by applying a multi-source data fusion technique, to obtain comprehensive disaster image data;

[0057] a feature extraction unit configured to extract disaster distribution features from the comprehensive disaster image data, and classify and label abnormal areas in the image to determine the spatial range and intensity level of disaster distribution;

[0058] a view construction unit configured to construct a comprehensive view, and perform visual processing on the disaster distribution by using a layered rendering technique to generate a multi-dimensional disaster distribution view.

[0059] Overall image data and local image data are respectively acquired by satellite remote sensing and unmanned aerial vehicle collection, preliminary format conversion and metadata extraction are performed on the two data sources to obtain a standardized image data set. According to the standardized image data set, spatial registration processing is performed on the overall image data and the local image data by using a geographic information system, projection transformation is performed on the coordinate system difference to determine a unified geographic reference framework. If there is a positional deviation in the spatially registered image data, the overall image and the local image are finely adjusted by a feature point matching method to obtain spatially consistent fused image data. For the fused image data, time alignment is performed on images of different sources by using timestamp information, and time interpolation correction is performed on the image data if the timestamp difference exceeds a preset threshold, to determine the time-synchronized image sequence. According to the time-synchronized image sequence, multi-source data fusion technology is applied, and pixel-level fusion is performed on the overall image and the local image by using a weighted average method to obtain comprehensive disaster image data. Disaster distribution features are extracted from the comprehensive disaster image data, and abnormal areas in the image are classified and labeled to determine the spatial range and intensity level of disaster distribution. After obtaining the spatial range and intensity level of disaster distribution, a comprehensive view is constructed, and visual processing is performed on the disaster distribution by using a layered rendering technique to generate a multi-dimensional disaster distribution view.

[0060] For the technical implementation of the fusion of overall range image data acquired by satellite remote sensing and local image data acquired by unmanned aerial vehicles, the following specific methods can be used. First, assuming that the satellite image covers an area of 500 square kilometers with a resolution of 10 meters per pixel and a data format of GeoTIFF, containing three bands of red, green, and blue, while the unmanned aerial vehicle image covers an area of 5 square kilometers with a resolution of 0.5 meters per pixel and a data format of GeoTIFF. When performing spatial registration through a geographic information system, a feature point-based registration algorithm such as SIFT (Scale-Invariant Feature Transform) can be used to extract key points from both types of images, calculate feature descriptors, match similar point pairs, set the matching threshold to 0.8, and after filtering out reliable matching points, use the least squares method to estimate the affine transformation matrix to ensure that the registration error is controlled within 2 meters, generating image data in a unified geographic coordinate system. Next, time alignment is completed using timestamps, assuming that the satellite image was collected at 08:00:00 on October 1, 2023, and the unmanned aerial vehicle image was collected at 09:30:00 on October 1, 2023, with a time difference of 1.5 hours. The data is aligned using a time interpolation algorithm, combined with a disaster dynamic change model to predict the disaster expansion trend within 1.5 hours, with an error controlled within 5%, ensuring time consistency. Subsequently, for multi-source data fusion, a weighted average-based pixel-level fusion method can be used, setting the satellite image weight to 0.4 and the unmanned aerial vehicle image weight to 0.6, to generate a comprehensive image. The impact of the disaster on vegetation is analyzed using NDVI (Normalized Difference Vegetation Index), with the formula NDVI = (NIR-Red) / (NIR+Red), where NIR is the near-infrared band value and Red is the red band value. The analysis result shows that the NDVI value in the disaster area drops below 0.2, indicating that the vegetation is severely damaged. Finally, when generating a comprehensive disaster distribution view, based on the fused image, a K-means clustering algorithm is used to divide the disaster area into three levels of light, moderate, and severe, setting the number of cluster centers K = 3 and the number of iterations to 10, generating a disaster distribution heat map. Combined with the GIS platform, administrative boundaries and road data are superimposed to output a disaster distribution view with a resolution of 1 meter per pixel, visually displaying the affected area and extent.

[0061] Further, the hotspot division module includes:

[0062] The classification processing unit is configured to obtain preliminary regional distribution information from the disaster distribution data by comprehensive view analysis, and process the data using a support vector machine classification algorithm to obtain classified disaster distribution categories.

[0063] The clustering analysis unit is configured to group the regions according to the classified disaster distribution categories using a spatial clustering method to determine the boundary range of the disaster hotspot area.

[0064] The feature extraction unit is configured to extract key feature data in the determined disaster hotspot region boundary range, obtain a distribution of feature point numbers, and determine a concentration degree of potential risks.

[0065] The risk marking unit is configured to mark the hotspot region as a high-risk area if the distribution of feature point numbers exceeds a preset threshold, and obtain a marked region data set.

[0066] The risk analysis unit is configured to analyze spatial correlation between potential risks and the high-risk area based on the marked region data set, and determine a risk level distribution map.

[0067] The region optimization unit is configured to optimize region determination based on the risk level distribution map and dynamic changes in disaster distribution, and obtain a final high-risk region division result.

[0068] The sorting generation unit is configured to generate a spatial distribution priority ranking based on the final high-risk region division result, and determine a distribution position of a key monitoring region.

[0069] Through comprehensive view analysis, preliminary region distribution information is obtained from disaster distribution data. A support vector machine classification algorithm is used to process the data to obtain classified disaster distribution categories. According to the classified disaster distribution categories, a spatial clustering method is applied to group process the regions to determine the boundary range of the disaster hotspot region. Key feature data in the determined disaster hotspot region boundary range is extracted, a distribution of feature point numbers is obtained, and a concentration degree of potential risks is determined. If the distribution of feature point numbers exceeds a preset threshold, the hotspot region is marked as a high-risk area, and a marked region data set is obtained. Spatial correlation between potential risks and the high-risk area is analyzed based on the marked region data set, and a risk level distribution map is determined. The risk level distribution map is used to optimize region determination in combination with dynamic changes in disaster distribution, and a final high-risk region division result is obtained. The final high-risk region division result is used to generate a spatial distribution priority ranking, and a distribution position of a key monitoring region is determined.

[0070] Based on the comprehensive disaster distribution view, the support vector machine classification algorithm is combined with the spatial clustering method to automatically divide the disaster hot spot area, and the key feature point data is extracted to determine the specific implementation method of the potential high-risk area as follows. First, assuming that we have a spatial data set containing 1000 disaster occurrence points, each point contains latitude and longitude coordinates and disaster intensity value (such as intensity range 1 to 10), the data is classified by support vector machine algorithm, the kernel function is set as radial basis function (RBF), the parameter C is set as 1.0, and the gamma is set as 0.1, the disaster points are divided into high-risk and low-risk two categories by the algorithm, and the classification result shows that about 600 points are classified as high-risk. Then, the high-risk points after classification are input into the DBSCAN spatial clustering algorithm, the neighborhood radius Eps is set as 0.5 kilometers, and the minimum point number MinPts is set as 5, the clustering result generates 3 main hot spot areas, containing 200, 250 and 150 disaster points respectively. Then, the key feature point data is extracted from each hot spot area, the center point coordinates and the average disaster intensity of each clustering area are calculated, for example, the center point of the first area is (latitude 39.904, longitude 116.407), and the average intensity is 8.5, and the density of points in the area (about 50 points per square kilometer) is combined as a feature value. Finally, the risk assessment model is constructed by using these feature values, and the risk index is obtained by weighted calculation (center point intensity accounts for 60%, and density accounts for 40%), the risk index of the first area is 8.5x0.6+50x0.4=7.1, and it is judged as a potential high-risk area.

[0071] Further, the risk triggering module comprises:

[0072] The data processing unit is used for acquiring feature data from the monitoring system for key features, and performing cleaning and standardization processing on the collected information by the pre-established data processing module to obtain a standardized feature data set;

[0073] The threshold judgment unit is used for determining that there is a potential disaster risk in the current area if the value in the feature data set exceeds the risk threshold range according to the standardized feature data set by using the preset threshold judgment logic, and outputting a high-risk area identifier;

[0074] The cruise triggering unit is used for starting the unmanned aerial vehicle cruise system for the high-risk area identifier, executing a dynamic cruise task, and acquiring real-time change information of the target area through the high-definition camera and sensor carried to generate a dynamic monitoring data stream;

[0075] The feature analysis unit is used for analyzing the collected video and image by using image processing technology according to the dynamic monitoring data stream, extracting the feature change related to the disaster risk, and obtaining a real-time risk feature set;

[0076] a risk assessment unit configured to classify and evaluate feature changes using a support vector machine algorithm for a real-time risk feature set, and if the classification result shows that the risk features continue to intensify, determine that the current area needs further monitoring, and output an urgent monitoring instruction;

[0077] a path adjustment unit configured to adjust the UAV cruising path and frequency according to the urgent monitoring instruction, obtain more detailed regional monitoring data by increasing the cruising density and data acquisition frequency, and determine the risk evolution trend;

[0078] a trend analysis unit configured to analyze the trend using a pre-established risk prediction model in combination with historical data and real-time monitoring data for the risk evolution trend, determine the development direction of the potential disaster, and generate a dynamic risk assessment result.

[0079] For key features, feature data is obtained from the monitoring system, and the collected information is cleaned and standardized by a pre-established data processing module to obtain a standardized feature data set. According to the standardized feature data set, a pre-set threshold judgment logic is used, and if the values in the feature data set exceed the risk threshold range, it is determined that there is a potential disaster risk in the current area, and a high-risk area identifier is output. For the high-risk area identifier, the UAV cruising system is started to perform a dynamic cruising task, and real-time change information of the target area is obtained through the high-definition camera and sensors carried to generate a dynamic monitoring data stream. According to the dynamic monitoring data stream, the collected video and images are analyzed using image processing technology to extract feature changes related to disaster risk to obtain a real-time risk feature set. For the real-time risk feature set, a support vector machine algorithm is used to classify and evaluate feature changes, and if the classification result shows that the risk features continue to intensify, it is determined that the current area needs further monitoring, and an urgent monitoring instruction is output. According to the urgent monitoring instruction, the UAV cruising path and frequency are adjusted, more detailed regional monitoring data is obtained by increasing the cruising density and data acquisition frequency, and the risk evolution trend is determined. For the risk evolution trend, a pre-established risk prediction model is used to analyze the trend in combination with historical data and real-time monitoring data, the development direction of the potential disaster is determined, and a dynamic risk assessment result is generated.

[0080] In the automated process of disaster risk assessment and UAV cruising, first, the key feature point data is collected through the sensor network, for example, assuming that the soil humidity sensor of a certain area collects data every hour, the current value is 85%, and the preset disaster risk threshold is 70%, the system automatically calculates the difference 85%-70%=15% through the comparison algorithm, and combines the historical data analysis to obtain the abnormal humidity growth rate of 20%, which exceeds the normal fluctuation range of 10%, and determines it as a high-risk signal. Then, the system matches the data with the preset threshold database, uses the decision tree algorithm to classify the risk level, and if the result is "high risk", it automatically generates a trigger instruction, which includes the target area coordinates (such as latitude 39.9042, longitude 116.4074) and the cruising radius of 500 meters. Subsequently, the system transmits the instruction to the UAV control platform, and the UAV optimizes the flight route according to the preset path planning algorithm (such as A* algorithm) to avoid obstacles, and cruises at a speed of 3 meters / second in the high-risk area, while carrying a high-definition camera and an infrared sensor to collect image and temperature data in real time, for example, detecting a local temperature anomaly of 45 degrees Celsius, which is 10 degrees Celsius higher than the normal value. The collected data is transmitted back to the cloud analysis system in real time through the 5G network, and the system uses a deep learning model (such as convolutional neural network CNN) to extract features from the image, identify potential cracks or water accumulation areas, and calculate the change rate, for example, the crack width increases from 0.5 centimeters to 1.2 centimeters, with a change rate of 140%, thereby generating a dynamic risk report. Finally, the system compares the analysis results with the historical disaster database, and if the similarity exceeds 80%, the risk warning level is automatically updated.

[0081] Further, the evolution analysis module comprises:

[0082] A data preprocessing unit is configured to acquire image data of a disaster area through a satellite remote sensing system, preprocess time series images to obtain a preliminarily arranged image data set, and determine a clear image set after processing.

[0083] A quality improvement unit is configured to improve the quality of the image data by using a denoising technology according to the preliminarily arranged image data set, and determine the clear image set after processing.

[0084] A feature extraction unit is configured to extract spatial distribution features of disaster hotspots from the clear image set after processing, and obtain initial distribution information of the disaster hotspots.

[0085] An evolution analysis unit is configured to analyze the dynamic changes of the disaster hotspots by using a long short-term memory network algorithm according to the initial distribution information and real-time information updates, and determine a characteristic value of the evolution trend.

[0086] A directional decomposition unit is configured to perform vector decomposition on the change direction to obtain detailed components of the change direction if the characteristic value of the evolution trend exceeds a preset threshold.

[0087] a speed calculation unit configured to calculate the distribution of the change speed according to the detailed components of the change direction, and determine the dynamic evolution rate of the disaster hot spot;

[0088] a prediction result unit configured to fuse the continuous image data of the time sequence according to the dynamic evolution rate, and obtain the long-term evolution prediction result of the disaster hot spot.

[0089] Image data of a disaster area is obtained through a satellite remote sensing system, and the time sequence image is preprocessed to obtain a preliminarily sorted image data set. According to the preliminarily sorted image data set, a denoising technology is used to improve the quality of the image data, and a clear image set after processing is determined. According to the clear image set after processing, the spatial distribution characteristics of the disaster hot spot are extracted, and initial distribution information of the disaster hot spot is obtained. Through the initial distribution information, combined with real-time information updating, a long short-term memory network algorithm is used to analyze the dynamic change of the disaster hot spot, and a characteristic value of the evolution trend is judged. If the characteristic value of the evolution trend exceeds a preset threshold value, the change direction is decomposed into a vector to obtain detailed components of the change direction. According to the detailed components of the change direction, the distribution of the change speed is calculated, and the dynamic evolution rate of the disaster hot spot is determined. According to the dynamic evolution rate, the continuous image data of the time sequence is fused to obtain the long-term evolution prediction result of the disaster hot spot.

[0090] By satellite remote sensing technology to obtain time series image data of disaster hotspots, assuming we use Landsat-8 satellite thermal infrared band data, time span from January to December 2023, get image once a month, resolution is 30 meters, cover 1000 square kilometers of thermal anomaly point distribution in a forest fire prone area. In the data preprocessing stage, use ENVI software to perform radiation correction and atmospheric correction on the image, ensure that the temperature value accuracy of the thermal anomaly point is within ±0.5 degrees Celsius, then extract the position and intensity of the thermal anomaly point at each time node to form a time series data set. Then, build a long short-term memory network (LSTM) model, use the TensorFlow framework of Python, set 3 layers of LSTM network, each layer has 64 hidden units, input data is the number and position coordinates of thermal anomaly points in the past 6 months, output is the prediction value of the next month, training data set accounts for 80%, validation set accounts for 20%, iterate 1000 times through Adam optimizer, loss function is mean square error (MSE), the final model prediction accuracy reaches 85%. In the dynamic evolution trend analysis, calculate the spatial moving track of the thermal anomaly point, find that the center of the hot spot moves about 2.5 kilometers northeast every month, combined with the temperature intensity change (monthly average rise of 0.8 degrees Celsius), infer that the fire risk is intensifying. Further, use spatial interpolation algorithm (Kriging method) to grid the hot spot distribution, generate a continuous thermal map with 10 meter resolution, calculate the evolution speed as an expansion rate of 3.2 square kilometers per month, the direction angle is 45 degrees. Through correlation analysis with historical meteorological data (such as monthly average wind speed of 5 meters per second, wind direction northeast), verify that the evolution trend is highly related to wind driven, form disaster warning basis.

[0091] Further, the diffusion prediction module comprises:

[0092] A data preprocessing unit is configured to obtain evolution direction and speed data, clean and format the original data through the data preprocessing module, and obtain a standardized data set;

[0093] A feature extraction unit is configured to construct a prediction model using a data-driven method for the standardized data set, extract features of the evolution direction and speed data using a random forest algorithm, and determine key influence factors;

[0094] A simulation unit is configured to simulate disaster diffusion behavior multiple times according to the key influence factors and using a Monte Carlo simulation method, and generate multiple possible diffusion paths;

[0095] A path analysis unit is configured to obtain probability density characteristics of path distribution by statistically analyzing the generated multiple diffusion paths, and determine a high-probability diffusion area;

[0096] a path refinement unit configured to, if the probability density of the high-probability diffusion region exceeds a preset threshold, refine simulation of path distribution of the region to obtain a more accurate predicted path set;

[0097] a diagram construction unit configured to construct a distribution diagram module according to the predicted path set, present the path distribution in a visual form, and determine a final disaster diffusion prediction diagram;

[0098] a risk analysis unit configured to obtain a potential diffusion risk region by boundary analysis of the final disaster diffusion prediction diagram, and determine a risk level distribution feature.

[0099] Evolution direction and speed data are obtained, the original data are cleaned and formatted by a data preprocessing module to obtain a standardized data set. A prediction model is constructed by using a data-driven method for the standardized data set, a random forest algorithm is used to extract features of the evolution direction and speed data, and key influence factors are determined. According to the key influence factors, a Monte Carlo simulation method is used to simulate disaster diffusion behavior multiple times to generate multiple possible diffusion paths. The probability density characteristics of the path distribution are obtained by statistical analysis of the generated multiple diffusion paths, and a high-probability diffusion region is determined. If the probability density of the high-probability diffusion region exceeds a preset threshold, the path distribution of the region is refined to obtain a more accurate predicted path set. According to the predicted path set, a distribution diagram module is constructed to present the path distribution in a visual form, and a final disaster diffusion prediction diagram is determined. The potential diffusion risk region is obtained by boundary analysis of the final disaster diffusion prediction diagram, and the risk level distribution feature is determined.

[0100] For the evolution direction and speed data, a data-driven disaster spread prediction model is constructed, and the possible spread path is generated through Monte Carlo simulation, and finally the prediction path distribution map is obtained. We can implement from data processing to result visualization. First, suppose we have collected the evolution data of forest fires in a certain area, including wind speed 5 meters per hour, wind direction 30 degrees north, fire spread speed 2 kilometers per hour, etc. Historical records are removed by data cleaning algorithm to remove outliers (such as wind speed exceeding 50 meters per second is considered abnormal), and interpolation method is used to fill in missing data to ensure data set integrity. Then, when building the prediction model, the random forest algorithm is used to regress the evolution direction and speed, the input features include wind speed, wind direction, terrain slope (assuming an average slope of 15 degrees), and the output is the predicted speed and direction of fire spread. The mean square error of the validation set after model training is 0.25, indicating high prediction accuracy. Subsequently, based on the Monte Carlo simulation method, 10000 random simulations are set, and in each simulation, a random disturbance of wind speed ±1 meter / second and wind direction ±5 degrees is introduced to generate the possible path of fire spread, and the coverage range (assuming an average coverage radius of 10 kilometers) within 24 hours is calculated. In the path generation process, the terrain data is combined to limit the spread range, for example, the fire speed is halved in areas with a slope greater than 30 degrees. Finally, all simulated paths are superimposed to form a probability distribution map, and the heat map visualization technology is used to represent the probability of high and low diffusion in the geographic information system software, with the highest probability (more than 80%) marked in red to help decision makers identify high-risk areas.

[0101] Further, the path adjustment module comprises:

[0102] A path model construction unit is configured to construct an initial path model through a predicted path and a distribution map, and compare the initial path model with a pre-established geographic information database to obtain a preliminary range of potential spread.

[0103] A trajectory adjustment unit is configured to adjust the cruise trajectory of the unmanned aerial vehicle according to the preliminary range of potential spread, set a high-frequency patrol plan for the core node, and obtain real-time image data.

[0104] A data fusion unit is configured to extract key features from the real-time image data, combine the wide-area coverage data monitored by the satellite, fuse the two types of data sources, and determine the dynamic change trend of the core node.

[0105] An optimization triggering unit is configured to trigger an update mechanism of path optimization if the dynamic change trend of the core node exceeds a preset threshold, recalculate the cruise trajectory and the monitoring priority area, and determine the new direction of potential spread.

[0106] A prediction updating unit is configured to track changes in the core node continuously, collect monitoring data in multiple time periods, and use a random forest algorithm to classify the data to obtain a prediction updating result of the diffusion path.

[0107] A monitoring adjusting unit is configured to adjust the key area of satellite monitoring according to the prediction updating result, increase the data collection frequency for the high-risk area, and obtain more detailed monitoring information.

[0108] A scheme updating unit is configured to restart the calculation process of the path optimization if the monitoring information shows that the core node has abnormal fluctuations, update the cruise trajectory in combination with the latest data, and determine the subsequent monitoring scheme.

[0109] An initial path model is constructed through a prediction path and a distribution diagram, a pre-established geographic information database is used for comparison, and a preliminary range of potential diffusion is obtained. According to the preliminary range of potential diffusion, the cruise trajectory of the unmanned aerial vehicle is adjusted, a high-frequency patrol plan is set for the core node, and real-time image data is obtained. Key features are extracted from the real-time image data, combined with wide-area coverage data of satellite monitoring, and two types of data sources are fused to determine the dynamic change trend of the core node. If the dynamic change trend of the core node exceeds a preset threshold, an update mechanism of path optimization is triggered, the cruise trajectory and the monitoring priority area are recalculated, and the new direction of potential diffusion is determined. Through continuous tracking of changes in the core node, monitoring data in multiple time periods is collected, and a random forest algorithm is used to classify the data to obtain a prediction updating result of the diffusion path. According to the prediction updating result, the key area of satellite monitoring is adjusted, the data collection frequency for the high-risk area is increased, and more detailed monitoring information is obtained. If the monitoring information shows that the core node has abnormal fluctuations, the calculation process of the path optimization is restarted, the cruise trajectory is updated in combination with the latest data, and the subsequent monitoring scheme is determined.

[0110] On the basis of the predicted path distribution map, first, through the geographic information system (GIS) combined with historical diffusion data and current environmental variables (such as wind speed 5 meters / second, humidity 60%), the Gaussian diffusion model is used to calculate the possible diffusion path, generate a probability distribution heat map, and identify the core diffusion path area with a probability of more than 80%, for example, the central coordinates of a certain area are (latitude 39.9042, longitude 116.4074), and the path within a radius of 5 kilometers is marked as high priority. Next, based on this heat map, adjust the UAV cruise route, use the A* path planning algorithm to calculate the optimal flight path from the starting point (latitude 39.9000, longitude 116.4000) to the target area, avoid obstacles and optimize the flight distance to 10.5 kilometers, while setting the UAV cruise height to 200 meters and the cruise speed to 15 meters / second, ensuring coverage of 3 nodes on the critical path, with each node staying for 5 minutes to collect image data. Subsequently, these path data are interfaced with the satellite monitoring system, and the key monitoring area is locked through the remote sensing image analysis algorithm (such as NDVI index calculation), and a high-frequency scan is performed on each square kilometer grid in the core area, with a scanning frequency of every 2 hours, and the surface change data (such as a 10% decrease in vegetation coverage) are extracted and correlated with meteorological data to obtain the correlation coefficient between diffusion trend and wind direction of 0.85. Finally, for the key nodes on the path, time series analysis algorithm is used to process the continuous tracking data, taking node 1 as an example, collecting the hourly pollutant concentration data (average value from 5 micrograms / cubic meter to 8 micrograms / cubic meter) in the past 24 hours, and predicting the concentration change trend in the next 12 hours through the ARIMA model, the prediction result shows that the concentration may reach 10 micrograms / cubic meter, and the system automatically generates warning information and updates the monitoring strategy.

[0111] Further, the prediction optimization module comprises:

[0112] a data preprocessing unit for extracting key features from monitoring data, using preprocessing techniques to clean and format the data, and obtaining a structured disaster-related data set;

[0113] a model construction unit for constructing an initial prediction path distribution map based on the structured disaster-related data set and applying a random forest algorithm to model the predicted path of disaster diffusion;

[0114] a comparative analysis unit for comparing the initial prediction path distribution map with the dynamic evolution trend, and if the deviation exceeds the preset threshold, adjusting the prediction path distribution map to obtain an adjusted prediction path distribution map;

[0115] A result updating unit is configured to continuously update the prediction result of disaster spread by adjusting the prediction path distribution map in combination with the latest monitoring data, and determine the matching degree of the prediction result and the actual dynamic evolution;

[0116] A model optimizing unit is configured to optimize the parameter configuration of the disaster spread prediction model according to the matching degree analysis result, and obtain an optimized prediction result data set;

[0117] A view updating unit is configured to obtain the optimized prediction result data set, update the comprehensive distribution view, generate a dynamic visualization layer of disaster distribution, and determine the latest distribution state of the disaster influence range.

[0118] A decision support unit is configured to generate a priority ranking of disaster management decisions in combination with response priorities and data support according to the latest distribution state, and output a reference basis for a targeted response scheme.

[0119] By extracting key features from monitoring data, the data is cleaned and formatted using preprocessing techniques to obtain a structured disaster-related data set. According to the structured disaster-related data set, a random forest algorithm is applied to initially model the prediction path of disaster spread, and an initial prediction path distribution map is determined. The initial prediction path distribution map is compared and analyzed with the dynamic evolution trend. If the comparison result shows that the deviation exceeds the preset threshold, the prediction path distribution map is adjusted to obtain an adjusted prediction path distribution map. The prediction result of disaster spread is continuously updated by the adjusted prediction path distribution map in combination with the latest monitoring data, and the matching degree of the prediction result and the actual dynamic evolution is determined. According to the matching degree analysis result, the parameter configuration of the disaster spread prediction model is optimized to obtain an optimized prediction result data set. The optimized prediction result data set is obtained to update the comprehensive distribution view, generate a dynamic visualization layer of disaster distribution, and determine the latest distribution state of the disaster influence range. According to the latest distribution state, a priority ranking of disaster management decisions is generated in combination with response priorities and data support, and a reference basis for a targeted response scheme is output.

[0120] With the updated monitoring data, the disaster spread prediction results are continuously verified and optimized by comparing with the predicted path distribution map and dynamic evolution trend, and the comprehensive disaster distribution view is updated to provide data support for the next response focus of disaster management. The specific implementation method is as follows: first, assuming that the latest monitoring data obtained through satellite remote sensing and ground sensors show that the flood water level in a certain area has risen by 0.5 meters in the past 24 hours, and the coverage area has expanded from 100 square kilometers to 120 square kilometers, these data are input into the disaster prediction system for processing. The system compares the monitoring data with the predicted path distribution map, and the prediction map is based on historical data and meteorological model (such as LSTM model based on neural network) to estimate that the flood will expand to 150 square kilometers within 48 hours, while the actual monitoring shows that the current expansion speed is slightly lower than the predicted value, therefore the system automatically adjusts the model parameters, reduces the expansion speed weight from 0.8 to 0.75, recalculates the predicted path, and obtains a new predicted coverage area of 140 square kilometers. At the same time, the system analyzes the dynamic evolution trend, combines the wind speed (10 meters per hour) and rainfall (5 millimeters per hour) data, and uses the Monte Carlo simulation algorithm to generate 1000 diffusion scenarios, and obtains that the area with the highest probability of flood diffusion is concentrated in the downstream 10 kilometers, with a probability of 85%. Then, the system fuses the optimized prediction results with the monitoring data, updates the comprehensive disaster distribution view, generates a heat map containing water depth distribution (average 1.2 meters) and affected population (about 5000 people), and automatically labels the high-risk areas. By analyzing the areas with the highest risk level (divided into 1 to 5 levels) in the view, the system automatically generates response focus suggestions, prioritizes resource allocation to the downstream high-risk area, and transmits the data to the disaster management platform through the API interface.

[0121] The above is only a preferred specific embodiment of the present application, but the protection scope of the present application is not limited thereto, any skilled person in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A disaster area mapping and deep learning analysis system based on low-altitude unmanned aerial vehicle and satellite remote sensing cooperation, characterized in that, Comprise: Preliminary monitoring module for preliminary scanning of suspected disaster area by low-altitude unmanned aerial vehicle, obtaining local image data and marking abnormal features, generating preliminary disaster distribution information; Wide area monitoring module for continuously observing larger area by satellite remote sensing according to preliminary disaster distribution information, obtaining long time sequence image data, and determining overall scope of disaster; Data fusion module for fusing satellite remote sensing data and unmanned aerial vehicle data to generate comprehensive disaster distribution view; Hot spot division module for dividing disaster hot spot area based on comprehensive view, extracting key feature point data, and judging potential high-risk area; Risk trigger module for triggering unmanned aerial vehicle to dynamically cruise high-risk area if key feature point data exceeds preset threshold to obtain real-time change information; Evolution analysis module for analyzing dynamic evolution trend of disaster hot spot by combining real-time change information and time sequence data; Diffusion prediction module for constructing disaster diffusion prediction model according to evolution trend to generate possible diffusion path; Path adjustment module for adjusting monitoring path of unmanned aerial vehicle and satellite according to diffusion path to continuously track key nodes; Prediction optimization module for optimizing disaster diffusion prediction result by using updated monitoring data to update comprehensive disaster distribution view.

2. The system of claim 1, wherein, The preliminary monitoring module comprises: Unmanned aerial vehicle scanning unit for obtaining high-resolution local image data by low-altitude unmanned aerial vehicle carrying scanning technology and storing as initial image set; Image preprocessing unit for denoising and enhancing initial image set to obtain clear image set; Feature recognition unit for recognizing abnormal features in clear image set by using pre-established convolutional neural network model to output abnormal feature distribution map; Deep analysis unit for performing deep analysis on image data of significant abnormal area detected in abnormal feature distribution map to determine specific category of abnormal feature; Mapping generation unit for generating preliminary mapping result of disaster distribution by combining specific category of abnormal feature and distribution map; Path adjustment unit for adjusting cruise path of low-altitude unmanned aerial vehicle according to key area of disaster distribution to collect more detailed image data to obtain supplementary image set; Information updating unit for updating abnormal feature distribution map and disaster distribution mapping result according to supplementary image set to generate final disaster distribution information.

3. The system of claim 1, wherein, The wide area monitoring module comprises: Satellite data acquisition unit for data acquisition of target area by satellite remote sensing system to obtain original image data covering disaster distribution, and preliminary labeling of abnormal features and feature positions to obtain initial image set; Data preprocessing unit for denoising and correcting initial image set to generate corrected first image set, and data integration in time dimension for long time sequence characteristics; Scope determination unit for extracting disaster impact features within region coverage by coverage analysis of wide view angle to obtain distribution layer of disaster impact; The regional expansion unit is configured to call a satellite remote sensing system to collect supplementary data of the expanded region and acquire image data of the newly added region to generate an expanded image set if the scope of disaster influence in the distribution layer exceeds the preset regional coverage boundary; The boundary division unit is configured to perform fusion processing on the expanded image set and the first image set, perform feature extraction by using a convolutional neural network model, and determine the overall scope of disaster influence; The trend analysis unit is configured to extract a change trend of disaster distribution by comparing and analyzing the final scope layer and long-time sequence image data, and determine dynamic evolution characteristics of disaster influence in the time dimension; The monitoring plan generation unit is configured to generate a targeted monitoring plan according to the dynamic evolution characteristics of the change trend, acquire subsequent image data, and determine a real-time update state of disaster distribution.

4. The system of claim 1, wherein, The data fusion module includes: The data standardization unit is configured to perform preliminary format conversion and metadata extraction on image data collected by satellite remote sensing and unmanned aerial vehicles to obtain a standardized image data set; The spatial registration unit is configured to perform spatial registration processing on the overall image data and the local image data by using a geographic information system, perform projection transformation on the coordinate system difference, and determine a unified geographic reference framework; The bias adjustment unit is configured to perform fine adjustment on the overall image and the local image by using a feature point matching method to obtain spatially consistent fused image data if there is a position bias in the image data after spatial registration; The time alignment unit is configured to perform time alignment processing on images of different sources by using timestamp information, and perform time interpolation correction on the image data if the time stamp difference exceeds a preset threshold; The pixel fusion unit is configured to apply a multi-source data fusion technology, perform pixel-level fusion on the overall image and the local image by using a weighted average method, and obtain comprehensive disaster image data; The feature extraction unit is configured to extract disaster distribution features from the comprehensive disaster image data, classify and label abnormal regions in the image, and determine the spatial range and intensity level of disaster distribution; The view construction unit is configured to construct a comprehensive view, perform visual processing on disaster distribution by using a layered rendering technology, and generate a multi-dimensional disaster distribution view.

5. The system of claim 1, wherein, The hotspot division module includes: The classification processing unit is configured to obtain preliminary regional distribution information from disaster distribution data by analyzing the comprehensive view, process the data by using a support vector machine classification algorithm, and obtain classified disaster distribution categories; The clustering analysis unit is configured to group regions by using a spatial clustering method according to the classified disaster distribution categories, and determine the boundary range of disaster hot spot regions; The feature extraction unit is configured to extract key feature data in the determined disaster hot spot region boundary range, obtain the distribution of feature points, and determine the concentration degree of potential risks; The risk marking unit is configured to mark the hot spot region as a high-risk area if the distribution of feature points exceeds a preset threshold, and obtain a marked region data set; The risk analysis unit is configured to analyze the spatial correlation between potential risks and high-risk areas by using the marked region data set, and determine a risk level distribution map. The regional optimization unit is configured to optimize the regional judgment according to the risk level distribution map and the dynamic change of the disaster distribution, and obtain a final high-risk region division result. The sorting generation unit is configured to generate a spatial distribution priority ranking by using the final high-risk region division result, and determine the distribution position of the key monitoring region.

6. The system of claim 1, wherein, The risk triggering module comprises: The data processing unit is configured to acquire feature data from the monitoring system for the key features, and perform cleaning and standardization processing on the collected information by using the pre-established data processing module, to obtain a standardized feature data set. The threshold judgment unit is configured to determine that there is a potential disaster risk in the current region and output a high-risk region identifier, if the values in the feature data set exceed the risk threshold range, by using a preset threshold judgment logic according to the standardized feature data set. The cruise triggering unit is configured to start the unmanned aerial vehicle cruise system for the high-risk region identifier, perform a dynamic cruise task, and acquire real-time change information of the target region by using the high-definition camera and the sensor carried thereon, to generate a dynamic monitoring data stream. The feature analysis unit is configured to analyze the collected video and image by using an image processing technology according to the dynamic monitoring data stream, extract feature changes related to the disaster risk, and obtain a real-time risk feature set. The risk assessment unit is configured to classify and evaluate the feature changes by using a support vector machine algorithm for the real-time risk feature set, and output an urgent monitoring instruction if the classification result shows that the risk features continue to intensify. The path adjustment unit is configured to adjust the unmanned aerial vehicle cruise path and frequency according to the urgent monitoring instruction, increase the cruise density and the data acquisition frequency, acquire more detailed regional monitoring data, and determine the risk evolution trend. The trend analysis unit is configured to analyze the risk evolution trend by using a pre-established risk prediction model in combination with the historical data and the real-time monitoring data, judge the development direction of the potential disaster, and generate a dynamic risk assessment result.

7. The system of claim 1, wherein, The evolution analysis module comprises: The data preprocessing unit is configured to acquire image data of the disaster region by using a satellite remote sensing system, pre-process the time series images, and obtain a preliminarily sorted image data set. The quality improvement unit is configured to improve the quality of the image data by using a denoising technology according to the preliminarily sorted image data set, and determine a clear image set after processing. The feature extraction unit is configured to extract spatial distribution features of the disaster hotspots for the clear image set after processing, and acquire initial distribution information of the disaster hotspots. The evolution analysis unit is configured to analyze the dynamic change of the disaster hotspots by using a long short-term memory network algorithm in combination with real-time information update, judge the characteristic value of the evolution trend, and output a change direction vector. The direction decomposition unit is configured to perform vector decomposition on the change direction if the characteristic value of the evolution trend exceeds a preset threshold value, to obtain detailed components of the change direction. The speed calculation unit is configured to calculate the distribution of the change speed according to the detailed components of the change direction, and determine the dynamic evolution rate of the disaster hotspots. The prediction result unit is configured to fuse continuous image data of a time series for a dynamic evolution rate, and obtain a long-term evolution prediction result of a disaster hotspot.

8. The system of claim 1, wherein, The diffusion prediction module comprises: The data preprocessing unit is configured to obtain evolution direction and speed data, clean and format the original data through the data preprocessing module, and obtain a standardized data set; The feature extraction unit is configured to use a data-driven method to construct a prediction model for the standardized data set, extract features of the evolution direction and speed data using a random forest algorithm, and determine key influence factors; The simulation simulation unit is configured to simulate the disaster diffusion behavior multiple times according to the key influence factors and in combination with a Monte Carlo simulation method, and generate multiple possible diffusion paths; The path analysis unit is configured to obtain probability density characteristics of path distribution by statistically analyzing the generated multiple diffusion paths, and determine a high-probability diffusion area; The path refinement unit is configured to refine the path distribution of the high-probability diffusion area if the probability density of the high-probability diffusion area exceeds a preset threshold, and obtain a more accurate prediction path set; The graph construction unit is configured to construct a distribution graph module according to the prediction path set, present the path distribution in a visual form, and determine a final disaster diffusion prediction graph; The risk analysis unit is configured to obtain a potential diffusion risk area by boundary analysis of the final disaster diffusion prediction graph, and determine risk level distribution characteristics.

9. The system of claim 1, wherein, The path adjustment module comprises: The path model construction unit is configured to construct an initial path model through the prediction path and the distribution graph, compare the initial path model with a pre-established geographic information database, and obtain a preliminary range of potential diffusion; The trajectory adjustment unit is configured to adjust the unmanned aerial vehicle cruising trajectory according to the preliminary range of potential diffusion, set a high-frequency patrol plan for the core node, and obtain real-time image data; The data fusion unit is configured to extract key features from the real-time image data, fuse the two types of data sources in combination with wide-area coverage data monitored by a satellite, and determine a dynamic change trend of the core node; The optimization triggering unit is configured to trigger an update mechanism of path optimization if the dynamic change trend of the core node exceeds a preset threshold, recalculate the cruising trajectory and the monitoring priority area, and determine a new direction of potential diffusion; The prediction update unit is configured to track the change of the core node, collect monitoring data of multiple time periods, classify the data using a random forest algorithm, and obtain a prediction update result of the diffusion path; The monitoring adjustment unit is configured to adjust the key area of satellite monitoring according to the prediction update result, increase the data collection frequency for the high-risk area, and obtain more detailed monitoring information; The scheme update unit is configured to restart the calculation process of path optimization if the monitoring information shows that the core node has an abnormal fluctuation, update the cruising trajectory in combination with the latest data, and determine a subsequent monitoring scheme.

10. The system of claim 1, wherein, The prediction optimization module comprises: The data preprocessing unit is configured to extract key features from the monitoring data, clean and format the data using preprocessing techniques, and obtain a structured disaster-related data set; A model construction unit is configured to apply a random forest algorithm to preliminarily model a predicted path of disaster spread according to the structured disaster-related data set, and determine an initial predicted path distribution map; A comparative analysis unit is configured to compare the initial predicted path distribution map with a dynamic evolution trend, and if a comparison result shows that a deviation exceeds a preset threshold, adjust the predicted path distribution map to obtain an adjusted predicted path distribution map; A result updating unit is configured to continuously update a predicted result of disaster spread by using the adjusted predicted path distribution map and combining latest monitoring data, and judge a matching degree of the predicted result and an actual dynamic evolution; A model optimization unit is configured to optimize parameter configuration of a disaster spread prediction model according to a matching degree analysis result, and obtain an optimized predicted result data set; A view updating unit is configured to obtain the optimized predicted result data set, update a comprehensive distribution view, generate a dynamic visualization layer of disaster distribution, and determine a latest distribution state of a disaster influence range; A decision support unit is configured to generate a priority order of disaster management decisions according to the latest distribution state and in combination with response priorities and data support, and output a reference basis of a targeted response scheme.

Images