River Basin Disaster Monitoring and Warning Method and System Based on Multi-source Monitoring Data Analysis

Abstract

The present invention provides a river basin disaster monitoring and early warning method and system based on multi-source monitoring data analysis. By acquiring multi-source monitoring data of the river basin, preprocessing the multi-source monitoring data, eliminating noise interference in the real-time location data, and performing time synchronization alignment on the channel image data and environmental parameter data, a standardized monitoring data set is generated. The dynamic characteristics of water flow, meteorological anomaly characteristics, and channel obstacle distribution characteristics of the river basin are extracted to generate a multi-dimensional disaster-related feature set. The feature set is input into a preset disaster early warning model for dynamic analysis, generating a disaster early warning signal, determining the type of disaster and the scope of impact, triggering the autonomous detachment mechanism of the towed airship and the rapid start command of the drone, broadcasting the early warning information to ships in the channel through the drone, and simultaneously transmitting the disaster type and scope of impact to the command center. The present invention can improve the timeliness of disaster early warning and the spatial adaptation accuracy of disposal measures.

Description

Technical Field

[0001] The present invention relates to the technical field of data monitoring and analysis, and in particular to a river basin disaster monitoring and early warning method and system based on multi-source monitoring data analysis. Background Art

[0002] With the increasing demand for waterway transportation safety, river basin disaster monitoring and early warning technologies are becoming crucial for waterway safety management. Existing technologies typically acquire disaster data through satellite remote sensing, fixed water level sensors, or manual inspections. Pre-set thresholds trigger alarm signals, and human intervention initiates emergency responses. However, these methods have significant limitations: satellite data is prone to blind spots due to cloud cover and limited revisit cycles; fixed sensors cannot cover complex, dynamically changing waterways; and multi-source data collection frequencies and spatial benchmarks make collaborative analysis difficult, making it difficult to identify complex disasters caused by the coupling of meteorological, hydrological, and obstacle factors. Furthermore, traditional early warning mechanisms rely on manual interpretation and response decision-making, which can lead to delays in response to rapidly evolving disasters. Existing hardware systems lack the autonomous disaster avoidance capabilities linked to warning signals. Airships, drones, and other equipment serve merely as independent data collection terminals, failing to form a closed-loop chain from data perception to emergency response execution. This results in insufficient warning timeliness and adaptability of measures. Summary of the Invention

[0003] In view of this, the embodiments of the present invention at least provide a river basin disaster monitoring and early warning method and system based on multi-source monitoring data analysis.

[0004] The technical solution of the embodiment of the present invention is achieved as follows:

[0005] In one aspect, an embodiment of the present invention provides a river basin disaster monitoring and early warning method based on multi-source monitoring data analysis, comprising the following steps:

[0006] Acquiring multi-source monitoring data for river basins, including real-time position data from towed airships, channel image data collected by drones, and environmental parameter data transmitted by satellite remote sensing modules;

[0007] Performing data preprocessing on the multi-source monitoring data to eliminate noise interference in the real-time position data, and performing time synchronization alignment on the waterway image data and environmental parameter data to generate a standardized monitoring data set;

[0008] Based on the standardized monitoring data set, the water flow dynamic characteristics, meteorological anomaly characteristics and waterway obstacle distribution characteristics of the river basin are extracted to generate a multi-dimensional disaster correlation feature set;

[0009] Inputting the multi-dimensional disaster correlation feature set into a preset disaster warning model for dynamic analysis, generating a disaster warning signal, and determining the disaster type and impact range;

[0010] According to the disaster warning signal, the autonomous detachment mechanism of the towed airship and the quick start instruction of the drone are triggered, and the warning information is broadcast to the ships in the waterway through the drone, and the disaster type and impact range are simultaneously transmitted to the command center.

[0011] On the other hand, an embodiment of the present invention provides a computer system including a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and the processor implements the steps in the above method when executing the program.

[0012] The river basin disaster monitoring and early warning method based on multi-source monitoring data analysis provided by the present invention dynamically and collaboratively collects the real-time position data of towed airships, the channel image data of drones, and the environmental parameter data of satellite remote sensing, performs spatiotemporal alignment and standardized fusion of multi-source heterogeneous data, constructs a multi-dimensional disaster correlation feature set including water flow dynamic characteristics, meteorological anomaly characteristics, and obstacle distribution characteristics, and realizes accurate identification of disaster types and impact ranges based on the disaster coupling analysis model, triggering a differentiated collaborative response mechanism for airships and drones. This method enables disaster monitoring to break through the perception limitations of traditional data sources. Through the spatial benchmark unification of airships and drones and cross-modal correlation analysis of multi-dimensional features, it can effectively capture the implicit triggering patterns of complex disasters. Combined with the dynamic matching of airship disengagement mode and drone path planning strategy based on disaster type, multi-dimensional data features are converted into executable hardware action instructions, achieving full-link connection from data perception to emergency response without human intervention, significantly improving the timeliness of disaster warnings and the spatial adaptation accuracy of disposal measures. At the same time, through the spatial overlap verification mechanism of satellite remote sensing data and multi-dimensional features, false alarms and missed alarms are avoided, ensuring the credibility of warning signals and the rationality of command decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 A schematic diagram of the implementation flow of a river basin disaster monitoring and early warning method based on multi-source monitoring data analysis provided by an embodiment of the present invention.

[0014] Figure 2 A schematic diagram of a hardware entity of a computer system provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0015] Embodiments of the present invention provide a river basin disaster monitoring and early warning method based on multi-source monitoring data analysis. The method can be executed by a processor of a computer system. The computer system can refer to a device with data processing capabilities, such as a server, laptop, tablet, or desktop computer, installed in the river basin disaster monitoring and early warning backend.

[0016] Figure 1 A schematic diagram of the implementation flow of a river basin disaster monitoring and warning method based on multi-source monitoring data analysis provided by an embodiment of the present invention is shown as follows: Figure 1 As shown, the method includes:

[0017] Step S100: Acquire multi-source monitoring data of the river basin, where the multi-source monitoring data includes real-time position data of the towed airship, channel image data collected by the drone, and environmental parameter data transmitted by the satellite remote sensing module.

[0018] In this step, the towed airship's real-time position data is the airship's real-time geographic coordinate location within the river channel. It reflects the dynamic changes in the airship's specific position in space. This data can be used to analyze the airship's trajectory and determine whether it has deviated from the preset channel. This data can be obtained using a high-precision Global Positioning System (GPS) device installed on the airship. The GPS device receives satellite signals in real time and calculates the airship's current latitude and longitude coordinates. For example, on the Yangtze River channel, the GPS device on each towed airship records the airship's position every second. The channel image data collected by the drone is image data captured by the drone using its high-definition camera during flight. These images can intuitively display the actual channel conditions, including channel boundaries and obstacle distribution. The drone can capture images according to a preset flight path and shooting parameters. For example, in a Yangtze River channel monitoring scenario, the drone can capture a channel image every 50 meters, with a resolution set to 4K.

[0019] The environmental parameter data transmitted by the satellite remote sensing module is river basin-related environmental information acquired through satellite remote sensing technology. This includes meteorological parameters such as temperature, humidity, wind speed, and air pressure, as well as hydrological parameters such as water level and current velocity. These parameters reflect the environmental status and changing trends of the river basin and are crucial for disaster monitoring and early warning. The satellite remote sensing module regularly monitors the river basin and transmits the collected environmental parameter data to a ground receiving station. For example, a meteorological satellite monitors the environmental parameters of the Yangtze River waterway every 30 minutes and transmits the data back in real time.

[0020] Step S200: pre-process the multi-source monitoring data to eliminate noise interference in the real-time position data, and perform time synchronization alignment on the channel image data and the environmental parameter data to generate a standardized monitoring data set.

[0021] Data preprocessing involves preliminary processing of raw multi-source monitoring data to improve its quality and usability. In this step, real-time location data must be subjected to noise removal. During actual monitoring, due to various factors (such as signal interference and equipment errors), real-time location data may contain noise, which can affect the accuracy of subsequent data analysis and disaster monitoring. This noise can be eliminated using filtering algorithms, such as the Kalman filter. This algorithm processes a series of noisy measurement data to estimate the system's true state, effectively eliminating noise from the real-time location data. Furthermore, time synchronization is required between the waterway image data and the environmental parameter data. Because different data sources may have different acquisition times, time synchronization is necessary to ensure that these data accurately reflect the river basin's conditions at the same moment. This can be achieved by adding a precise timestamp to each data record and then matching and aligning the data from different data sources based on the timestamp. Finally, after data preprocessing and time synchronization, a standardized monitoring dataset is generated. This dataset has a unified data format and time tags, facilitating subsequent data analysis and processing.

[0022] As an implementation manner, step S200 may specifically include the following steps S210 to S240:

[0023] Step S210: Dynamic drift compensation is performed on the real-time position data of the towed airship. Based on the inertial navigation data of the airship and the preset course coordinate reference point, the airship position offset vector is calculated, and the corrected position data is generated by reverse displacement superposition.

[0024] Dynamic drift compensation is designed to correct for potential drift errors in the towed airship's real-time position data. During flight, the airship's actual position may deviate from its theoretical position due to external factors such as wind and currents. This deviation is known as drift error. Inertial navigation data is acquired by the airship's inertial measurement unit (IMU). It includes acceleration recorded by a triaxial accelerometer and the rate of change of heading angle recorded by a gyroscope. This data provides information about the airship's motion and attitude.

[0025] The preset channel coordinate reference points are fixed reference points on a pre-set channel, such as the longitude and latitude coordinates of fixed lighthouses on either side of the channel and their corresponding visual feature points. By combining the airship's inertial navigation data with the preset channel coordinate reference points, the airship's position offset vector can be calculated. The specific calculation process is as follows: First, the original airship trajectory data is generated based on the inertial navigation data. That is, the theoretical position of the airship at different times is calculated using kinematic principles, combined with the initial coordinates of the airship's anchor point. Then, the airship's onboard camera captures the channel boundary image in real time and matches it with the preset visual feature points to obtain the visual positioning deviation data of the airship's current position relative to the reference point. Finally, the original airship trajectory data and the visual positioning deviation data are temporally and spatially aligned. An instantaneous motion model of the airship is constructed based on the airship's acceleration and heading angle change rate. The accumulated displacement error caused by wind and water flow within a continuous time window is calculated to obtain the airship's position offset vector.

[0026] Corrected position data is generated through reverse displacement superposition, which involves superimposing the opposite of the calculated airship position offset vector onto the original airship trajectory data, thereby compensating and correcting the original trajectory data. For example, if the calculated airship position offset vector is (x offset, y offset), the offset vector is subtracted from each coordinate point in the original trajectory data to obtain the corrected position data.

[0027] As an implementation manner, step S210 may specifically include the following steps S211 to S215:

[0028] Step S211: acquiring the inertial navigation data of the towed airship, including the airship motion acceleration recorded by the three-axis accelerometer and the heading angle change rate recorded by the gyroscope, and generating original airship trajectory data in combination with the initial coordinates of the airship anchor point.

[0029] The triaxial accelerometer installed on the towed airship records the airship's acceleration in the x, y, and z directions in real time. The gyroscope installed on the airship records the airship's heading rate of change, or the rate of change of the airship's heading. The initial coordinates of the airship's anchor point are the latitude and longitude coordinates of the airship's initial position at the start of the monitoring mission. By combining the acceleration data recorded by the triaxial accelerometer, the heading rate of change data recorded by the gyroscope, and the initial coordinates of the airship's anchor point, kinematic principles can be used to generate raw airship trajectory data. Specifically, the acceleration data is integrated to obtain velocity data, which is then integrated to obtain displacement data. Combining the initial coordinates with the heading rate of change data, the theoretical position of the airship at each moment can be calculated, thereby generating raw airship trajectory data. For example, if the initial coordinates of the airship anchor point are (x0, y0), at time t, the displacement calculated based on the acceleration data and the heading angle change rate data is (Δx, Δy), then the theoretical position coordinates of the airship at time t are (x0+Δx, y0+Δy).

[0030] Step S212: Extract a preset channel coordinate reference point sequence, which includes the latitude and longitude coordinates of the fixed lighthouses on both sides of the channel and the corresponding visual feature points. The channel boundary image is captured in real time by the camera carried by the airship and the visual feature points are matched to generate the visual positioning deviation data of the current position of the airship relative to the reference point.

[0031] The preset channel coordinate reference point sequence is a series of pre-determined reference points on the channel. These reference points include the latitude and longitude coordinates of fixed lighthouses on both sides of the channel and their corresponding visual feature points. Visual feature points are representative features of the lighthouse in the image, such as the top and edges of the lighthouse. The airship's onboard camera captures channel boundary images in real time. The captured images are analyzed using image processing algorithms to identify visual feature points in the images and match them with the preset visual feature points. This matching process can utilize feature matching algorithms such as the Scale-Invariant Feature Transform (SIFT) algorithm or the Speeded Up Robust Features (SURF) algorithm. These algorithms can accurately identify and match feature points in images under different scales and rotation conditions.

[0032] By matching visual feature points, the visual positioning deviation data of the airship's current position relative to the reference point can be calculated. The specific calculation method is to use geometric relationships based on the position of the matched visual feature point in the image and the latitude and longitude coordinates of the corresponding reference point to calculate the position deviation of the airship relative to the reference point. For example, if there is a pixel offset between the matched visual feature point in the image and the preset visual feature point, the pixel offset can be converted into an actual geographic distance offset through a pre-calibrated conversion relationship between image pixels and actual geographic distances, thereby obtaining the visual positioning deviation data of the airship's current position relative to the reference point.

[0033] Step S213: The original airship trajectory data and the visual positioning deviation data are aligned in time and space, and an airship instantaneous motion model is constructed based on the airship motion acceleration and heading angle change rate. The cumulative displacement error of the airship caused by wind and water flow in the continuous time window is calculated to generate the airship position offset vector.

[0034] Spatiotemporal alignment involves matching and aligning the original airship trajectory data and visual positioning deviation data in time and space, ensuring they accurately reflect the airship's position at the same moment. This is achieved by correlating the original airship trajectory data and visual positioning deviation data at the same moment based on the timestamps of the data records.

[0035] An instantaneous motion model of the airship is constructed based on its acceleration and heading rate. This model describes the airship's motion and forces at a given moment. This step considers the effects of wind and current on the airship. By analyzing the acceleration and heading rate, and combining the principles of fluid mechanics and aerodynamics, we construct an instantaneous motion model that reflects the airship's motion under the influence of wind and current.

[0036] Using the constructed instantaneous motion model of the airship, the cumulative displacement error caused by wind and water currents within a continuous time window is calculated. The calculation process involves dividing the continuous time window into multiple small time intervals. Within each time interval, the displacement error caused by wind and water currents is calculated based on the airship's acceleration and heading angle rate of change, combined with the instantaneous motion model. These displacement errors are then accumulated to obtain the cumulative displacement error within the continuous time window. Ultimately, the accumulated displacement error is used as the airship position offset vector. For example, if the continuous time window is divided into n time intervals, and the displacement error calculated within the i-th time interval is (Δxi, Δyi), then the cumulative displacement error (i.e., the airship position offset vector) is (ΣΔxi, ΣΔyi), where i ranges from 1 to n.

[0037] Step S214: reversely compensate the original airship trajectory data according to the airship position offset vector, use the visual positioning deviation data as a correction constraint, perform track fitting optimization on the compensated trajectory, eliminate the drift component of the inertial navigation data, and generate corrected position data.

[0038] Inverse compensation involves superimposing the opposite vector of the airship's position offset onto the original airship trajectory data to correct for position offsets caused by factors such as wind and current. For example, if the airship's position offset vector is (x offset, y offset), then this offset vector is subtracted from each coordinate point in the original trajectory data to obtain the preliminarily compensated trajectory data.

[0039] The purpose of using visual positioning deviation data as a correction constraint is to further improve the accuracy of the correction. Visual positioning deviation data reflects the actual position deviation of the airship relative to the preset course coordinate reference point. During the track fitting optimization process, by introducing visual positioning deviation data, the compensated trajectory can be constrained and adjusted to better match the actual situation.

[0040] Track fitting optimization can utilize curve fitting algorithms, such as the least-squares curve fitting algorithm. This algorithm seeks a curve that minimizes the sum of squared errors between the curve and the compensated trajectory data, thereby obtaining the optimal fitting curve. During the fitting process, the constraints of the visual positioning deviation data are considered, and the fitting parameters are adjusted to eliminate the drift component of the inertial navigation data. Ultimately, track fitting optimization yields corrected position data that more accurately reflects the actual position of the airship.

[0041] Step S215: Perform residual analysis on the corrected position data and the channel coordinate reference point. If the residual value exceeds the preset tolerance range, re-match the visual feature points and iteratively perform dynamic drift compensation processing until the corrected position data that meets the accuracy requirements is generated.

[0042] Residual analysis calculates the difference between the corrected position data and the channel coordinate reference point, i.e., the residual value. This residual value reflects the degree of deviation between the corrected position data and the preset reference point. By comparing the residual value with the preset tolerance range, it can be determined whether the corrected position data meets the accuracy requirements. The preset tolerance range is a pre-set allowable error range, determined based on actual monitoring needs and accuracy requirements.

[0043] If the residual value exceeds the preset tolerance range, it means that the corrected position data still contains large errors, and it is necessary to re-match the visual feature points and iteratively perform dynamic drift compensation. Re-matching the visual feature points can be done by processing and analyzing the channel boundary image captured by the airship's camera again, and using the feature matching algorithm to find a more accurate visual feature point matching result. Then, based on the new visual positioning deviation data, the airship position offset vector is recalculated, the original airship trajectory data is reversely compensated and track fitting is optimized, and the residual analysis is performed again. The above process is repeated until the residual value is within the preset tolerance range, and the corrected position data that meets the accuracy requirements is generated.

[0044] Step S220: Adaptively defogging the channel image data collected by the drone, adjusting the intensity parameters of the defogging algorithm according to the real-time visibility value in the environmental parameter data, extracting the contour information of the channel boundary and obstacles in the image, and generating enhanced channel image data.

[0045] Adaptive defogging automatically adjusts the parameters of the defogging algorithm according to the actual conditions of the image and the environmental conditions to achieve the best defogging effect. In this step, the real-time visibility value in the environmental parameter data is used to adjust the intensity parameter of the defogging algorithm. The real-time visibility value reflects the clarity of the current channel environment. The lower the visibility, the more serious the impact of the fog, and a stronger defogging algorithm is required. The defogging algorithm can adopt a defogging algorithm based on a physical model, such as a dark channel prior defogging algorithm. This algorithm is based on the atmospheric scattering model and estimates the atmospheric light and transmittance by analyzing the dark channel information of the image, thereby achieving image defogging. In practical applications, the transmittance estimation parameters of the dark channel prior defogging algorithm are dynamically adjusted according to the real-time visibility value to adapt to different fog concentrations.

[0046] Extracting the outlines of channel boundaries and obstacles in images is crucial for better identifying channel conditions and obstacle distribution. Edge detection algorithms, such as the Canny edge detection algorithm, can be used to process dehazed images to detect the edges of these boundaries and obstacles. Morphological processing algorithms, such as dilation and erosion, can then be used to optimize and refine these edges, ultimately yielding the outlines of these boundaries and obstacles.

[0047] Finally, the extracted contour information is fused with the dehazed image to generate enhanced waterway image data. This image data can more clearly show the location and shape of waterway boundaries and obstacles, facilitating subsequent disaster monitoring and analysis.

[0048] As an implementation manner, step S220 may specifically include the following steps S221 to S226:

[0049] Step S221: Obtain the real-time visibility value from the original channel image data and environmental parameter data collected by the UAV, divide the image fog concentration level based on the real-time visibility value, and determine the defogging intensity coefficient corresponding to different levels.

[0050] The raw channel image data collected by drones is directly captured during flight. These images may be affected by fog and have low clarity. The real-time visibility value in the environmental parameter data is visibility information of the current channel environment obtained through satellite remote sensing modules or other meteorological monitoring equipment.

[0051] Image fog concentration levels are divided based on real-time visibility values. Fog concentration can be divided into different levels, such as light fog, moderate fog, and heavy fog, according to pre-set visibility thresholds. Each level corresponds to a defogging intensity coefficient, which is used to adjust the intensity of the defogging algorithm. For example, when the real-time visibility value is between 500 meters and 1000 meters, it is classified as light fog, and the corresponding defogging intensity coefficient is 0.3; when the real-time visibility value is between 200 meters and 500 meters, it is classified as moderate fog, and the corresponding defogging intensity coefficient is 0.6; when the real-time visibility value is less than 200 meters, it is classified as heavy fog, and the corresponding defogging intensity coefficient is 0.9.

[0052] Step S222: Dynamically adjust the transmittance estimation parameters of the adaptive defogging algorithm according to the defogging intensity coefficient, perform multi-scale filtering on the original channel image data, separate the high-frequency details and the low-frequency fog background in the image, and generate a preliminary defogging image.

[0053] The transmittance estimation parameter of the adaptive defogging algorithm is a key parameter that affects the defogging effect. This parameter is dynamically adjusted based on the defogging intensity coefficient determined in step S221. For example, in the dark channel prior defogging algorithm, the transmittance estimation parameter is proportional to the defogging intensity coefficient. The larger the defogging intensity coefficient, the smaller the transmittance estimation parameter, and the stronger the defogging effect.

[0054] Multiscale filtering involves filtering the raw channel image data at different scales to separate high-frequency details from the low-frequency haze background. Multiscale decomposition can be performed using either the Gaussian pyramid or Laplacian pyramid algorithms. The Gaussian pyramid algorithm continuously smoothes and downsamples the image to produce images of varying scales; the Laplacian pyramid algorithm subtracts two adjacent layers of the Gaussian pyramid to extract high-frequency detail information. Multiscale filtering can separate the haze background from the high-frequency detail information in the image. The haze background information is then processed to remove the effects of fog, resulting in a preliminary dehazed image.

[0055] Step S223: Perform local contrast enhancement processing on the preliminary dehazed image, adaptively adjust the brightness compensation weight based on the geometric feature distribution of the channel boundary, suppress overexposed areas and enhance dark textures, and generate a balanced intermediate image.

[0056] Local contrast enhancement is used to improve contrast in local areas of an image, making it clearer. The CLAHE algorithm can be used to process the initial dehazed image. This algorithm divides the image into multiple small regions and performs histogram equalization on each region to enhance local contrast.

[0057] Adaptively adjusting the brightness compensation weight based on the geometric distribution of the channel boundary is designed to adjust the image brightness according to the actual channel boundary conditions. The geometric characteristics of the channel boundary include information such as its length and curvature. By analyzing the distribution of the geometric features of the channel boundary, the brightness compensation weights for different areas can be determined. For example, near the channel boundary, due to its high importance, the brightness compensation weight can be appropriately increased to highlight the channel boundary; whereas, in some overexposed areas, the brightness compensation weight can be reduced to suppress overexposure. Through local contrast enhancement and brightness compensation adjustment, overexposed areas are suppressed and dark textures are enhanced, generating a balanced intermediate image. This image has a more balanced contrast and brightness, which facilitates subsequent contour information extraction.

[0058] Step S224: Extract edge gradient information of the channel boundary and obstacles from the equalized intermediate image, fuse the gradient amplitudes of consecutive pixels through a multi-directional edge detection algorithm, and generate a contour mask containing the channel boundary outline and obstacle shape.

[0059] Edge gradient information is the rate of change of grayscale at an image pixel in different directions. It reflects the location and direction of the edge of an object in the image. Edge detection operators such as the Sobel operator or the Prewitt operator can be used to process the equalized intermediate image to calculate the edge gradient information for each pixel in the image.

[0060] The multi-directional edge detection algorithm is designed to more comprehensively detect edge information in an image. It performs edge detection in multiple directions and then fuses the gradient magnitudes of consecutive pixels. For example, edge detection operators can be used to detect in multiple directions, such as horizontal, vertical, and diagonal. The gradient magnitudes in different directions are then fused to obtain more accurate edge information.

[0061] Based on the extracted edge gradient information and multi-directional edge detection results, a contour mask is generated that contains the outline of the channel boundary and obstacle shapes. The contour mask is a binary image in which white areas represent the outlines of the channel boundary and obstacles, and black areas represent other areas. The contour mask is generated by thresholding the edge gradient magnitude, marking pixels with gradient magnitudes greater than the set threshold as white and pixels with gradient magnitudes less than the threshold as black.

[0062] Step S225: The contour mask and the equalized intermediate image are superimposed and fused, and the image sharpening intensity is optimized according to the continuity characteristics of the channel boundary in the contour mask, the residual artifacts of defogging are eliminated, and the contrast of the obstacle edge is enhanced to generate enhanced channel image data.

[0063] The purpose of overlaying and fusing the contour mask with the equalized intermediate image is to combine the extracted channel boundary and obstacle outline information with the original image, making the image more clearly display the channel conditions. A simple image overlay method can be used to add the grayscale values ​​of the corresponding pixels in the contour mask and the equalized intermediate image to obtain the fused image.

[0064] The image sharpening intensity is optimized based on the continuity characteristics of the channel boundary in the contour mask. This continuity characteristic includes information such as the boundary length and smoothness. By analyzing these characteristics, the image sharpening intensity can be determined. For example, when the channel boundary is long and smooth, the image sharpening intensity can be increased to enhance the contrast of the obstacle edge. When the channel boundary is short and discontinuous, the image sharpening intensity can be reduced to avoid excessive noise.

[0065] By optimizing the image sharpening intensity, removing residual artifacts from defogging and enhancing the contrast of obstacle edges, enhanced channel image data is generated. This image data removes the effects of fog while more clearly showing the location and shape of channel boundaries and obstacles.

[0066] Step S226: Verify the contour integrity of the enhanced channel image data. If a channel boundary break or a blurred obstacle contour is detected, readjust the defogging intensity coefficient and iteratively perform adaptive defogging processing until enhanced channel image data that meets the recognition requirements is generated.

[0067] Contour integrity verification checks the integrity of channel boundaries and obstacle outlines in enhanced channel image data. Morphological analysis methods, such as connected component analysis, can be used to analyze contours in the image. By calculating contour parameters such as connectivity and area, it can be determined whether channel boundaries are broken or obstacle outlines are blurred.

[0068] If the channel boundary is broken or the obstacle outline is blurred, the dehazing process is ineffective and the dehazing intensity coefficient needs to be readjusted. Depending on the actual situation, the dehazing intensity coefficient should be appropriately increased or decreased, and then adaptive dehazing processing should be performed again, including multi-scale filtering, local contrast enhancement, edge detection, and outline fusion. This process is repeated until the generated enhanced channel image data meets the recognition requirements, that is, the channel boundary is intact and the obstacle outline is clear.

[0069] Step S230: Perform outlier repair processing on the environmental parameter data transmitted by the satellite remote sensing module. Based on the distribution pattern of environmental parameters of adjacent airship monitoring nodes in the same time period, replace the data points that exceed the preset fluctuation range to generate a repaired environmental parameter sequence.

[0070] Outlier repair is used to remove possible outliers in environmental parameter data, improving data accuracy and reliability. The distribution pattern of environmental parameters between adjacent airship monitoring nodes during the same time period reflects the normal range of environmental parameter variation in that area. The preset fluctuation range is an allowable range of parameter fluctuations determined based on historical data and actual monitoring experience.

[0071] When performing outlier repair processing, the environmental parameter data of adjacent airship monitoring nodes within the same time period are first analyzed, and statistical quantities such as their mean and standard deviation are calculated to determine the normal distribution range of the environmental parameters. Then, each data point in the environmental parameter data transmitted by the satellite remote sensing module is checked to determine whether it exceeds the preset fluctuation range. If a data point exceeds the preset fluctuation range, it is considered an outlier. For outliers, the environmental parameter data of adjacent airship monitoring nodes are used to replace them. A specific replacement method can be to take the average value or median of the environmental parameters of adjacent nodes as the replacement value. For example, if the preset fluctuation range of a certain environmental parameter is the mean ± 2 times the standard deviation, when a data point is detected to exceed this range, the average value of the environmental parameter of the adjacent airship monitoring nodes is calculated, and the outlier data point is replaced with this average value. Finally, through the outlier repair processing, a repaired environmental parameter sequence is generated.

[0072] Step S240: Perform spatiotemporal alignment processing on the corrected position data, enhanced channel image data, and restored environmental parameter sequence. According to the mapping relationship between the airship position and the drone aerial photography timestamp, interpolate and complete the data of the missing time window to generate a standardized monitoring data set with complete spatiotemporal labels.

[0073] Spatiotemporal alignment involves matching and aligning the corrected position data, enhanced channel image data, and restored environmental parameter sequences in time and space, ensuring they accurately reflect the river basin conditions at the same time and location. This is achieved by associating data at the same time and location based on the timestamp and location information of each data record.

[0074] The mapping relationship between airship positions and drone aerial photography timestamps represents the correspondence between the airship's position at different times and the timestamps of the drone's aerial images captured at the corresponding times. This mapping relationship allows for the temporal and spatial correlation between the corrected position data and the enhanced aerial image data. Interpolation is performed to complete missing time windows to ensure the integrity of the standardized monitoring dataset. In actual monitoring, data may be missing for certain time windows due to various reasons (such as equipment failure or signal interruption). Interpolation algorithms, such as linear interpolation or spline interpolation, can be used to estimate and complete the data for the missing time windows based on the values ​​of known data points. For example, if environmental parameter data is known at times t1 and t2, but data for time t between t1 and t2 is missing, a linear interpolation algorithm can be used to calculate an estimated value at time t based on the data values ​​at t1 and t2.

[0075] Finally, through spatiotemporal alignment and data interpolation, a standardized monitoring dataset with complete spatiotemporal labels is generated. This dataset has a unified data format and time tags, facilitating subsequent data analysis and processing.

[0076] Step S300: Based on the standardized monitoring data set, the water flow dynamic characteristics, meteorological anomaly characteristics and channel obstacle distribution characteristics of the river basin are extracted to generate a multi-dimensional disaster correlation feature set.

[0077] The standardized monitoring dataset contains information such as corrected position data, enhanced waterway imagery, and restored environmental parameter sequences. This information comprehensively reflects the actual conditions of the river basin. In this step, we extract flow dynamics, meteorological anomalies, and the distribution of waterway obstacles from the standardized monitoring dataset.

[0078] Water flow dynamics reflect the state and changing patterns of water flow in a river basin, such as the gradient of water velocity. Meteorological anomaly characteristics are unusual changes in meteorological parameters (such as temperature and humidity), such as periods of simultaneous decreases in temperature gradients and sudden increases in humidity. Waterway obstacle distribution characteristics describe the distribution of obstacles within a waterway, such as their projected area and spatial density.

[0079] By analyzing and processing standardized monitoring datasets, these features are extracted and fused in space and time to generate a multidimensional disaster-related feature set. This set encompasses multiple aspects of disaster-related features and can more comprehensively reflect the potential disaster risks in river basins.

[0080] As an implementation manner, step S300 may specifically include the following steps S310 to S340:

[0081] Step S310: Based on the corrected position data, the displacement acceleration of the towed airship in continuous time intervals is calculated, and the water velocity gradient variation characteristics are deduced by combining the channel water depth data and the airship draft.

[0082] The corrected position data accurately reflects the actual position of the towed airship within the river basin. By calculating the displacement acceleration of the towed airship within consecutive time intervals, changes in the airship's motion state can be understood. The specific calculation method is to calculate the displacement difference between two adjacent time points based on the corrected position data, then divide it by the time interval to obtain the velocity value; then calculate the difference between two adjacent velocity values ​​and divide it by the time interval to obtain the displacement acceleration. Channel depth data, which provides depth information at different locations within the river channel, can be obtained using water depth measurement equipment such as sonar. The airship draft is the depth to which the airship is submerged in the water. Combining channel depth data with the airship draft allows analysis of the impact of water currents on the airship.

[0083] By analyzing the displacement acceleration, channel depth, and draft of the towed airship, the characteristics of the water velocity gradient can be derived. This derivation can be based on the principles of fluid mechanics and kinematics, establishing a mathematical model between the water flow and the airship's motion. For example, if the force exerted by the water on the airship is proportional to the water velocity, the longitudinal gradient of the water velocity can be calculated by analyzing the force and motion of the airship, combined with factors such as the channel depth and the airship's draft.

[0084] Step S320: extract the coordinated change trend of temperature and humidity from the repaired environmental parameter sequence, identify the concurrent intervals of temperature gradient decrease and humidity surge, and generate meteorological anomaly correlation features.

[0085] The repaired environmental parameter sequence contains meteorological parameter data such as temperature and humidity, after outlier correction. To extract the coordinated trend of temperature and humidity from this sequence, time series analysis of the temperature and humidity data is required. A sliding window approach can be used to calculate the rate of change of temperature and humidity within a set time window and analyze their correlation.

[0086] Identify periods of concurrent temperature gradient decline and humidity surges. Specifically, identify periods within the same timeframe where the temperature drops rapidly and the humidity increases sharply. You can set temperature drop thresholds and humidity increase thresholds. When the temperature drop exceeds the temperature drop threshold and the humidity increase exceeds the humidity increase threshold, the period is considered to be a period of concurrent temperature drop and humidity surges.

[0087] Based on the identified concurrent intervals, meteorological anomaly correlation features are generated. These features can be described using parameters such as the time range of the concurrent interval, the magnitude of the temperature drop, and the magnitude of the humidity increase. They reflect the abnormal changes in meteorological parameters and may be related to the occurrence of meteorological disasters.

[0088] Step S330: Perform multi-scale obstacle segmentation processing on the enhanced channel image data, distinguish floating objects, fixed obstacles and dynamic targets of ships based on contour information, calculate the projected area and spatial density of obstacles in combination with the drone altitude data, and generate the channel obstacle distribution characteristics.

[0089] Multi-scale obstacle segmentation involves segmenting enhanced airway image data at different scales to more accurately identify and distinguish different types of obstacles. Deep learning-based semantic segmentation algorithms, such as the U-Net network, can be used for image segmentation. The U-Net network is a convolutional neural network with an encoder-decoder structure that automatically learns the characteristics of different objects in an image and segments the image into different categories.

[0090] Based on the segmented contour information, dynamic targets such as floating objects, fixed obstacles, and ships can be distinguished. This distinction can be made by analyzing contour features such as shape, size, and motion. For example, the contours of floating objects are typically small and irregular, potentially moving with the current; the contours of fixed obstacles are relatively stable and their position remains constant; dynamic targets such as ships exhibit distinct motion characteristics, and their contours may resemble the outer shape of the ship.

[0091] The projected area and spatial density of obstacles are calculated by combining drone altitude data. Drone altitude data can be obtained using an altitude sensor mounted on the drone. Based on the drone altitude and the pixel size of obstacles in the image, geometric relationships can be used to calculate the projected area of ​​obstacles on the ground. Spatial density, the number of obstacles per unit volume, can be calculated by counting the number of obstacles in the image and their corresponding spatial extent.

[0092] Finally, the calculated obstacle projected area and spatial density are used to generate a waterway obstacle distribution feature. This feature can describe the distribution of obstacles in the waterway and is of great significance for waterway safety and disaster monitoring.

[0093] Step S340: The water velocity gradient change characteristics, meteorological anomaly correlation characteristics and obstacle spatial density characteristics are temporally and spatially superimposed and fused, and a multidimensional disaster correlation feature set containing temporal and spatial correlation is generated based on a preset disaster feature weight table.

[0094] Spatiotemporal fusion combines the characteristics of water velocity gradients, meteorological anomaly correlations, and obstacle density across time and space, integrating them into a comprehensive feature set. This is achieved by correlating and combining features at the same time and location based on each feature's timestamp and location information.

[0095] The preset disaster feature weight table is a pre-set table that records the importance of different disaster features in disaster assessment, i.e., their weights. By combining the fused features with the preset disaster feature weight table and weighting each feature, a multidimensional disaster correlation feature set with spatiotemporal correlations is obtained.

[0096] Step S400: Input the multi-dimensional disaster correlation feature set into a preset disaster warning model for dynamic analysis, generate a disaster warning signal, and determine the disaster type and impact range.

[0097] The pre-set disaster warning model is a trained and optimized model used to analyze and process a multi-dimensional set of disaster-related features to predict potential disasters. This model can use machine learning or deep learning algorithms, such as neural networks and decision trees.

[0098] A multidimensional set of disaster-related features is input into the disaster warning model, which then dynamically analyzes the input features. The specific analysis process includes steps such as feature extraction, feature selection, and model inference. For example, in a neural network model, the input multidimensional set of disaster-related features is passed from the input layer to the hidden layer. Neurons in the hidden layer perform nonlinear transformations and processing on the features, and finally output the prediction results through the output layer. Based on the model's analysis results, a disaster warning signal is generated. Disaster warning signals can be represented by different levels and identifiers, such as level one, level two, and level three. The disaster type and impact range are also determined. Disaster types may include floods, meteorological disasters, and obstacle accumulation disasters. The impact range can be described using geographic coordinates and regional boundaries.

[0099] As an implementation method, the iterative optimization process of the disaster warning model includes the following steps S401 to S406:

[0100] Step S401: Collect complete response chain data of historical disaster events, including warning trigger delay, airship deviation from trajectory, number of drone path replanning times, and final disaster loss assessment indicators.

[0101] Complete response chain data for historical disaster events records relevant information from the entire monitoring, early warning, and response process during past disasters. Warning trigger latency is the time interval from the occurrence of a disaster to the issuance of the warning signal, reflecting the timeliness of the warning system. Airship trajectory deviation is the deviation of a towed airship from its preset trajectory during a disaster, reflecting the stability of the airship in disaster response. The number of drone re-route planning is the number of times a drone re-planned its flight path due to changes in the disaster situation during its early warning mission, reflecting the drone's flexibility in responding to disasters. The final disaster loss assessment indicator is a quantitative measure of the losses caused by a disaster, such as economic losses and casualties.

[0102] By collecting the complete response chain data of these historical disaster events, we can provide rich samples and data support for the iterative optimization of disaster warning models.

[0103] Step S402: Construct a multi-dimensional model training feature set, including the coupling degree of water flow characteristics and meteorological characteristics, the spatial correlation between obstacle distribution and ship avoidance behavior, and the correspondence between warning signal level and rescue resource consumption.

[0104] The multidimensional model training feature set is designed to more comprehensively describe the process of disaster occurrence and development, as well as the effectiveness of early warning and rescue responses. The coupling degree between water flow and meteorological characteristics reflects the impact of the interaction between water flow and meteorological factors on disasters. For example, in flood disasters, there may be a correlation between water flow velocity and rainfall. By analyzing their coupling degree, we can better predict the occurrence and development of floods. The spatial correlation between obstacle distribution and ship avoidance behavior describes the impact of the distribution of obstacles in the waterway on ship avoidance behavior. For example, when obstacles are densely distributed, ships may need to make more avoidance maneuvers. By analyzing their spatial correlation, we can optimize ship avoidance strategies. The correspondence between warning signal level and rescue resource consumption reflects the amount and type of rescue resources required for different warning signal levels. For example, a level 1 warning requires more rescue personnel and equipment. By analyzing this correspondence, we can rationally allocate rescue resources. By constructing a multidimensional model training feature set, we can improve the accuracy and reliability of disaster warning models.

[0105] Step S403: A phased reinforcement learning strategy is used to train the model: the first phase learns the feature recognition pattern of a single disaster type, the second phase integrates the multi-disaster coupling influencing factors, and the third phase optimizes the matching relationship between the warning level and the allocation of rescue resources.

[0106] The phased reinforcement learning strategy divides the model training process into multiple stages, gradually improving the model's performance and capabilities. In the first stage, the model learns the characteristic recognition patterns of a single disaster type. By analyzing historical disaster event data, the model learns the characteristics and patterns of different disaster types (such as floods, meteorological disasters, and obstruction accumulation disasters) to accurately identify a single disaster type. In the second stage, the model integrates the influencing factors of multiple disasters. Real-world disasters are often the result of multiple disasters coupled together. For example, floods may be accompanied by meteorological disasters and obstruction accumulation disasters. In this stage, the model learns how to comprehensively consider the impact of multiple disasters, improving its predictive capabilities for coupled multi-hazard situations. In the third stage, the model optimizes the matching relationship between warning levels and rescue resource allocation. Based on different warning signal levels, rescue resources are rationally allocated to achieve optimal rescue results. By learning the corresponding relationship between warning signal levels and rescue resource consumption in historical data, the model continuously adjusts and optimizes the matching strategy between warning levels and rescue resource allocation, enabling more scientific and efficient deployment of rescue forces when disasters occur.

[0107] During training, reinforcement learning algorithms such as the Deep Q-Network (DQN) or the Policy Gradient Algorithm are employed. For example, DQN maintains an action-value function Q(s, a) to estimate the expected cumulative reward for taking action a in state s. At each training step, the model selects an action (such as issuing a warning signal of varying levels or allocating different amounts and types of rescue resources) based on the current state (i.e., the state represented by the multidimensional model training feature set). A reward is then assigned based on environmental feedback (i.e., the actual disaster development and rescue effectiveness). Through continuous iterative training, the model adjusts its action-value function to maximize the long-term cumulative reward.

[0108] In the first stage, the model primarily learns the characteristic recognition patterns of a single disaster type. For each disaster type, a preset reward function is defined. For example, for flood disasters, if the model accurately identifies the occurrence of a flood disaster and issues a correct warning signal, a positive reward is given; if it misidentifies or misses a warning, a negative reward is given. The model continuously adjusts its parameters through interaction with the environment to improve the accuracy of identifying a single disaster type.

[0109] In the second stage, we consider the case of multiple disasters coupled together. In this case, the reward function needs to comprehensively account for the impact of multiple disasters. For example, if a flood and a meteorological disaster occur simultaneously, if the model can accurately identify the coupling of the two disasters, issue appropriate warning signals, and allocate appropriate rescue resources, a higher positive reward will be given. If it only identifies one disaster or misjudges the coupling of the disasters, a lower positive or even negative reward will be given. In this way, the model gradually learns the role of multi-hazard coupling factors, improving its predictive capabilities for complex disaster situations.

[0110] In the third stage, the focus is on optimizing the relationship between warning levels and rescue resource allocation. The reward function is related to the effective utilization of rescue resources and the reduction of disaster losses. If the model can rationally allocate rescue resources based on the severity and development trend of the disaster, minimizing disaster losses, a high reward will be given. If rescue resources are not allocated rationally, resulting in large disaster losses or waste of resources, a low or negative reward will be given. Through continuous training and optimization, the model will find the optimal matching strategy between warning levels and rescue resource allocation.

[0111] Step S404: After the model is deployed, real-time disaster response data is continuously collected, and the spatial overlap between the actual disaster impact range and the predicted range is used as a reward function to dynamically adjust the feature weight distribution coefficient in the model.

[0112] After a disaster warning model is deployed in real-world applications, it's crucial to continuously collect real-time disaster response data. This data includes the type of disaster that actually occurred, the scope of the disaster, the implementation of warning signals, and the effectiveness of rescue resources. By comparing the actual disaster impact area with the model's predicted impact area, the spatial overlap between them can be calculated.

[0113] Spatial overlap is an important metric for measuring model prediction accuracy. A high degree of spatial overlap between the actual disaster impact area and the predicted area indicates a relatively accurate model prediction. Conversely, a low degree of overlap indicates a significant error. Spatial overlap is used as a reward function, with positive rewards awarded for high overlap and negative rewards for low overlap.

[0114] Based on the feedback from the reward function, the model's feature weighting coefficients are dynamically adjusted. These coefficients determine the importance of each feature in the model's decision-making. For example, if, in a disaster, water flow characteristics are found to have a greater impact than expected, while meteorological characteristics have a relatively smaller impact, the weight of water flow characteristics can be appropriately increased, while the weight of meteorological characteristics can be reduced. By continuously adjusting the feature weighting coefficients, the model can better adapt to actual conditions and improve prediction accuracy.

[0115] Step S405: When a new disaster pattern is detected or the prediction error rate of the original model continues to rise, the incremental learning mechanism is started to retain the existing knowledge while integrating new features to generate an extended disaster classification tree.

[0116] In real-world applications, new disaster patterns may emerge, or the prediction error rate of the original model may continue to increase due to factors such as environmental changes. When these situations are detected, incremental learning mechanisms need to be activated. Incremental learning is a learning method that can continuously absorb new data and new features without destroying the original model knowledge.

[0117] After the incremental learning mechanism is initiated, the existing knowledge and parameters in the model are first retained. Then, newly collected data is analyzed to extract new features. These new features may include those of new disasters or environmental changes. These new features are integrated with existing features to generate an extended disaster classification tree. The extended disaster classification tree expands and updates the existing disaster classification tree. It can identify more types of disaster patterns, including new ones. Decision tree algorithms, such as the C4.5 algorithm or the random forest algorithm, can be used to generate the extended disaster classification tree. These algorithms can construct more accurate and comprehensive disaster classification models based on new and existing features.

[0118] Step S406: Regularly encrypt and transmit the optimized model parameters to all airship nodes, ensure the consistency of the model version of each node through a distributed verification protocol, and perform silent update operations during low-load periods.

[0119] To ensure that the disaster warning model on all airship nodes has the same performance and predictive capabilities, it is necessary to regularly encrypt and transmit the optimized model parameters to all airship nodes. Encrypted transmission can use symmetric encryption algorithms (such as AES) or asymmetric encryption algorithms (such as RSA) to ensure the security of model parameters during transmission.

[0120] After model parameters are transmitted, a distributed verification protocol is used to ensure model version consistency across all nodes. This distributed verification protocol can utilize Byzantine fault-tolerant algorithms (such as PBFT) or blockchain technology. These protocols ensure data consistency and reliability in a distributed environment. For example, in the PBFT algorithm, each node verifies the received model parameters and communicates and negotiates with other nodes. Only when a majority of nodes have verified the parameters will the local model parameters be updated.

[0121] To minimize the impact on normal monitoring tasks, silent updates are performed during low-load periods. These periods typically occur when river monitoring tasks are relatively light, such as late at night or early in the morning. During these periods, the airship node's computing and communication resources are relatively idle, allowing model parameter updates to be performed without significantly disrupting real-time monitoring and early warning operations.

[0122] As an implementation method, the hierarchical processing and dynamic upgrade mechanism of the early warning signal includes the following steps S4021 to S4026:

[0123] Step S4021: Generate three-level warning signs according to the comprehensive disaster level: level one corresponds to the multi-disaster coupling and diffusion state, level two corresponds to the single disaster continuous intensification state, and level three corresponds to the potential risk warning state.

[0124] The comprehensive disaster level is derived from the analysis of disaster warning models, taking into account multiple disaster factors and their coupling relationships. Generating three-level warning signs based on the comprehensive disaster level more clearly conveys the severity and development of the disaster.

[0125] A Level 1 warning corresponds to a multi-hazard coupled diffusion state, meaning multiple disasters are likely to occur simultaneously, interacting and spreading, potentially causing severe damage and impacts to river basins. For example, floods, meteorological disasters, and debris accumulation disasters may occur simultaneously, and the impact of these disasters is expanding.

[0126] The Level 2 warning sign corresponds to a single disaster that is continuously intensifying, meaning the intensity of a particular disaster is increasing, potentially causing significant impacts in a local area. For example, the water level of a flood disaster continues to rise, or meteorological indicators such as wind speed and rainfall continue to increase.

[0127] Level 3 warning corresponds to a potential risk warning state. Although no obvious disaster has occurred yet, there are potential risk factors that may trigger a disaster. For example, there are abnormal changes in meteorological parameters or potential obstacles in the waterway.

[0128] Step S4022: When the first-level warning sign is generated, the full-channel strobe warning mode is triggered, all airships are controlled to switch to red warning lights, and drones use an alternating circling strategy to block the entrance to the core disaster area.

[0129] When a Level 1 warning sign is generated, it indicates that the disaster situation is extremely serious and urgent warning and preventive measures are required. The full-channel strobe warning mode uses warning lights installed on both sides of the river channel or lighting equipment on airships and drones to emit strobe light signals to attract the attention of ships and personnel within the channel.

[0130] All airships were controlled to switch their lights to red. Red lights typically indicate danger and emergency situations, more intuitively conveying the severity of a disaster. Drones employed an alternating circling strategy to block the entrance to the core disaster zone, preventing vessels from entering and causing further damage. Drones could alternately circle around the entrance to the core disaster zone according to a pre-set trajectory, emitting sound and light signals to warn and intercept vessels.

[0131] Step S4023: For the second-level warning sign, the directional warning mode is activated. Based on the real-time position and heading data of the ship, a personalized obstacle avoidance path is generated for each ship and displayed on the channel surface through the laser projection device of the airship.

[0132] When a Level 2 warning is generated, although the severity of the disaster is lower than that of a Level 1 warning, effective warning and guidance of ships within the waterway is still necessary. Targeted warning mode provides warning and guidance based on the specific situation of each ship, rather than a comprehensive warning.

[0133] Based on the vessel's real-time position and heading data, along with channel maps and obstacle distribution information, a personalized obstacle avoidance path is generated for each vessel. This personalized obstacle avoidance path is calculated based on the vessel's current position, heading, and speed, as well as the distribution of obstacles in the channel, resulting in a safe and feasible navigation path. The personalized obstacle avoidance path is displayed on the channel surface via the airship's laser projection system, allowing crew members to visually visualize the path and navigate accordingly, avoiding collisions.

[0134] Step S4024: For the third-level warning sign, a progressive warning strategy is enabled, warning information is scrolled on the advertising screen on the surface of the airship, and the drone is controlled to conduct low-altitude reconnaissance according to a preset period without starting broadcasting.

[0135] When a Level 3 warning is generated, it primarily serves as a warning of potential risks, alerting vessels and personnel within the waterway to take precautions. A progressive warning strategy involves gradually increasing the intensity and scope of warnings to avoid disruption to normal navigation caused by excessive warnings.

[0136] Warning information scrolls across the airship's surface display screens, including the type of potential risk and the likely location. Vessel personnel can view the warning information by observing the screens. Drones are controlled to conduct low-altitude reconnaissance at preset intervals, providing real-time monitoring of the waterway and identifying potential risk factors. During this phase, no broadcast warnings will be initiated to avoid unnecessary panic.

[0137] Step S4025: Monitor the evolution trend of disaster-related characteristics in real time. When a sudden change in the water flow velocity gradient is detected or the increase in obstacle density exceeds the historical average, the existing warning level is raised by one level and the corresponding response strategy is activated.

[0138] Real-time monitoring of the evolving trends in disaster-related characteristics is crucial, as it allows for timely detection of changes in the disaster situation and for the implementation of appropriate measures. Sudden changes in the flow velocity gradient may indicate a sudden shift in flow speed or direction, potentially causing flooding or compromising navigation safety. An increase in obstacle density exceeding the historical average may indicate the presence of more obstacles in the waterway, increasing the risk of collisions.

[0139] If a sudden change in the water velocity gradient is detected, or if the increase in obstacle density exceeds the historical average, the existing warning level will be raised by one level. For example, if the current warning level is level 3, it will be raised to level 2; if it is level 2, it will be raised to level 1. At the same time, the corresponding response strategy will be activated, such as triggering a channel-wide strobe warning mode for a level 1 warning and a directional warning mode for a level 2 warning.

[0140] Step S4026: During the warning downgrade process, the operating status of the core monitoring nodes at the previous level is retained until it is confirmed that the disaster characteristics have stabilized and subsided, and a composite monitoring strategy for the downgrade transition period is generated and synchronously updated to all mobile nodes.

[0141] After the disaster situation has eased, the warning level needs to be downgraded. However, during this downgrade, monitoring and prevention measures at the previous level cannot be immediately discontinued. The core monitoring nodes at the previous level must remain operational to ensure timely detection of a resurgence of the disaster situation. Core monitoring nodes, such as some airships and drones, are key monitoring equipment and nodes in the disaster monitoring and early warning process.

[0142] Generate a composite monitoring strategy for the downgrade transition period. This strategy integrates multiple monitoring methods and approaches. For example, during the downgrade process, drones can continue to conduct low-altitude reconnaissance while satellite remote sensing modules can be used for macro-monitoring of the river basin. This composite monitoring strategy is then synchronized and updated to all mobile nodes, ensuring that all mobile nodes (such as airships, drones, and ships) are able to operate according to the new monitoring strategy to ensure river basin safety.

[0143] As an embodiment, step S400 inputs a multi-dimensional disaster correlation feature set into a preset disaster warning model for dynamic analysis, generates a disaster warning signal, and determines the disaster type and impact range. Specifically, the following steps S410 to S460 may be included:

[0144] Step S410: Match the water flow velocity gradient change characteristics with the water flow patterns in the historical flood disaster database section by section, extract the time correlation between the flow velocity mutation interval and the historical disaster events, and generate a flood disaster probability distribution map.

[0145] The historical flood disaster database records information about past flood disasters, including water flow patterns, disaster occurrence time, and impact area. The purpose of matching the flow velocity gradient characteristics with the flow patterns in the historical flood disaster database is to identify similarities between current water flow conditions and historical flood flow patterns.

[0146] The segment-by-segment matching process can utilize pattern matching algorithms, such as the Dynamic Time Warping (DTW) algorithm. This algorithm calculates the similarity between two time series while taking into account differences in their lengths. Through segment-by-segment matching, the temporal correlation between velocity mutation intervals and historical disaster events can be extracted. Velocity mutation intervals are intervals where water velocity experiences significant changes within a short period of time, and these intervals are likely to be closely associated with flood disasters.

[0147] Based on the matching results and temporal correlation, a flood probability distribution map is generated. This map, based on geographic space, displays the probability of flood disasters occurring in different regions. For example, based on the frequency of historical flood disasters and the current flow velocity gradient, the probability of flood disasters occurring in each region can be calculated and represented on the map using color depth and other methods.

[0148] Step S420: Based on the concurrent intervals of temperature gradient decrease and humidity surge in the meteorological anomaly correlation characteristics, the meteorological disaster classifier is called to perform time series analysis, identify the disaster triggering threshold of temperature-humidity coordinated changes, and generate meteorological disaster type labels and confidence levels.

[0149] The meteorological disaster classifier is a trained classification model used to classify and identify meteorological disasters. Based on the correlation characteristics of meteorological anomalies, which show periods of simultaneous temperature gradient decreases and humidity surges, time series analysis of temperature and humidity data within these intervals is performed. Time series analysis can employ methods such as the Autoregressive Integrated Moving Average (ARIMA) model or Long Short-Term Memory (LSTM) network to identify patterns and trends in temperature and humidity data.

[0150] Through time series analysis, we identify the disaster triggering thresholds for temperature and humidity co-variations. These thresholds are critical values ​​that, when temperature and humidity changes reach a certain level, could trigger a particular meteorological disaster. Based on the identified disaster triggering thresholds, we call a meteorological disaster classifier to classify the current meteorological conditions and generate a meteorological disaster type label, such as heavy rain or high winds. Furthermore, we calculate a confidence level based on the classification results. This confidence level indicates the reliability of the classification result and is typically expressed as a probability value.

[0151] Step S430: Based on the spatial expansion trend of the obstacle spatial density characteristics and the dynamic density threshold preset in the waterway traffic safety model, the obstacle diffusion rate in the obstacle accumulation area is calculated to generate an emergency level indicator for the obstacle accumulation disaster.

[0152] The spatial expansion trend of the obstacle density characteristic reflects the changes in the obstacle accumulation area in the waterway. By performing a time series analysis of the obstacle density characteristic, the obstacle diffusion rate of the obstacle accumulation area is calculated. The obstacle diffusion rate is the expansion area or volume of the obstacle accumulation area per unit time.

[0153] The dynamic density threshold preset in the waterway safety model is a safety threshold determined based on factors such as the width, depth, and vessel traffic volume of the waterway. When the spatial density of obstacles exceeds the dynamic density threshold, it may impact waterway safety. The dynamic density threshold and the obstacle diffusion rate within the obstacle accumulation area are combined to generate an emergency indicator for the obstacle accumulation disaster. This emergency indicator can be expressed in different levels, such as mild emergency, moderate emergency, and severe emergency, reflecting the degree of threat posed by the obstacle accumulation disaster to waterway safety.

[0154] Step S440: Input the flood disaster probability distribution map, meteorological disaster type label and obstacle urgency mark into the disaster coupling analysis model, and based on the preset disaster chain effect rules, judge the trigger correlation between different disaster types to generate a comprehensive disaster level and potential superimposed impact range.

[0155] The disaster coupling analysis model is used to analyze the interactions and impacts between different disaster types. Flood hazard probability distribution maps, meteorological hazard type labels, and obstacle urgency indicators are input into the disaster coupling analysis model. The model then analyzes the impact based on pre-defined disaster chain reaction rules. These rules are based on triggering relationships and impact mechanisms between different disaster types, as summarized from historical disaster events and related research. For example, floods may cause the accumulation of obstacles, which in turn may affect the speed and direction of water flow, further exacerbating the impact of floods. Meteorological disasters (such as heavy rain) may also trigger floods. The disaster coupling analysis model determines the triggering correlations between different disaster types and generates a comprehensive disaster level and potential overlapping impact range. The comprehensive disaster level, derived by comprehensively considering multiple hazard factors and their coupling relationships, reflects the severity of the disaster. The potential overlapping impact range describes the geographical area and scope that may be affected by the combined effects of different disaster types.

[0156] Step S450: Based on the comprehensive disaster level, the real-time surface temperature data and thermal infrared images of the satellite remote sensing module are called to verify the spatial overlap between the thermal distribution of the disaster-affected area and the multi-dimensional disaster correlation characteristics. If the overlap exceeds the preset verification threshold, a disaster warning signal containing the disaster type label, level identifier and spatial boundary coordinates is generated.

[0157] Real-time surface temperature data and thermal infrared imagery from satellite remote sensing modules can provide information on the thermal distribution of disaster-affected areas. This information can reflect the impact of disasters on the surface environment. For example, floods can cause surface temperatures to drop, while fires can cause surface temperatures to rise.

[0158] Based on the comprehensive disaster level, real-time surface temperature data and thermal infrared imagery from satellite remote sensing modules are used to calculate the spatial overlap between the thermal distribution of the disaster-affected area and the multidimensional disaster correlation characteristics. Spatial overlap is the degree of geographic overlap between the thermal distribution and the disaster-affected area represented by the multidimensional disaster correlation characteristics.

[0159] The preset verification threshold is a pre-defined criterion used to determine the degree of match between the thermal distribution and the multidimensional disaster correlation characteristics. If the overlap exceeds the preset verification threshold, it indicates that the thermal distribution of the disaster-affected area matches the multidimensional disaster correlation characteristics. A disaster warning signal is generated, containing a disaster type label, a level identifier, and spatial boundary coordinates. The disaster type label identifies the specific type of disaster, the level identifier indicates the severity of the disaster, and the spatial boundary coordinates determine the specific geographical scope of the disaster's impact.

[0160] Step S460: If the spatial overlap between the thermal distribution and the multidimensional features does not reach the verification threshold, the matching weights of the water flow velocity gradient change features and the meteorological anomaly correlation features are readjusted, and the disaster coupling analysis is iteratively performed until a disaster warning signal that meets the verification conditions is generated.

[0161] If the spatial overlap between the thermal distribution and the multidimensional features does not reach the validation threshold, the current disaster analysis results may contain errors and the matching weights between the flow velocity gradient characteristics and the meteorological anomaly correlation characteristics need to be readjusted. The matching weights determine the importance of the flow velocity gradient characteristics and the meteorological anomaly correlation characteristics in disaster analysis.

[0162] By adjusting the matching weights, the contribution ratios of the flow velocity gradient characteristics and meteorological anomaly correlation characteristics in the disaster coupling analysis are changed. The disaster coupling analysis is then performed again, including inputting the adjusted characteristics into the disaster coupling analysis model and analyzing it according to the pre-set disaster chain effect rules to generate a new comprehensive disaster level and potential superimposed impact range.

[0163] Repeat the above process, continuously adjust the matching weights and iteratively perform disaster coupling analysis until the spatial overlap between the generated thermal distribution of the disaster-affected area and the multidimensional disaster association characteristics exceeds the preset verification threshold, thus generating a disaster warning signal that meets the verification conditions.

[0164] Step S500: According to the disaster warning signal, the autonomous detachment mechanism of the towed airship and the quick start command of the drone are triggered, and the warning information is broadcast to the ships in the channel through the drone, and the disaster type and impact range are simultaneously transmitted to the command center.

[0165] Disaster warning signals contain crucial information such as the disaster type, level, and spatial boundary coordinates. This information triggers the towed airship's autonomous detachment mechanism. When a disaster strikes, the airship must be released from its towed position to protect its safety. This autonomous detachment mechanism can be achieved by controlling the airship's anchor cable release mechanism and vertical thrusters. For example, upon receiving a flood warning signal, the towed airship is controlled to release its anchor cable and activate its vertical thrusters, allowing it to ascend to a safe altitude.

[0166] At the same time, the drone's quick start command is triggered. The drone slot is pre-loaded with a drone and equipped with a quick charging protocol. When a disaster warning signal is received, the quick charging protocol is activated, rapidly charging the drone and releasing it from the slot.

[0167] Drones can be used to broadcast warnings to ships in the waterway. These drones can be equipped with directional acoustic and optical transmitters. The directional acoustic transmitters broadcast multilingual warnings and simultaneously emit strobe lights corresponding to the disaster severity level to attract the attention of shipboard personnel. Warning information can include the type of disaster, the scope of impact, and response measures.

[0168] The airship is equipped with a multi-band relay module, which transmits disaster warning signals, including disaster type and impact area, to the command center. The command center uses this information to make further decisions and dispatch, coordinate rescue resources, and direct rescue operations.

[0169] As an implementation manner, step S500 may specifically include the following steps S510 to S560:

[0170] Step S510: Analyze the disaster type label in the disaster warning signal. If it is a flood disaster label, control the towed airship to release the anchor cable and activate the vertical thruster. Calculate the airship's ascent path based on the spatial boundary coordinates to avoid the flood impact area and hover to a preset flood control height.

[0171] The purpose of analyzing the disaster type tag in the disaster warning signal is to determine the specific disaster type so that appropriate response measures can be taken. If the disaster type tag indicates flooding, the towed airship must be protected from the impact of the flood. The towed airship is controlled to release the anchoring cable, allowing the airship to break free from its towing position. Simultaneously, the vertical thrusters are activated to provide the airship with upward momentum.

[0172] The airship's ascent path is calculated based on the spatial boundary coordinates from the disaster warning signal. These coordinates define the impact area of ​​the flood. By analyzing these coordinates and the airship's current position, a path planning algorithm (such as the A* algorithm) is used to calculate a safe ascent path, allowing the airship to avoid the flood zone. Finally, the airship is controlled to hover at a preset flood-control altitude, determined based on historical flood data and airship safety requirements, to ensure its safety during floods.

[0173] Step S511: If the obstacle is a piled-up disaster label, the towed airship is controlled to move laterally along the channel, and the airship movement speed is dynamically adjusted based on the obstacle diffusion rate until it reaches the preset obstacle avoidance anchor point and re-anchors.

[0174] When the disaster type is labeled "obstacle accumulation," the towed airship must be controlled to move laterally along the route to avoid collisions. The obstacle diffusion rate reflects the expansion rate of the obstacle accumulation area, and the airship's movement speed is dynamically adjusted based on the obstacle diffusion rate. When the obstacle diffusion rate is fast, the airship's movement speed is increased; when the obstacle diffusion rate is slow, the airship's movement speed is appropriately reduced.

[0175] The preset obstacle avoidance anchor point is a pre-set safe location outside of the obstacle accumulation area. After controlling the airship to move to the preset obstacle avoidance anchor point, re-anchor the airship to ensure the stability of the airship in this position.

[0176] Step S512: During the airship detachment process, the fast charging protocol of the drone slot is started, the spiral search path of the drone is generated according to the geometric shape of the space boundary coordinates, and the multilingual warning voice template and optical signal coding rules corresponding to the disaster type label are loaded.

[0177] During the airship's detachment process, the drone's fast charging protocol is activated to ensure it can perform its early warning mission. This protocol fully charges the drone in a short period of time, ensuring it has sufficient power for flight and early warning operations.

[0178] A spiral search path for the drone is generated based on the geometric shape of the spatial boundary coordinates in the disaster warning signal. This spiral search path enables the drone to conduct comprehensive and efficient search and monitoring within the disaster-affected area. By geometrically analyzing the spatial boundary coordinates, the starting point, radius, pitch, and other parameters of the spiral search path are determined, and the drone's flight path is generated.

[0179] At the same time, the system loads multilingual warning voice templates and optical signal coding rules corresponding to the disaster type label. Different disaster types require different warning information and signals. Based on the disaster type label, the corresponding multilingual warning voice template and optical signal coding rules are selected.

[0180] Step S513: Control the drone to fly along a spiral path, broadcast warning voice through the directional sound wave transmitter, synchronously emit a strobe light signal that matches the disaster level identifier, and collect image data of the disaster area in real time, and transmit it back to the command center through the multi-band relay module of the airship.

[0181] The drone is controlled to fly in a spiral path, enabling it to cover all parts of the disaster-affected area. During flight, a directional acoustic transmitter broadcasts warning messages, ensuring that ships within the waterway receive clear warning information. Simultaneously, a strobe light signal matching the disaster level indicator is emitted, effectively conveying the severity of the disaster.

[0182] The drone also collects real-time image data of the disaster area, including images and videos of the disaster site, providing the command center with intuitive disaster information. The collected image data is transmitted back to the command center via the airship's multi-band relay module, allowing the command center to conduct real-time monitoring and make decisions based on this data.

[0183] Step S514: Receive the rescue priority instruction fed back by the command center. If the coordinates of a ship that has not responded to the warning are detected, dynamically adjust the UAV's spiral search path to a focused tracking path, control the UAV to approach the target ship and activate the emergency strobe mode until a response signal from the ship is obtained.

[0184] The command center formulates rescue priority instructions based on the disaster information and image data it receives and feeds them back to the drones. After receiving the rescue priority instructions fed back by the command center, the drones carry out their work according to the instructions.

[0185] If the coordinates of vessels that did not respond to the warning are detected, it indicates that these vessels may not have received the warning information in a timely manner or failed to take appropriate countermeasures. In this case, the drone's spiral search path is dynamically adjusted to a focused tracking path. A focused tracking path is a search and tracking path specifically targeted at the target vessel, allowing the drone to quickly approach the target vessel.

[0186] After the drone approaches the target vessel, it activates emergency strobe mode. In emergency strobe mode, the drone emits a high-frequency strobe light signal to attract the attention of personnel on the target vessel. The emergency strobe signal continues until a response signal from the vessel is received. The vessel's response signal is transmitted to the drone via wireless communication equipment, indicating that the vessel has received the warning and taken appropriate measures.

[0187] Step S515: After the disaster warning signal is lifted, according to the return command issued by the command center, the drone is controlled to return to the airship slot and enter a low-power standby state. At the same time, the airship is controlled to descend or move sideways to return to the initial towing position, re-anchor and upload the disaster response log.

[0188] When the disaster warning signal is lifted, indicating the disaster situation has been alleviated, the drone and airship must be controlled to return to normal. Following the return command from the command center, the drone is controlled to return to the airship slot. During the return process, the drone follows the pre-set path and lands accurately in the airship slot.

[0189] After returning to the airship's slot, the drone enters a low-power standby state. This reduces energy consumption and extends its lifespan. Simultaneously, the airship is controlled to descend or move sideways back to its initial towing position. If the airship reaches a preset flood control altitude during a disaster, it is controlled to descend. If the airship moves sideways along the waterway to a preset obstacle avoidance anchor point during a disaster, it is controlled to move sideways back to its initial towing position.

[0190] After returning to its initial towing position, the airship re-anchored to ensure stability. At the same time, a disaster response log was uploaded. This log records relevant information from the entire disaster response process, including the triggering time of the warning signal, the trajectory of the airship and drone, and the response of the vessel. Uploading the log provides data support for subsequent disaster analysis and improvements.

[0191] As an implementation method, the method further includes a disaster response collaborative optimization and dynamic feedback mechanism, which may specifically include the following steps S570 to S5120:

[0192] Step S570: Receive the disaster area image data sent back by the UAV, extract the heading angle change rate and speed attenuation characteristics of the ship in the disaster area image data, and generate a ship response behavior data set.

[0193] During its early warning mission, drones collect and transmit real-time image data of the disaster area. After receiving this data, they analyze and process it. Using image processing and object detection algorithms, they identify the vessel's position and posture from the image data of the disaster area.

[0194] Extract the vessel's heading angle change rate and speed decay characteristics. The heading angle change rate reflects the speed at which a vessel changes direction during navigation, while the speed decay characteristic reflects the decrease in speed during a disaster. By comparing and analyzing image data at different times, the vessel's heading angle change rate and speed decay characteristics are calculated.

[0195] Based on the extracted heading angle change rate and speed attenuation characteristics, a vessel response behavior dataset was generated. This vessel response behavior dataset records the response behavior characteristics of each vessel during a disaster, providing data support for subsequent analysis and optimization.

[0196] Step S571: Based on the disaster type label and spatial boundary coordinates, the waterway is divided into the core disaster area, the diffusion impact area, and the safe passage area. Communication channels of different frequency bands are allocated to each area, and an airship-UAV collaborative task table is generated.

[0197] The waterway is divided into zones based on the disaster type label and spatial boundary coordinates in the disaster warning signal. The core disaster zone is the area most severely affected by the disaster, usually the center of the disaster. The diffuse impact zone is the area around the core disaster zone that is affected by the spread of the disaster. The safe passage zone is a relatively safe area that is not directly affected by the disaster.

[0198] Different frequency bands are allocated to each area to prevent interference. These channels ensure stable and efficient communication between airships, drones, and ships in different areas. For example, high-frequency channels are allocated to the core disaster area to ensure communication quality in complex environments, while mid- and low-frequency channels are allocated to the diffusion impact zone and safe passage zone.

[0199] Based on the regional divisions and communication channel allocations, an airship-UAV collaborative task table is generated. This table clearly defines the division of labor and collaboration between airships and UAVs within different areas. For example, in the core disaster zone, UAVs will be responsible for detailed disaster monitoring and early warning information broadcast, while airships will provide communication relay and data transmission support. In the diffusion impact zone, airships and UAVs will jointly monitor the spread of the disaster and provide early warning. In the safe passage zone, UAVs can conduct patrols and information collection, while airships can provide necessary navigation and guidance services.

[0200] Step S572: Based on the obstacle diffusion rate and flood impact direction prediction model, the airship deployment density in the core disaster area is adjusted, and the redundant airships are controlled to move in the predicted diffusion direction and release warning buoys.

[0201] The Obstacle Diffusion Rate and Flood Impact Direction Prediction Model, based on historical and real-time monitoring data, is used to predict the barrier diffusion rate and flood impact direction. Based on the model's predictions, the density of airship deployment in core disaster areas is adjusted.

[0202] If the predicted spread of obstacles is rapid or the flood's impact direction is likely to change, the density of airship deployment in the core disaster area will be increased to strengthen monitoring and early warning capabilities in that area. Redundant airships, which are temporarily not needed in other areas, will be controlled to move in the predicted spread direction. This movement of redundant airships allows for early monitoring and early warning, allowing for timely detection of the spread of the disaster.

[0203] At the same time, the airship was controlled to release warning buoys. Warning buoys are surface-mounted warning devices that emit audible and visual signals to alert ships to safety. Releasing warning buoys in the predicted direction of spread can expand the warning range and improve its effectiveness.

[0204] Step S573: Analyze the coordinates of the non-responding warning ships in the ship response behavior dataset, call the airships within the preset range to release the backup drone cluster, and use the ring encirclement strategy to cover the target ship with multi-angle warning signals.

[0205] The coordinates of vessels that did not respond to warnings are parsed from the vessel response behavior dataset to identify the locations of vessels that did not respond promptly to warnings. Airships within a preset range are then called upon to deploy a backup drone swarm. The preset range is a geographical area determined based on actual conditions within which airships can quickly respond and deploy a backup drone swarm. The backup drone swarm is a pre-prepared group of drones to respond to vessels that do not respond to warnings.

[0206] A ring-shaped encirclement strategy is employed to provide multi-angle warning signal coverage to the target vessel. This involves a swarm of backup drones forming a ring around the target vessel, sending warning signals from multiple angles. The drones broadcast warning messages and simultaneously emit radio frequency flash signals via directional acoustic transmitters, ensuring that personnel on the target vessel receive clear and strong warning signals from multiple directions. This multi-angle warning signal coverage enhances the effectiveness of the warning and increases the likelihood that the target vessel will respond.

[0207] Step S574: Real-time monitoring of the displacement data of the warning buoy and the warning feedback rate of the drone cluster. If abnormal drift of the buoy is detected or the warning feedback rate is lower than the preset index, the secondary separation mechanism of the airship is triggered and the drone search radius is expanded.

[0208] Real-time monitoring of the displacement data of warning buoys. Warning buoys are typically equipped with positioning equipment that transmits their position information in real time. By analyzing the displacement data of the warning buoys, it can be determined whether they are drifting abnormally. Abnormal drift may indicate changes in water flow conditions or other external forces, which may affect the warning buoy's effectiveness.

[0209] At the same time, the drone swarm's warning response rate is monitored in real time. This rate is the percentage of target vessels responding to drone warning signals. It is calculated by counting the number of vessels receiving a reply signal relative to the total number of target vessels. If abnormal buoy drift is detected or the warning response rate falls below a preset threshold, it indicates that current warning measures may be ineffective and further action is needed.

[0210] Triggering the airship's secondary disengagement mechanism means it will again detach from its current location, adjusting its position and attitude to better respond to disasters. Simultaneously, the drone's search radius and range will be expanded to detect more vessels that haven't responded to warnings or to obtain more comprehensive disaster information. These measures can improve the effectiveness of warnings and disaster response capabilities.

[0211] Step S575: After the disaster warning signal is lifted, a channel communication network optimization plan and an airship deployment heat map are generated based on the communication channel load data and airship displacement trajectory of each area, and uploaded to the command center for subsequent disaster prevention planning.

[0212] After the disaster warning signal is lifted, data from the entire disaster response process is analyzed and summarized. Communication channel load data for each region is collected. This data reflects the usage of communication channels in different areas during the disaster response, including information such as communication traffic volume and signal strength. The displacement trajectory of the airship is also recorded, reflecting its movement and operating range during the disaster response.

[0213] Based on the communication channel load data for each region, the efficiency and bottleneck issues of communication channels are analyzed to generate an optimization plan for the waterway communication network. This optimization plan may include adjusting the communication channel allocation strategy, adding communication relay equipment, optimizing communication protocols, and other measures to improve the reliability and stability of the waterway communication network.

[0214] Based on the airship's displacement trajectory, an airship deployment heat map is generated. This geospatial heat map shows the density and frequency of airship deployments in different regions. By analyzing this heat map, we can understand the use and rationality of airship distribution during disaster response, providing a reference for subsequent airship deployment.

[0215] The channel communication network optimization plan and airship deployment heat map are uploaded to the command center, which can then use this information to make subsequent disaster prevention plans. For example, the channel communication network can be improved and upgraded according to the channel communication network optimization plan, and the initial deployment location and number of airships can be adjusted according to the airship deployment heat map to improve the efficiency of disaster monitoring and early warning.

[0216] As an embodiment, the method further includes a dynamic credibility assessment and self-repair mechanism for multi-source monitoring data, specifically including the following steps S5130 to S5180:

[0217] Step S5130: Acquire the position data and channel image data of adjacent airship nodes within the same time period, calculate the position correlation coefficient and image content matching degree between the nodes, and generate the airship data credibility score.

[0218] Acquire the location data and channel image data of adjacent airship nodes within the same time period. The same time period is to ensure the temporal consistency of the data and accurately reflect the monitoring situation at the same moment. Adjacent airship nodes are airships that are adjacent in geographical space, and their monitoring data are correlated.

[0219] Calculate the position correlation coefficient between nodes. This coefficient measures the similarity between the position data of adjacent airship nodes. This can be calculated using methods such as the Pearson correlation coefficient. The coefficient ranges from -1 to 1. Values ​​closer to 1 indicate a stronger correlation between the position data of two nodes, while values ​​closer to -1 indicate a weaker correlation.

[0220] At the same time, the image content matching degree is calculated. This degree measures the similarity between the channel image data collected by adjacent airship nodes. A feature matching algorithm, such as the SIFT algorithm or the ORB algorithm, can be used to extract and match features from the channel images of adjacent airship nodes. The ratio of the number of matched feature points to the total number of feature points is calculated as the image content matching degree.

[0221] The airship data credibility score is generated based on the position correlation coefficient and the image content match. The airship data credibility score can be obtained by setting a weight coefficient and taking a weighted sum of the position correlation coefficient and the image content match. For example, if the position correlation coefficient has a weight of 0.6 and the image content match has a weight of 0.4, the airship data credibility score = 0.6 × position correlation coefficient + 0.4 × image content match. A higher airship data credibility score indicates a more reliable data source for that airship node.

[0222] Step S5131: Perform multi-frame continuity analysis on the channel image data collected by the drone, extract the pixel change gradient of adjacent frame images, verify the cause of image blur in combination with the wind speed data in the environmental parameters, and generate the image quality credibility level.

[0223] Multi-frame continuity analysis is performed on the airway image data collected by the drone. This analysis analyzes multiple frames of continuously acquired airway images to detect moving objects and changes in the images. By comparing the pixel values ​​of adjacent frames, the pixel change gradients between the frames are extracted. The pixel change gradient reflects the speed and direction of pixel value changes in the image and can demonstrate the dynamic changes in the image.

[0224] Combined with wind speed data from environmental parameters, the cause of image blur can be verified. Image blur can be caused by factors such as drone vibration during flight or excessive wind speed. High wind speeds can cause the drone to vibrate, resulting in blurred images. By analyzing the relationship between wind speed data and pixel gradients, the cause of image blur can be determined.

[0225] Based on the analysis results, an image quality confidence level is generated. This level can be categorized as high, medium, or low. For example, when the pixel gradient is small and the wind speed is normal, the image quality confidence level is high; when the pixel gradient is large and the wind speed is high, the image quality confidence level is low. The image quality confidence level can provide a reference for subsequent data analysis and processing. For image data with a low confidence level, appropriate processing measures such as filtering and deblurring can be implemented.

[0226] Step S5132: Based on the environmental parameter data transmitted by the satellite remote sensing module, a parameter difference matrix of multiple satellite data sources in the same geographic grid is constructed, the deviation degree of abnormal data points is verified based on the historical data fitting curve, and an environmental parameter credibility mark is generated.

[0227] Based on the environmental parameter data transmitted by satellite remote sensing modules, a geographic grid divides the river basin into multiple equally sized geographic regions, each corresponding to a geographic grid. Within the same geographic grid, multiple satellite data sources may provide environmental parameter data, and these data may differ.

[0228] Construct a parameter difference matrix for multiple satellite data sources within the same geographic grid. This matrix records the differences between environmental parameter data provided by different satellite data sources. This matrix is ​​constructed by calculating the parameter differences between different satellite data sources. For example, for temperature parameters, the differences between temperature values ​​provided by different satellite data sources are calculated and organized into a matrix.

[0229] Verify the degree of deviation of abnormal data points based on a fitted curve of historical data. Historical data records the past changes in environmental parameters within the same geographic grid. By analyzing and fitting historical data, a normal variation curve of environmental parameters can be obtained. Compare the current environmental parameter data with the fitted curve to determine whether each data point is an abnormal data point and calculate the degree of deviation of the abnormal data point.

[0230] Based on the degree of deviation of the abnormal data point, an environmental parameter credibility indicator is generated. This indicator can be represented by different symbols or colors. For example, data points with a small degree of deviation are marked as credible, while data points with a large degree of deviation are marked as suspicious or unreliable. This indicator can help users quickly identify the reliability of environmental parameter data and allow for further verification and processing of unreliable data points.

[0231] Step S5133: Input the airship data credibility score, image quality credibility level and environmental parameter credibility identifier into the data fusion decision model to adjust the weight ratio of different data sources in the standardized monitoring data set.

[0232] The data fusion decision model comprehensively considers the credibility of different data sources to integrate and make decisions. The data fusion decision model inputs the airship data credibility score, image quality credibility level, and environmental parameter credibility identifiers. Based on this credibility information, the model adjusts the weights of different data sources within the standardized monitoring dataset.

[0233] For data sources with higher credibility, their weight in the standardized monitoring dataset is increased to give them a greater impact on the final monitoring results. For data sources with lower credibility, their weight is reduced to minimize their impact on the monitoring results. For example, if a certain airship node has a high credibility score, the weight of the location data and channel image data provided by the airship node will be increased during the data fusion process. If the credibility of the environmental parameter data provided by a certain satellite data source is marked as unreliable, the weight of the data provided by this data source in the standardized monitoring dataset will be reduced.

[0234] By adjusting the weight ratios of different data sources, the quality and reliability of standardized monitoring datasets can be improved, making the monitoring results more accurately reflect the actual conditions of river basins.

[0235] Step S5134: If it is detected that the credibility score of any airship node is continuously lower than the average value of the adjacent nodes, the node data replacement mechanism is activated, and a spatiotemporal interpolation algorithm is used to generate an alternative data set and mark it as data to be verified.

[0236] During continuous monitoring, if the credibility score of any airship node is consistently lower than the average of its neighboring nodes, this indicates that there may be issues with the data of that airship node and that it needs to be processed. A node data replacement mechanism is initiated to ensure the quality of the standardized monitoring dataset.

[0237] A surrogate dataset is generated using a spatiotemporal interpolation algorithm. This algorithm estimates and predicts unknown data points based on the values ​​of known data points. In this step, the data of the adjacent airship nodes are interpolated using the spatiotemporal interpolation algorithm based on their location data, channel image data, and temporal and spatial information to generate a surrogate dataset.

[0238] The generated alternative dataset is marked as pending data. This means that the dataset is generated through an interpolation algorithm and requires further verification and validation. During subsequent processing, the alternative dataset is verified to ensure its accuracy and reliability.

[0239] Step S5135: After generating the disaster warning signal, the alternative data set is cross-validated with the original data. If the difference exceeds the preset tolerance, a manual review instruction is triggered and the credibility assessment model parameters are updated.

[0240] After generating a disaster warning signal, the alternative dataset is cross-validated against the original data. Cross-validation involves comparing and analyzing the alternative dataset with the original data to examine any differences. The degree of similarity between the alternative dataset and the original data is assessed by calculating metrics such as the difference and correlation coefficient.

[0241] If the difference exceeds the preset tolerance, which is a pre-defined range of allowable error, it indicates a significant discrepancy between the alternative dataset and the original data, potentially indicating a problem. A manual review is triggered, notifying relevant personnel to manually review the data for that airship node. This manual review can include checking the operating status of the equipment and recollecting data to confirm the accuracy of the data.

[0242] At the same time, the credibility assessment model parameters are updated. The credibility assessment model is used to assess the credibility of each data source. Based on the results of this cross-validation, the credibility assessment model parameters are adjusted to improve the model's accuracy and reliability. For example, if the interpolated data of a particular airship node differs significantly from the original data, this indicates that the credibility assessment of that node may be inaccurate, and the credibility assessment parameters for that node need to be adjusted.

[0243] As an implementation manner, the method provided in the embodiment of the present invention further includes an intelligent optimization process in the post-disaster recovery phase, which may specifically include the following steps S600 to S1100:

[0244] Step S600: After the disaster warning signal is lifted, the turbulence intensity data of the waterway flow and the residual density of floating objects on the water surface are continuously collected, and the waterway recovery degree evaluation index is generated in combination with the anchoring stability data of the airship.

[0245] After the disaster warning signal is lifted, the post-disaster recovery phase begins. Turbulence intensity data is continuously collected from the waterway. Turbulence intensity reflects the degree of water disturbance. Turbulence sensors installed in the waterway can collect turbulence intensity data in real time. Simultaneously, the residual density of floating debris on the water surface is collected. This represents the number or mass of floating debris per unit area of ​​water. This can be analyzed and calculated using waterway images captured by drones.

[0246] Combined with the airship's anchoring stability data, which can be obtained through sensors installed on the airship, including information such as the tension of the anchoring cable and the airship's attitude changes, this anchoring stability data reflects the airship's stability after a disaster and indirectly reflects the recovery of the waterway environment.

[0247] The channel recovery index is generated based on the turbulence intensity data of the waterway flow, the residual density of floating debris on the water surface, and the anchoring stability data of the airship. By setting a weight coefficient, this data can be weighted and summed to obtain the channel recovery index.

[0248] Step S601: Analyze the airship displacement trajectory data throughout the disaster response cycle, extract the channel sections where trajectory corrections frequently occur, generate an airship deployment density optimization plan, and mark the coordinates of potential blind spots.

[0249] Analyze airship trajectory data throughout the entire disaster response cycle, from the time a disaster warning signal is triggered to its release. This data identifies sections of the waterway where frequent trajectory corrections occur. These corrections can be caused by obstacles, changes in water flow, and other factors. Frequent corrections indicate a complex environment requiring increased monitoring and attention.

[0250] Based on the analysis results, an optimized plan for airship deployment density is generated. For sections of the route where trajectory corrections are frequent, the airship deployment density is increased to improve monitoring capabilities. For sections of the route where trajectories are more stable, the airship deployment density can be appropriately reduced. Furthermore, the coordinates of potential blind spots are marked. Potential blind spots are areas that are beyond the reach of airships due to terrain, obstacles, and other factors. This marking of potential blind spot coordinates provides a reference for subsequent monitoring and improvement, allowing appropriate measures to be taken to eliminate potential blind spots.

[0251] Step S602: Based on the charging times and flight range data of the drones in disaster response, a slot charging efficiency attenuation model is established to dynamically adjust the drone rotation scheduling strategy of different airship nodes.

[0252] Based on the number of charging times and flight range data of drones during disaster response, the number of charging times reflects the drone's energy consumption during the disaster response process, while the flight range data reflects the drone's flight distance and range. By analyzing this data, a slot charging efficiency decay model was established. The slot charging efficiency decay model describes how the slot charging efficiency changes with usage and time. This gradual decrease in charging efficiency may be caused by factors such as battery aging and charging equipment wear.

[0253] Based on the slot charging efficiency decay model, the rotation scheduling strategy for drones at different airship nodes is dynamically adjusted. For airship nodes with slots with low charging efficiency, the number of drones charging at those nodes is reduced, while the number of drones charging at other nodes with higher charging efficiency is increased. At the same time, drone flight missions and ranges are rationally scheduled to avoid excessive energy consumption, thereby improving drone utilization efficiency and overall disaster response capabilities.

[0254] Step S603: Integrate historical disaster data with the current response log, reconstruct a virtual disaster scenario, and inject it into the disaster warning model for stress testing to identify the weak links in the model's response under complex disaster scenarios.

[0255] By integrating historical disaster data, which records information about past disasters, with current response logs, we can obtain more comprehensive and accurate disaster data.

[0256] Reconstruct virtual disaster scenarios using computer simulation technology based on the integrated disaster data. Virtual disaster scenarios can simulate different types of disasters and their coupling, such as the superposition of flood disasters and meteorological disasters, and the mutual influence of obstacle accumulation disasters and water flow changes.

[0257] Reconstructed virtual disaster scenarios are injected into the disaster warning model for stress testing. Stress testing tests the model's performance and reliability under extreme or complex circumstances. Through stress testing, weaknesses in the model's response to complex disaster scenarios are identified. For example, the model may be unable to accurately predict the occurrence and development of disasters in the event of multiple disasters, or it may be unable to promptly adjust warning levels and rescue resource allocation strategies during disaster response.

[0258] Based on the identified weak links in response, the disaster warning model is improved and optimized to enhance the performance and reliability of the model in complex disaster scenarios.

[0259] Step S604: Generate a set of suggestions for improving the waterway infrastructure, including adding auxiliary positioning buoys in sections with frequent trajectory corrections, upgrading wireless charging modules at nodes with reduced charging efficiency, and deploying underwater monitoring robots for warning blind spots.

[0260] Based on the previous analysis and test results, a set of recommendations for improving waterway infrastructure was generated. Auxiliary positioning buoys were added in sections where track corrections were frequent. These buoys could provide more accurate position reference information, helping airships and ships navigate and position themselves better, reducing the number of track corrections and improving navigation safety.

[0261] Upgrading the wireless charging module at the point where charging efficiency decreases can improve charging efficiency and stability, reduce drone charging time, and increase drone usage efficiency. At the same time, it can extend the battery life and reduce maintenance costs.

[0262] Underwater monitoring robots are deployed to address warning blind spots. They can monitor underwater conditions in waterways, filling these blind spots and providing more comprehensive waterway information. Equipped with a variety of sensors, such as sonar and cameras, they can collect real-time information on underwater obstacles and currents, transmitting the data back to the command center.

[0263] Step S605: The optimized early warning model parameters, infrastructure improvement plan and scheduling strategy are packaged to generate an intelligent upgrade package, which is distributed to all airship nodes via satellite links and the nighttime silent deployment mode is activated.

[0264] The optimized warning model parameters, infrastructure improvement plans, and dispatch strategies are integrated into a packaged intelligent upgrade. The package includes improvements and optimizations to the disaster warning model, waterway infrastructure, and dispatch strategies.

[0265] The smart upgrade package is distributed to all airship nodes via satellite links. Satellite links have the advantages of wide coverage and fast transmission speed, which can ensure that the smart upgrade package is transmitted to each airship node quickly and accurately.

[0266] Activate nighttime silent deployment mode. Nighttime is typically a time when waterway monitoring tasks are relatively low. Choosing to perform upgrade deployments at night can minimize the impact on normal monitoring operations. In this silent deployment mode, the airship node automatically downloads and installs the intelligent upgrade package, updating warning model parameters, implementing infrastructure improvement plans, and adjusting scheduling strategies. After the upgrade is complete, the system automatically performs testing and verification to ensure the upgraded system is functioning properly.

[0267] It is understandable that the various algorithms involved in the above-mentioned introductions of the embodiments of the present invention, such as curve fitting algorithms, dark channel prior dehazing algorithms, SIFT algorithms, ORB algorithms, interpolation algorithms, etc., can all be learned from the relevant content in the prior art. In order to save space, they will not be expanded too much in the embodiments of the present invention. In addition, when implementing the scheme of the present invention, those skilled in the art can supplement the details according to the common knowledge in this field. For example, according to the common knowledge in this field, normalization can be used to eliminate dimensional conflicts before feature fusion, interpolation can be used to eliminate dimensional differences, and historical data, experience or business scenario requirements can be combined to reasonably set thresholds, and models can be trained based on a general model training method, etc. The present invention will no longer provide redundant introductions to overly detailed implementation processes.

[0268] Figure 2 A hardware entity diagram of a computer system provided by an embodiment of the present invention is as follows Figure 2As shown, the hardware entity of the computer system 1000 includes: a processor 1001 and a memory 1002, wherein the memory 1002 stores a computer program that can be run on the processor 1001, and when the processor 1001 executes the program, the steps in the method of any of the above embodiments are implemented.

Claims

1. A river basin disaster monitoring and warning method based on multi-source monitoring data analysis, characterized in that: The following steps are involved: Acquiring multi-source monitoring data for river basins, including real-time position data from towed airships, channel image data collected by drones, and environmental parameter data transmitted by satellite remote sensing modules; Performing data preprocessing on the multi-source monitoring data to eliminate noise interference in the real-time position data, and performing time synchronization alignment on the waterway image data and environmental parameter data to generate a standardized monitoring data set; Based on the standardized monitoring data set, the water flow velocity gradient variation characteristics, meteorological anomaly correlation characteristics, and waterway obstacle distribution characteristics of the river basin are extracted to generate a multi-dimensional disaster correlation feature set, wherein the waterway obstacle distribution characteristics include the projected area and spatial density characteristics of the obstacles; Matching the water flow velocity gradient variation characteristics with the water flow patterns in the historical flood disaster database section by section, extracting the temporal correlation between the velocity mutation interval and the historical disaster events, and generating a flood disaster probability distribution map; Based on the concurrent intervals of temperature gradient decrease and humidity surge in the meteorological anomaly correlation characteristics, a meteorological disaster classifier is called to perform time series analysis, identify the disaster triggering threshold of temperature-humidity coordinated change, and generate a meteorological disaster type label and confidence level; Based on the spatial expansion trend of the obstacle density characteristics and the dynamic density threshold preset in the waterway traffic safety model, the obstacle diffusion rate in the obstacle accumulation area is calculated to generate an emergency level indicator for the obstacle accumulation disaster; The flood disaster probability distribution map, meteorological disaster type label, and obstacle urgency identifier are input into a disaster coupling analysis model. Based on the preset disaster chain effect rules, the trigger correlation between different disaster types is determined to generate a comprehensive disaster level and potential superimposed impact range; the disaster coupling analysis model is used to analyze the disaster effects and impacts between different disaster types; Based on the comprehensive disaster level, real-time surface temperature data and thermal infrared images from a satellite remote sensing module are used to verify the spatial overlap between the thermal distribution of the disaster-affected area and the multi-dimensional disaster correlation characteristics. If the overlap exceeds a preset verification threshold, a disaster warning signal containing a disaster type label, a level identifier, and spatial boundary coordinates is generated. If the spatial overlap between the thermal distribution and the multidimensional features does not reach the verification threshold, the matching weights of the water flow velocity gradient change features and the meteorological anomaly correlation features are readjusted, and the disaster coupling analysis is iteratively performed until a disaster warning signal that meets the verification conditions is generated; According to the disaster warning signal, the autonomous detachment mechanism of the towed airship and the quick start instruction of the drone are triggered, and the warning information is broadcast to the ships in the waterway through the drone, and the disaster type and impact range are simultaneously transmitted to the command center.

2. The method according to claim 1, characterized in that The data preprocessing of the multi-source monitoring data, eliminating noise interference in the real-time position data, and time synchronization alignment of the channel image data and environmental parameter data to generate a standardized monitoring data set includes: Performing dynamic drift compensation processing on the real-time position data of the towed airship, calculating the airship position offset vector based on the airship's inertial navigation data and a preset course coordinate reference point, and generating corrected position data through reverse displacement superposition; Adaptively defogging the channel image data collected by the drone, adjusting the intensity parameters of the defogging algorithm according to the real-time visibility value in the environmental parameter data, extracting the contour information of the channel boundary and obstacles in the image, and generating enhanced channel image data; Performing outlier repair processing on the environmental parameter data transmitted by the satellite remote sensing module, replacing data points that exceed a preset fluctuation range based on the distribution pattern of environmental parameters of adjacent airship monitoring nodes within the same time period, and generating a repaired environmental parameter sequence; The corrected position data, enhanced channel image data and repaired environmental parameter sequence are subjected to spatiotemporal alignment processing. According to the mapping relationship between the airship position and the drone aerial photography timestamp, the data of the missing time window is interpolated and supplemented to generate a standardized monitoring data set containing complete spatiotemporal labels.

3. The method according to claim 2, characterized in that The method extracts the flow velocity gradient variation characteristics, meteorological anomaly correlation characteristics, and waterway obstacle distribution characteristics of the river basin to generate a multi-dimensional disaster correlation feature set, including: Based on the corrected position data, the displacement acceleration of the towed airship in continuous time intervals is calculated, and the water velocity gradient variation characteristics are deduced by combining the waterway depth data and the airship draft; Extracting the coordinated change trend of temperature and humidity from the repaired environmental parameter sequence, identifying the concurrent intervals of temperature gradient decrease and humidity surge, and generating meteorological anomaly correlation features; Performing multi-scale obstacle segmentation processing on the enhanced waterway image data, distinguishing floating objects, fixed obstacles, and dynamic targets such as ships based on contour information, calculating the projected area and spatial density of obstacles in combination with the drone altitude data, and generating a distribution feature of waterway obstacles; The water flow velocity gradient change characteristics, meteorological anomaly correlation characteristics and obstacle spatial density characteristics are temporally and spatially superimposed and fused, and a multidimensional disaster correlation feature set containing temporal and spatial correlation is generated based on a preset disaster feature weight table.

4. The method according to claim 1, wherein The method includes triggering the autonomous detachment mechanism of the towed airship and the rapid start-up command of the drone based on the disaster warning signal, broadcasting warning information to ships in the waterway through the drone, and simultaneously transmitting the disaster type and impact range to the command center, including: parsing the disaster type label in the disaster warning signal; if it is a flood disaster label, controlling the towed airship to release the anchor cable and activate the vertical thruster, calculating the airship's ascent path based on the spatial boundary coordinates, avoiding the flood impact area and hovering to a preset flood control altitude; If the obstacle accumulation disaster label is set, the towed airship is controlled to move laterally along the route, and the airship movement speed is dynamically adjusted based on the obstacle diffusion rate until it reaches the preset obstacle avoidance anchor point and re-anchors; During the airship's separation process, the rapid charging protocol of the drone slot is activated, a spiral search path for the drone is generated based on the geometric shape of the spatial boundary coordinates, and multilingual warning voice templates and optical signal coding rules corresponding to the disaster type label are loaded; The drone is controlled to fly along a spiral path, broadcasting warning voices through a directional acoustic transmitter, simultaneously emitting a strobe light signal that matches the disaster level, and collecting real-time image data of the disaster area, which is then transmitted back to the command center via the airship's multi-band relay module. Receive rescue priority instructions from the command center. If the coordinates of a ship that has not responded to the warning are detected, the drone will dynamically adjust its spiral search path to a focused tracking path, control the drone to approach the target ship, and activate emergency strobe mode until it receives a response signal from the ship. After the disaster warning signal is lifted, according to the return command issued by the command center, the drone is controlled to return to the airship slot and enter a low-power standby state. At the same time, the airship is controlled to descend or move sideways to return to the initial towing position, re-anchor and upload the disaster response log.

5. The method according to claim 4, characterized in that The method also includes a disaster response collaborative optimization and dynamic feedback mechanism: receiving the disaster area image data transmitted by the UAV, extracting the heading angle change rate and speed attenuation characteristics of the ship in the disaster area image data, and generating a ship response behavior data set; Based on the disaster type label and spatial boundary coordinates, the waterway is divided into a core disaster area, a diffusion impact area, and a safe passage area. Communication channels of different frequency bands are allocated to each area, and an airship-UAV collaborative task table is generated. Based on the obstacle diffusion rate and flood impact direction prediction model, the airship deployment density in the core disaster area is adjusted, and redundant airships are controlled to move in the predicted diffusion direction and release warning buoys; Analyze the coordinates of the non-responding warning ships in the ship response behavior dataset, call airships within a preset range to release a backup drone cluster, and use a ring-shaped encirclement strategy to cover the target ship with multi-angle warning signals; Real-time monitoring of the displacement data of the warning buoy and the warning feedback rate of the drone cluster. If abnormal buoy drift is detected or the warning feedback rate falls below a preset indicator, the airship's secondary disengagement mechanism is triggered and the drone search radius is expanded. The warning feedback rate is the response rate of target ships to the drone's warning signal, calculated by counting the ratio of ships that receive the reply signal to the total number of target ships. After the disaster warning signal is lifted, a channel communication network optimization plan and airship deployment heat map are generated based on the communication channel load data and airship displacement trajectory of each area, and uploaded to the command center for subsequent disaster prevention planning.

6. The method according to claim 5, characterized in that The method also includes a dynamic credibility assessment and self-repair mechanism for multi-source monitoring data: Obtain the position data and channel image data of adjacent airship nodes within the same period, calculate the position correlation coefficient and image content matching degree between the nodes, and generate the airship data credibility score; Performing multi-frame continuity analysis on the channel image data collected by the UAV, extracting pixel change gradients of adjacent frame images, verifying the cause of image blur in combination with wind speed data in environmental parameters, and generating an image quality credibility level; Based on the environmental parameter data transmitted by the satellite remote sensing module, a parameter difference matrix of multiple satellite data sources in the same geographic grid is constructed, the deviation degree of abnormal data points is verified based on the historical data fitting curve, and an environmental parameter credibility mark is generated; Inputting the airship data credibility score, image quality credibility level and environmental parameter credibility identifier into the data fusion decision model, and adjusting the weight ratio of different data sources in the standardized monitoring data set; If it is detected that the credibility score of any airship node is continuously lower than the average value of the adjacent nodes, the node data replacement mechanism is activated, and a spatiotemporal interpolation algorithm is used to generate an alternative data set and mark it as data to be verified; After the disaster warning signal is generated, the alternative data set is cross-validated with the original data. If the difference exceeds the preset tolerance, a manual review instruction is triggered and the credibility assessment model parameters are updated.

7. The method according to claim 2, characterized in that The method of performing dynamic drift compensation processing on the real-time position data of the towed airship, calculating the airship position offset vector based on the airship's inertial navigation data and a preset course coordinate reference point, and generating corrected position data by reverse displacement superposition includes: Obtain the towed airship's inertial navigation data, including the airship's acceleration recorded by the three-axis accelerometer and the heading angle change rate recorded by the gyroscope, and combine this with the initial coordinates of the airship's anchor point to generate the original airship trajectory data; Extracting a preset sequence of channel coordinate reference points, which includes the latitude and longitude coordinates of fixed lighthouses on both sides of the channel and their corresponding visual feature points, using the airship's onboard camera to capture the channel boundary image in real time and match the visual feature points to generate visual positioning deviation data for the airship's current position relative to the reference points; The original airship trajectory data and the visual positioning deviation data are aligned in time and space, an airship instantaneous motion model is constructed based on the airship motion acceleration and heading angle change rate, and the cumulative displacement error of the airship caused by wind and water flow in a continuous time window is calculated to generate an airship position offset vector; performing reverse compensation on the original airship trajectory data according to the airship position offset vector, using the visual positioning deviation data as a correction constraint, performing track fitting optimization on the compensated trajectory, eliminating the drift component of the inertial navigation data, and generating corrected position data; The corrected position data and the channel coordinate reference point are subjected to residual analysis. If the residual value exceeds a preset tolerance range, the visual feature points are re-matched and the dynamic drift compensation process is iteratively performed until the corrected position data that meets the accuracy requirements is generated.

8. The method according to claim 2, characterized in that The adaptive defogging process is performed on the channel image data collected by the UAV, the intensity parameter of the defogging algorithm is adjusted according to the real-time visibility value in the environmental parameter data, the contour information of the channel boundary and obstacles in the image is extracted, and the enhanced channel image data is generated, including: Obtaining the original channel image data collected by the UAV and the real-time visibility value in the environmental parameter data, dividing the image fog concentration level based on the real-time visibility value, and determining the defogging intensity coefficient corresponding to the different levels; Dynamically adjust the transmittance estimation parameter of the adaptive defogging algorithm based on the defogging intensity coefficient, perform multi-scale filtering on the original channel image data, separate the high-frequency details from the low-frequency fog background in the image, and generate a preliminary defogging image, wherein the defogging intensity coefficient is used to adjust the intensity of the defogging algorithm, and the transmittance estimation parameter is a parameter that affects the defogging effect and is proportional to the defogging intensity coefficient; Performing local contrast enhancement processing on the preliminary dehazed image, adaptively adjusting the brightness compensation weight based on the geometric feature distribution of the channel boundary, suppressing overexposed areas and enhancing dark textures, and generating a balanced intermediate image; Extracting edge gradient information of the channel boundary and obstacles from the equalized intermediate image, fusing the gradient amplitudes of consecutive pixels using a multi-directional edge detection algorithm, and generating a contour mask containing the contour of the channel boundary and the shape of the obstacle; The contour mask is superimposed and fused with the equalized intermediate image, and the image sharpening intensity is optimized according to the continuity characteristics of the channel boundary in the contour mask, the residual artifacts of defogging are eliminated, and the contrast of the obstacle edge is enhanced to generate enhanced channel image data; The enhanced channel image data is subjected to contour integrity verification. If a channel boundary break or a blurred obstacle contour is detected, the defogging intensity coefficient is readjusted and adaptive defogging processing is iteratively performed until enhanced channel image data that meets the recognition requirements is generated.

9. A computer system comprising a memory and a processor, wherein the memory stores a computer program that can be run on the processor, wherein: When the processor executes the program, the steps of the method according to any one of claims 1 to 8 are implemented.

Images