Dam landslide danger area identification system and method based on unmanned aerial vehicle and remote sensing image

Abstract

The application discloses a dam landslide danger area identification system and method based on a UAV and remote sensing images, and belongs to the technical field of dam slope detection. The application comprises a task planning module, a data acquisition module, a feature extraction module, a region division module and a danger release module. Important features of key parts of a dam are highlighted and focused, shallow features are constructed into a graph structure, local feature information is expanded into deep semantic features containing global semantic information through feature propagation and aggregation between nodes, spatial multi-scale information and time change trends of deep semantic features are captured, the dynamic change process of the dam can be more comprehensively described, feature patterns related to landslides are identified, a probability value of belonging to a landslide area is assigned to each pixel point, a multi-index evaluation system is established for quantitative evaluation of landslide degree and risk level, and multiple physical parameters and feature parameters are selected as evaluation indexes for more comprehensive risk evaluation of the landslide area.

Description

Technical Field

[0001] This invention relates to the field of dam slope detection technology, and in particular to a system and method for identifying dam landslide hazard areas based on UAVs and remote sensing images. Background Technology

[0002] Due to factors such as changes in geological structure, extreme weather, and water erosion, landslides frequently occur at dikes. Traditional manual inspection methods are not only inefficient and costly, but also suffer from limited monitoring range and strong subjectivity, making it difficult to meet the real-time and accuracy requirements of modern water conservancy projects for dike safety monitoring. Summary of the Invention

[0003] The purpose of this invention is to provide a system and method for identifying landslide hazard areas along embankments based on unmanned aerial vehicles (UAVs) and remote sensing images, in order to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a dam landslide hazard identification system based on UAV and remote sensing images, comprising:

[0005] The mission planning module is configured to build a multi-dimensional environmental perception model and generate the drone's flight path.

[0006] The data acquisition module is configured to acquire remote sensing images and multimodal data of the dam area through sensors integrated on the UAV, register the remote sensing images and multimodal data and generate a fused dataset;

[0007] The feature extraction module is configured to use a convolutional neural network based on an attention mechanism to perform shallow feature extraction on the fused dataset, and to construct a topological graph based on the shallow features to generate deep semantic features.

[0008] The region division module is configured to identify landslide features from deep semantic features, capture the changing trend of deep semantic features over time and generate landslide features, determine the regional boundaries of remote sensing images based on the probability distribution of landslide features and generate landslide regions, assess the degree of landslide and divide the landslide regions into sub-regions with different risk levels.

[0009] The hazard reporting module is configured to generate landslide hazard reports based on the landslide area, landslide severity, and risk level of sub-areas. It also classifies different warning levels according to the severity of the landslide hazard and sends warning information through the corresponding reporting channels.

[0010] Furthermore, the data acquisition module also performs the following steps:

[0011] Aerial images captured by the UAV are retrieved in real time; wherein, all aerial images captured by the UAV are used to generate remote sensing images;

[0012] Extract remote sensing images acquired from two consecutive aerial image acquisitions as a set of aerial image data;

[0013] Collect the overlapping area of ​​aerial images corresponding to each group of aerial image data;

[0014] Retrieve the overlapping areas of all aerial image data and obtain the standard deviation of the area change corresponding to the overlapping area of ​​all groups of aerial image data.

[0015] Extract the drone's flight speed corresponding to each aerial image acquisition;

[0016] For each set of aerial image data, obtain the corresponding drone flight speed;

[0017] Based on the drone flight speed corresponding to each set of aerial image data, obtain the difference in drone flight speed corresponding to each set of aerial image data.

[0018] The standard deviation of the drone flight speed difference corresponding to each set of aerial image data is obtained based on the drone flight speed difference corresponding to each set of aerial image data.

[0019] The standard deviation of the area change is normalized to obtain the first standard deviation parameter after normalization.

[0020] The standard deviation of the flight speed difference of the UAV is normalized to obtain the second standard deviation parameter after normalization.

[0021] Adjust the drone's flight speed based on the first standard deviation parameter and the second standard deviation parameter.

[0022] Furthermore, adjusting the UAV flight speed based on the first standard deviation parameter and the second standard deviation parameter includes:

[0023] Retrieve the weighting coefficients corresponding to the first and second standard deviation parameters from the database;

[0024] Retrieve global sensitivity coefficients;

[0025] The global sensitivity coefficient is used to limit the speed adjustment range and prevent over-adjustment; and the value range of the global sensitivity coefficient is 0.18-0.32.

[0026] The minimum overlapping area of ​​the required aerial image is obtained by retrieving remote sensing images;

[0027] The area ratio parameter is obtained by comparing the standard deviation of the area change with the minimum value of the overlapping area of ​​the aerial image.

[0028] The flight speed of the UAV is adjusted by using the area ratio parameter in combination with the first standard deviation parameter and the second standard deviation parameter.

[0029] Furthermore, the feature extraction module includes:

[0030] The shallow feature enhancement unit is configured to construct a convolutional neural network based on an attention mechanism, and to construct a channel attention submodule and a spatial attention submodule;

[0031] The channel attention submodule obtains global information for different channels by performing global average pooling and global max pooling on the feature map in the channel dimension, and generates channel attention weights after processing by a multilayer perceptron; the spatial attention submodule obtains spatial attention weights by performing max pooling and average pooling operations on the feature map in the spatial dimension.

[0032] The channel attention weights and spatial attention weights are used to enhance channel features and spatial features related to dam structure, texture, and temperature anomalies;

[0033] The input fusion dataset is fed into a convolutional neural network to extract basic visual features, spatial geometric features, and temperature distribution features of the dam area in the remote sensing image, and the output is shallow features;

[0034] The topology association mining unit is configured to encode shallow features, map each feature vector to a node in a graph structure, establish edge connections between nodes and generate a feature association topology graph based on the spatial positional relationship and semantic similarity between features.

[0035] The feature association topology graph is input into the graph convolutional network for feature propagation and aggregation. Each node updates its feature representation by aggregating the feature information of its neighboring nodes and combining it with its own features.

[0036] Through multi-layer graph convolutional network iteration, the local feature information of nodes is expanded into deep semantic features containing global semantic information.

[0037] Furthermore, the region division module includes:

[0038] The landslide feature recognition unit is configured to construct a multi-scale feature analysis model and input deep semantic features, perform multi-scale convolution operations on the deep semantic features, and obtain feature representations at different scales.

[0039] For the deep semantic features of multiple frames in the same monitoring area, a recurrent neural network structure is used to analyze the trend of feature changes over time and capture the dynamic change information of the embankment area, identify the feature patterns related to landslides, and generate landslide feature vectors containing spatiotemporal information.

[0040] The region boundary determination unit is configured to enhance the landslide feature vector to generate more feature data simulating landslide scenarios. The enhanced landslide feature vector and the feature vector of non-landslide areas are input into the probability classification model. The probability value of each pixel belonging to the landslide area is obtained through model training, and a probability distribution map is generated.

[0041] The probability distribution map is processed by calculating the statistical characteristics of the probability distribution and dynamically determining the optimal segmentation threshold to divide the probability distribution map into landslide areas and non-landslide areas.

[0042] For the segmented region, isolated noise points are removed and the region boundaries are smoothed to determine the landslide area boundary in the remote sensing image and output the landslide area range.

[0043] Furthermore, the region division module also includes:

[0044] The landslide severity assessment unit is configured to select physical parameters and characteristic parameters of the landslide area as assessment indicators. The physical parameters include area, volume, slope change, and soil displacement velocity. The characteristic parameters include the intensity and rate of change of landslide characteristics.

[0045] The specific values ​​of each evaluation indicator are obtained by using sensor data and image processing results;

[0046] Based on the importance of each assessment indicator to the degree of landslide, the weight of each assessment indicator is calculated. The values ​​of each assessment indicator are combined with their corresponding weights to calculate the quantitative assessment value of the degree of landslide. The degree level of the current landslide is determined according to the preset degree level standard.

[0047] The risk level classification unit is configured to build a risk assessment model, taking the landslide severity level, combined with the geographical location of the landslide area, surrounding geological conditions, and historical disaster data as input parameters for the risk assessment model;

[0048] The input parameters are comprehensively analyzed, and the landslide area is divided into sub-regions with different risk levels according to the risk level corresponding to different parameter combinations. These sub-regions are divided into low-risk areas, medium-risk areas, and high-risk areas.

[0049] Furthermore, the hazard reporting module includes:

[0050] The hazard report generation unit is configured to build a hazard report template library. The templates in the hazard report template library include a text description module, a data statistics module, a chart display module, and an image data module.

[0051] Obtain the landslide area, landslide degree, and risk level information of sub-areas output by the area division module, select the appropriate report template structure from the template library, and fill in the report template;

[0052] Visualize the spatial distribution information of landslide areas on a two-dimensional map or three-dimensional terrain model to generate a landslide area distribution map. Based on multi-period monitoring data, analyze the changing trends of landslide characteristics and generate a landslide development trend curve. Fill the landslide area distribution map and landslide development trend curve into the report template to generate a complete landslide hazard report.

[0053] The graded early warning release unit is configured to divide different early warning levels according to the severity of the landslide hazard. The early warning levels include blue, yellow, orange and red warnings, and each warning level corresponds to a different range of hazard severity and corresponding response measures.

[0054] Warning information is sent through the channels corresponding to the warning level. The warning information includes the specific location of the landslide hazard, the risk level, and suggestions for countermeasures.

[0055] Furthermore, when constructing a multi-dimensional environmental perception model, the task planning module integrates the geographical location of the dam, historical landslide data, and meteorological conditions. When planning the UAV's flight path, it dynamically plans the UAV's flight altitude, speed, and shooting angle, enabling the UAV to collect high-resolution remote sensing images and terrain data containing the entire dam in the best posture.

[0056] Furthermore, the sensors integrated in the data acquisition module include an optical camera, a lidar, and a thermal imager.

[0057] Furthermore, the multimodal data collected by the sensor includes optical images, three-dimensional terrain point cloud data, and thermal radiation data, and the fusion dataset generated by the data acquisition module contains visual, spatial, and temperature information.

[0058] Furthermore, the method for identifying landslide hazard areas of dams based on UAVs and remote sensing images, applied to the aforementioned system for identifying landslide hazard areas of dams based on UAVs and remote sensing images, includes the following steps:

[0059] Construct a multi-dimensional environmental perception model, plan the flight path of the UAV, collect remote sensing images and multimodal data, and adjust the flight speed of the UAV based on the standard deviation of the area change of the overlapping area of ​​the aerial images and the standard deviation of the difference in the flight speed of the UAV.

[0060] The basic visual, spatial geometric, and temperature distribution shallow features of the dam area in remote sensing images are extracted; the shallow features are encoded into a graph structure, a feature association topology graph is established, and deep semantic features containing global semantic information are generated through graph convolutional network iteration.

[0061] A multi-scale feature analysis model is constructed to identify landslide feature patterns, generate landslide feature vectors and input them into a probability classification model to determine the probability value of a pixel belonging to a landslide area. The segmentation threshold is dynamically determined by calculating the statistical features of the probability distribution, and the boundary of the landslide area is determined.

[0062] Physical parameters such as the area and volume of the landslide area, as well as characteristic intensity parameters, are selected as evaluation indicators. The values ​​of the indicators are obtained and the weights are calculated to obtain the quantitative evaluation value and grade of the landslide degree.

[0063] Based on the geographical location of the landslide area, surrounding geological conditions, and historical disaster data, a risk assessment model was constructed to divide the landslide area into low, medium, and high-risk sub-regions.

[0064] Based on the regional division results, a landslide hazard report is generated, and early warning information including the location of the hazard, risk level, and response suggestions is released through appropriate channels according to the severity of the landslide hazard.

[0065] Compared with the prior art, the beneficial effects of the present invention are:

[0066] 1. The feature extraction module of this invention uses global pooling and multilayer perceptron to assign different weights to different channels, highlighting important features and focusing on features of key parts of the dam. It suppresses the interference of irrelevant information, improves the targeting and accuracy of feature extraction, constructs shallow features into a graph structure, and uses graph convolutional networks to mine the spatial relationships between features. Through feature propagation and aggregation between nodes, local feature information is expanded into deep semantic features containing global semantic information, capturing the complex interactions between features. This enables the model to adaptively learn different types of feature associations. Whether facing complex dam structures or changing environmental conditions, it can effectively extract landslide-related feature information, improving the application effect of the system in different scenarios.

[0067] 2. The region segmentation module of this invention, by constructing a multi-scale feature analysis model and a recurrent neural network structure, simultaneously captures the spatial multi-scale information and temporal change trends of deep semantic features. This enables a more comprehensive description of the dynamic change process of the embankment, accurately identifies feature patterns related to landslides, accurately determines the boundaries of landslide areas, assigns a probability value to each pixel belonging to the landslide area, and dynamically determines the optimal segmentation threshold. This avoids the limitations of traditional fixed threshold methods. By establishing a multi-index evaluation system and a comprehensive risk assessment model, it achieves a quantitative assessment of the degree and risk level of landslides. Multiple physical parameters and feature parameters are selected as evaluation indicators, and the severity of landslides is comprehensively assessed through weighted calculations, resulting in a more comprehensive risk assessment of landslide areas.

[0068] 3. The hazard release module of this invention standardizes and regulates the landslide hazard report generation process by constructing a hazard report template library, ensuring the integrity and standardization of the report content. This ensures that reports generated by different people at different times have a unified format and standard, facilitating information comparison and analysis. The spatial distribution information and changing trends of landslide areas are visualized in the form of charts and curves, making complex hazard information more intuitive and easy to understand. The landslide area distribution map and landslide development trend curve can clearly show the location, scope and development trend of the landslide, helping decision-makers to quickly grasp key information about the hazard and make timely and effective decisions. Attached Figure Description

[0069] Figure 1 This is a schematic diagram of the dam landslide hazard area identification system module of the present invention.

[0070] Figure 2 This is a schematic diagram of the process for identifying landslide hazard areas along embankments according to the present invention. Detailed Implementation

[0071] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0072] Please see Figure 1 The present invention provides the following technical solutions:

[0073] A system for identifying landslide hazard areas along embankments based on UAVs and remote sensing images includes:

[0074] The task planning module is configured to build a multi-dimensional environmental perception model and generate a drone flight path. When building the multi-dimensional environmental perception model, it integrates the geographical location of the dam, historical landslide data and meteorological conditions. When planning the drone flight path, it dynamically plans the drone's flight altitude, speed and shooting angle, so that the drone can collect high-resolution remote sensing images and terrain data containing the entire dam in the best posture.

[0075] The data acquisition module is configured to acquire remote sensing images and multimodal data of the dam area through sensors integrated on the UAV. The sensors include an optical camera, a lidar, and a thermal imager. The multimodal data acquired by the sensors includes optical images, three-dimensional terrain point cloud data, and thermal radiation data. The fusion dataset generated by the data acquisition module contains visual, spatial, and temperature information. The remote sensing images and multimodal data are registered and a fusion dataset is generated.

[0076] The feature extraction module is configured to use a convolutional neural network based on an attention mechanism to perform shallow feature extraction on the fused dataset, and to construct a topological graph based on the shallow features to generate deep semantic features.

[0077] The region division module is configured to identify landslide features from deep semantic features, capture the changing trend of deep semantic features over time and generate landslide features, determine the regional boundaries of remote sensing images based on the probability distribution of landslide features and generate landslide regions, assess the degree of landslide and divide the landslide regions into sub-regions with different risk levels.

[0078] The hazard reporting module is configured to generate landslide hazard reports based on the landslide area, landslide severity, and risk level of sub-areas. It also classifies different warning levels according to the severity of the landslide hazard and sends warning information through the corresponding reporting channels.

[0079] In the above embodiments, the mission planning module plans the optimal flight path for the UAV by integrating the geographical location of the dam, historical landslide data and meteorological conditions, ensuring that the UAV can collect high-resolution remote sensing images and terrain data containing the entire dam in the best attitude under different environmental conditions. The data acquisition module realizes the synchronous acquisition and fusion of multimodal data by integrating multiple sensors. By accurately registering the remote sensing images and multimodal data collected by different sensors, a fused dataset is generated, which solves the problem of spatial and temporal inconsistency between multi-source data.

[0080] Specifically, the data acquisition module also performs the following steps:

[0081] Aerial images captured by the UAV are retrieved in real time; wherein, all aerial images captured by the UAV are used to generate remote sensing images;

[0082] Extract remote sensing images acquired from two consecutive aerial image acquisitions as a set of aerial image data;

[0083] Collect the overlapping area of ​​aerial images corresponding to each group of aerial image data;

[0084] Retrieve the overlapping areas of all aerial image data and obtain the standard deviation of the area change corresponding to the overlapping area of ​​all groups of aerial image data.

[0085] Extract the drone's flight speed corresponding to each aerial image acquisition;

[0086] For each set of aerial image data, obtain the corresponding drone flight speed;

[0087] Based on the drone flight speed corresponding to each set of aerial image data, obtain the difference in drone flight speed corresponding to each set of aerial image data.

[0088] The standard deviation of the drone flight speed difference corresponding to each set of aerial image data is obtained based on the drone flight speed difference corresponding to each set of aerial image data.

[0089] The standard deviation of the area change is normalized to obtain the first standard deviation parameter after normalization.

[0090] The standard deviation of the flight speed difference of the UAV is normalized to obtain the second standard deviation parameter after normalization.

[0091] Adjust the drone's flight speed based on the first standard deviation parameter and the second standard deviation parameter.

[0092] The technical effect of the above solution is as follows: In UAV aerial photography operations, the working principle of this solution is based on the close correlation between the UAV's flight speed and the area of ​​the overlapping region of the aerial images. When the UAV flies, its speed directly affects the degree of overlap between adjacent aerial images. A higher speed results in a longer flight distance per unit time, increasing the interval between adjacent image captures and reducing the overlapping area; conversely, a slower speed has the opposite effect. Simultaneously, the stability of the flight speed is also crucial; drastic speed fluctuations lead to uneven intervals between adjacent image captures, further affecting the regularity of the overlapping area. This solution acquires remote sensing image data generated from adjacent aerial images in real time, extracts the overlapping region, and calculates the standard deviation of its area change to measure the rationality of the speed in the spatial dimension. Simultaneously, it obtains the difference in UAV flight speed and its standard deviation to reflect the stability of the speed in the temporal dimension. After normalizing these two standard deviation parameters, they are used as feedback. Through the coordinated adjustment of these two parameters, the UAV's flight speed is corrected, the speed mean is corrected, and speed jitter is suppressed, achieving "position-speed" dual-domain coordinated control.

[0093] Compared to existing technologies, this solution demonstrates significant technical advantages in performance indicators. Regarding overlap rate stability, existing technologies rely on a preset fixed speed, making them susceptible to external factors and resulting in large overlap rate fluctuations. This solution, however, dynamically adjusts the overlap rate to minimize fluctuations, significantly improving stability. In terms of image stitching accuracy, existing technologies suffer from poor stitching quality due to large image parallax. The solution described in this embodiment significantly reduces errors and improves image stitching quality. Regarding data acquisition efficiency, existing technologies require repeated manual speed adjustments, leading to long operation times. The solution described in this embodiment features fully automatic real-time adjustment, greatly shortening operation time. Regarding speed fluctuation range, the solution employs dual standard deviation feedback control, effectively suppressing speed fluctuations. Furthermore, this solution exhibits strong anti-interference capabilities, enabling stable operation in complex environments. It greatly expands the UAV's operational window, achieving dual optimization of UAV flight stability and data acquisition quality, and possesses higher engineering practicality and environmental adaptability.

[0094] Specifically, adjusting the drone's flight speed based on the first standard deviation parameter and the second standard deviation parameter includes:

[0095] Retrieve the weighting coefficients corresponding to the first and second standard deviation parameters from the database;

[0096] Retrieve global sensitivity coefficients;

[0097] The global sensitivity coefficient is used to limit the speed adjustment range and prevent over-adjustment, and the value range of the global sensitivity coefficient is 0.18-0.32.

[0098] The minimum overlapping area of ​​the required aerial image is obtained by retrieving remote sensing images;

[0099] The area ratio parameter is obtained by comparing the standard deviation of the area change with the minimum value of the overlapping area of ​​the aerial image.

[0100] The flight speed of the UAV is adjusted by using the area ratio parameter in combination with the first standard deviation parameter and the second standard deviation parameter.

[0101] The adjusted flight speed is obtained using the following formula:

[0102]

[0103] Where V represents the adjusted flight speed; V0 represents the original flight speed; and k represents the global sensitivity coefficient. Specifically, the global sensitivity coefficient can be derived from control theory, drawing on the damping ratio of a second-order system in automatic control. This can be achieved by establishing a simplified model for UAV speed adjustment, combining parameters such as system time constant and damping ratio, and then verifying it through ground testing. Alternatively, an engineering experience mapping method can be used to construct a historical operating condition database, select cases without overshoot and with excellent image quality, statistically analyze the distribution range of k, and then appropriately adjust it based on scene differences. Furthermore, an intelligent dynamic adaptation self-learning optimization method can be employed, defining a reward function that includes avoiding overshoot and improving image quality. Through online reinforcement learning, the system can dynamically optimize k based on the environment and system state. All of the above methods for obtaining the global sensitivity coefficient are known in existing technologies, and this embodiment will not elaborate on the specific implementation methods. In actual deployment, if simplicity and operability are desired, the simplified experience mapping method is recommended. First, the average value of k from qualified tasks is taken as the initial value through ground testing. During flight, the step size of k is dynamically increased or decreased based on overshoot and image quality conditions to ensure that k remains within a reasonable range and adapts to the needs of UAV aerial photography scenarios. α and β represent the weighting coefficients corresponding to the first and second standard deviation parameters, respectively. The values ​​of α and β range from 0.1 to 0.8. The system will automatically adjust these values ​​based on the priority of the two standard deviation parameters in the actual UAV image acquisition scenario. However, the adjustment process will always ensure the rationality of the weight allocation logic and formula calculations.01 and S 02 Let α and β represent the first and second standard deviation parameters, respectively; B represents the area ratio parameter; and w represents the adaptive adjustment reference coefficient, used to prevent the denominator from being zero or negative. The value of the adaptive adjustment reference coefficient is set according to the actual situation of UAV image acquisition, and its upper limit is slightly greater than the possible maximum value of the area ratio parameter B, with a greater rate not exceeding 30%. For example, when the minimum overlap area ratio required for remote sensing image generation by the UAV is 15%, and the maximum identifiable area change corresponding to the area change acquisition resolution of the UAV aerial photography equipment is 50% of the aerial photography area, and the standard deviation of the area change of the most likely adjacent aerial photography images is 1, the maximum possible value of the area ratio parameter B is 6.67. Then, the system automatically adjusts the adaptive adjustment reference coefficient w to be greater than the maximum possible value of the area ratio parameter B, for example, the upper limit of the adaptive adjustment reference coefficient w is automatically adjusted to 7.00. Meanwhile, α and β, as weighting coefficients, are used to adjust the "overlap area fluctuation (S... 01 "Correlated image quality" and "velocity fluctuation (S)" 02 The quantification of the impact of "correlated flight stability" can be achieved through existing offline calibration combined with online self-learning in real-world scenarios.

[0104] Offline calibration: In a laboratory environment (simulating stable airflow and fixed terrain), the effects of different combinations of α and β on imaging quality (such as stitching error and texture continuity) were tested by controlling the single variables of overlapping area fluctuation and velocity fluctuation. A "weight-imaging quality" mapping library was established as the initial value of the database.

[0105] Online self-learning: Real-time monitoring of imaging quality during flight (such as registration error and blurriness of the stitched remote sensing image). When the imaging quality deteriorates due to overlapping areas, the α weight is automatically increased; when it deteriorates due to speed fluctuations (such as image ghosting), the β weight is automatically increased, achieving dynamic adaptation. The above weight setting methods are all existing technologies in this field, and this embodiment will not elaborate on the principle and specific method.

[0106] Specifically, The first standard deviation parameter S 01 This is obtained after normalizing the standard deviation of area variation. The standard deviation of area variation reflects the fluctuation of the area of ​​overlapping regions in adjacent aerial images. The normalized S... 01 This facilitates a standardized quantification of the impact of such fluctuations on speed adjustment. The weighting coefficient α is used to measure S. 01 The importance of speed adjustment. The area ratio parameter B is obtained by dividing the standard deviation of area change by the minimum overlapping area of ​​the aerial image. It reflects the proportional relationship between the current overlapping area change and the minimum required overlapping area.

[0107] As w−B is used as the denominator, the denominator approaches 0 when B is close to w. The value of this item will increase, and the change in the area of ​​the overlapping region will have a greater impact on the speed adjustment. At this point, we are in a situation close to the limit of shooting requirements, and we need to pay more attention to the change in area to adjust the speed. The second standard deviation parameter S 02 This is the result after normalizing the standard deviation of the drone's flight speed difference. The standard deviation of the flight speed difference reflects the degree of fluctuation in the drone's flight speed during adjacent image acquisitions. Large fluctuations indicate unstable flight speed, which will affect the aerial photography quality and requires speed adjustment. The weighting coefficient β determines S. 02 Its importance in speed adjustment. Adjust S according to the relative magnitudes of β and B 02 Contribution to speed adjustment. When the area ratio parameter B exceeds the weight corresponding to β, the adjustment is primarily driven by area demand; if B is small, the weight β plays a major role, balancing the impact of area and speed fluctuations on flight control, reflecting the trade-off between the impact of speed difference fluctuations based on actual conditions. The formula integrates the changes in the overlapping area (via S...) 01 (reflected) and fluctuations in flight speed difference (via S) 02 Factors such as flight status and environmental changes are comprehensively reflected to determine the speed adjustment requirements of the drone. This avoids the one-sidedness of adjusting speed based on a single factor, making speed adjustments more closely match actual flight conditions and improving flight stability and safety.

[0108] The speed adjustment range is controlled by k, allowing for the setting of appropriate values ​​based on different flight scenarios and requirements. In complex environments (such as complex terrain and variable wind speeds), increasing the k value allows the drone to respond more sensitively to changes; in stable environments, decreasing the k value results in smoother speed adjustments and ensures reasonable speed regulation. The variation in the overlapping area (S) is also considered. 01 (Related) This ensures that the drone's speed is adapted to the shooting requirements in different shooting scenarios. By reasonably adjusting the speed, an appropriate overlapping area in the image is ensured, facilitating subsequent image stitching and remote sensing image generation. At the same time, considering fluctuations in flight speed differences, sudden speed changes can be avoided, leading to blurry or shaky images and improving the quality of aerial photographs. Furthermore, based on the different requirements for overlapping area and speed stability of different shooting tasks (such as panoramic shooting and close-up shooting), the formula can adjust the speed through various parameters. For example, panoramic shooting requires a large overlapping area and stable speed; by adjusting the weighting coefficients α and β, the drone speed can be made to meet the shooting task requirements, enhancing the compatibility with the aerial photography task.

[0109] The technical effects of the above-mentioned solution are as follows: Existing technologies often struggle to balance the area of ​​overlapping regions and speed stability, leading to large fluctuations in the overlapping region. This solution, through multi-parameter collaborative speed adjustment, can reduce the fluctuation range of the overlapping region area by approximately 30%-50%, improving the stability of the overlapping region and ensuring the quality of remote sensing image generation. Compared to traditional speed adjustment methods, the above-mentioned solution in this embodiment considers the standard deviation of the speed difference and assigns weights for adjustment, reducing the standard deviation of the UAV's flight speed fluctuation by approximately 40%-60%, resulting in a more stable speed and a significant advantage for tasks requiring stable shooting. Because it comprehensively considers multiple factors and has a global sensitivity coefficient limiting the adjustment range, the UAV can adjust its speed more quickly and accurately when facing external interference such as wind speed changes. Compared to existing technologies, it can maintain the stability of the overlapping region area and speed even with wind speed changes of ±2m / s, improving anti-interference capability by approximately 30%-40%. Existing speed adjustment technologies may suffer from over-adjustment or under-adjustment. This solution, through precise formula calculations, adjusts the speed reasonably based on various parameters, improving speed adjustment accuracy by approximately 20%-30% and reducing repetitive work or data errors caused by inaccurate speed adjustment.

[0110] The feature extraction module includes:

[0111] The shallow feature enhancement unit is configured to construct a convolutional neural network based on an attention mechanism, and to construct a channel attention submodule and a spatial attention submodule;

[0112] The channel attention submodule obtains global information for different channels by performing global average pooling and global max pooling on the feature map in the channel dimension, and generates channel attention weights after processing by a multilayer perceptron; the spatial attention submodule obtains spatial attention weights by performing max pooling and average pooling operations on the feature map in the spatial dimension.

[0113] The channel attention weights and spatial attention weights are used to enhance channel features and spatial features related to dam structure, texture, and temperature anomalies;

[0114] The input fusion dataset is fed into a convolutional neural network to extract basic visual features, spatial geometric features, and temperature distribution features of the dam area in the remote sensing image, and the output is shallow features;

[0115] The topology association mining unit is configured to encode shallow features, map each feature vector to a node in a graph structure, establish edge connections between nodes and generate a feature association topology graph based on the spatial positional relationship and semantic similarity between features.

[0116] The feature association topology graph is input into the graph convolutional network for feature propagation and aggregation. Each node updates its feature representation by aggregating the feature information of its neighboring nodes and combining it with its own features.

[0117] Through multi-layer graph convolutional network iteration, the local feature information of nodes is expanded into deep semantic features containing global semantic information.

[0118] In the above embodiments, the feature extraction module, by constructing a channel attention submodule and a spatial attention submodule, can automatically identify and enhance channel features and spatial features related to dam structure, texture, and temperature anomalies. The channel attention submodule assigns different weights to different channels through global pooling and a multilayer perceptron, highlighting important features; the spatial attention submodule, through spatial dimension pooling operations, focuses on features of key parts of the dam, suppressing interference from irrelevant information and improving the targeting and accuracy of feature extraction.

[0119] In the above embodiments, shallow features are constructed into a graph structure, and a graph convolutional network is used to mine the spatial relationships between features. Through feature propagation and aggregation between nodes, local feature information is expanded into deep semantic features containing global semantic information, capturing the complex interactions between features. For example, the correlation between different parts of a dam can be discovered, and potential landslide hazard areas can be identified, providing a more comprehensive and in-depth feature representation for subsequent landslide identification. This enables the model to adaptively learn different types of feature associations, exhibiting strong adaptability and generalization ability. Whether facing complex dam structures or changing environmental conditions, it can effectively extract landslide-related feature information, improving the system's application performance in different scenarios.

[0120] The region division module includes:

[0121] The landslide feature recognition unit is configured to construct a multi-scale feature analysis model and input deep semantic features, perform multi-scale convolution operations on the deep semantic features, and obtain feature representations at different scales.

[0122] For the deep semantic features of multiple frames in the same monitoring area, a recurrent neural network structure is used to analyze the trend of feature changes over time and capture the dynamic change information of the embankment area, identify the feature patterns related to landslides, and generate landslide feature vectors containing spatiotemporal information.

[0123] The region boundary determination unit is configured to enhance the landslide feature vector to generate more feature data simulating landslide scenarios. The enhanced landslide feature vector and the feature vector of non-landslide areas are input into the probability classification model. The probability value of each pixel belonging to the landslide area is obtained through model training, and a probability distribution map is generated.

[0124] The probability distribution map is processed by calculating the statistical characteristics of the probability distribution and dynamically determining the optimal segmentation threshold to divide the probability distribution map into landslide areas and non-landslide areas.

[0125] For the segmented region, isolated noise points are removed and the region boundaries are smoothed to determine the landslide area boundary in the remote sensing image and output the landslide area range.

[0126] The landslide severity assessment unit is configured to select physical parameters and characteristic parameters of the landslide area as assessment indicators. The physical parameters include area, volume, slope change, and soil displacement velocity. The characteristic parameters include the intensity and rate of change of landslide characteristics.

[0127] The specific values ​​of each evaluation indicator are obtained by using sensor data and image processing results;

[0128] Based on the importance of each assessment indicator to the degree of landslide, the weight of each assessment indicator is calculated. The values ​​of each assessment indicator are combined with their corresponding weights to calculate the quantitative assessment value of the degree of landslide. The degree level of the current landslide is determined according to the preset degree level standard.

[0129] The risk level classification unit is configured to build a risk assessment model, taking the landslide severity level, combined with the geographical location of the landslide area, surrounding geological conditions, and historical disaster data as input parameters for the risk assessment model;

[0130] The input parameters are comprehensively analyzed, and the landslide area is divided into sub-regions with different risk levels according to the risk level corresponding to different parameter combinations. These sub-regions are divided into low-risk areas, medium-risk areas, and high-risk areas.

[0131] In the above embodiments, the region segmentation module, by constructing a multi-scale feature analysis model and a recurrent neural network structure, simultaneously captures the spatial multi-scale information and temporal variation trends of deep semantic features. This enables a more comprehensive description of the dynamic change process of the embankment and accurate identification of landslide-related feature patterns. For example, by analyzing the changing trends of multi-period monitoring data, early signs of landslides can be detected in advance, providing more time for early warning.

[0132] In the above embodiments, a probabilistic classification model and an adaptive threshold segmentation algorithm, combined with morphological processing methods, can accurately determine the boundary of the landslide area and assign a probability value belonging to the landslide area to each pixel. The adaptive threshold segmentation algorithm dynamically determines the optimal segmentation threshold based on the statistical characteristics of the probability distribution, avoiding the limitations of traditional fixed threshold methods. By establishing a multi-index evaluation system and a comprehensive risk assessment model, a quantitative assessment of the degree and risk level of landslides is achieved. Multiple physical parameters and feature parameters are selected as evaluation indicators, and the severity of the landslide is comprehensively assessed through weighted calculation. Combining factors such as geographical location, surrounding geological conditions, and historical disaster data, a more comprehensive risk assessment of the landslide area is conducted.

[0133] The hazard reporting module includes:

[0134] The hazard report generation unit is configured to build a hazard report template library. The templates in the hazard report template library include a text description module, a data statistics module, a chart display module, and an image data module.

[0135] Obtain the landslide area, landslide degree, and risk level information of sub-areas output by the area division module, select the appropriate report template structure from the template library, and fill in the report template;

[0136] Visualize the spatial distribution information of landslide areas on a two-dimensional map or three-dimensional terrain model to generate a landslide area distribution map. Based on multi-period monitoring data, analyze the changing trends of landslide characteristics and generate a landslide development trend curve. Fill the landslide area distribution map and landslide development trend curve into the report template to generate a complete landslide hazard report.

[0137] The graded early warning release unit is configured to divide different early warning levels according to the severity of the landslide hazard. The early warning levels include blue, yellow, orange and red warnings, and each warning level corresponds to a different range of hazard severity and corresponding response measures.

[0138] Warning information is sent through the channels corresponding to the warning level. The warning information includes the specific location of the landslide hazard, the risk level, and suggestions for countermeasures.

[0139] In the above embodiments, the hazard release module standardizes and normalizes the landslide hazard report generation process by building a hazard report template library, ensuring the integrity and standardization of the report content. This makes reports generated by different people at different times have a unified format and standard, which facilitates information comparison and analysis and improves the scientificity and accuracy of decision-making.

[0140] In the above embodiments, the spatial distribution information and changing trends of landslide areas are visualized in the form of charts and curves, making complex hazard information more intuitive and easier to understand. The landslide area distribution map and landslide development trend curve clearly show the location, extent, and development status of the landslide, helping decision-makers quickly grasp key information about the hazard and make timely and effective decisions. Different warning levels are defined according to the severity of the landslide hazard, and corresponding combinations of dissemination channels are preset for each warning level, ensuring that warning information is accurately conveyed to relevant personnel based on the severity of the hazard. For example, a red warning can simultaneously trigger multiple dissemination channels, ensuring that the affected population and emergency personnel receive hazard information as soon as possible and take appropriate countermeasures to minimize disaster losses.

[0141] Please see Figure 2 The method for identifying landslide hazard areas of embankments based on UAVs and remote sensing images, applied to the aforementioned system for identifying landslide hazard areas of embankments based on UAVs and remote sensing images, includes the following steps:

[0142] Construct a multi-dimensional environmental perception model, plan the flight path of the UAV, collect remote sensing images and multimodal data, and adjust the flight speed of the UAV based on the standard deviation of the area change of the overlapping area of ​​the aerial images and the standard deviation of the difference in the flight speed of the UAV.

[0143] The basic visual, spatial geometric, and temperature distribution shallow features of the dam area in remote sensing images are extracted; the shallow features are encoded into a graph structure, a feature association topology graph is established, and deep semantic features containing global semantic information are generated through graph convolutional network iteration.

[0144] A multi-scale feature analysis model is constructed to identify landslide feature patterns, generate landslide feature vectors and input them into a probability classification model to determine the probability value of a pixel belonging to a landslide area. The segmentation threshold is dynamically determined by calculating the statistical features of the probability distribution, and the boundary of the landslide area is determined.

[0145] Physical parameters such as the area and volume of the landslide area, as well as characteristic intensity parameters, are selected as evaluation indicators. The values ​​of the indicators are obtained and the weights are calculated to obtain the quantitative evaluation value and grade of the landslide degree.

[0146] Based on the geographical location of the landslide area, surrounding geological conditions, and historical disaster data, a risk assessment model was constructed to divide the landslide area into low, medium, and high-risk sub-regions.

[0147] Based on the regional division results, a landslide hazard report is generated, and early warning information including the location of the hazard, risk level, and response suggestions is released through appropriate channels according to the severity of the landslide hazard.

[0148] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A dam landslide danger area identification system based on a UAV and a remote sensing image, characterized in that, The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) and a landslide risk assessment system based on the method. The application relates to a landslide risk assessment method based on unmanned aerial vehicle (UAV) For the multi-frame deep semantic features of the same monitoring area, a recurrent neural network structure is used to analyze the change trend of the features over time and capture the dynamic change information of the dam area, identify the feature patterns related to landslides, and generate a landslide feature vector containing spatio-temporal information; The region boundary determination unit is configured to perform enhancement processing on the landslide feature vector to generate more feature data simulating landslide scenarios, input the enhanced landslide feature vector and the feature vector of the non-landslide area into the probability classification model, and obtain the probability value of each pixel point belonging to the landslide area through model training to generate a probability distribution map; The probability distribution map is processed, the statistical characteristics of the probability distribution are calculated, and the optimal segmentation threshold is dynamically determined to divide the probability distribution map into landslide and non-landslide areas; For the segmented areas, remove isolated noise points and smooth the area boundaries to determine the landslide area boundary in the remote sensing image and output the landslide area range. 2.The dam landslide risk area identification system based on a UAV and remote sensing images of claim 1, wherein, The data acquisition module further performs the following steps: Real-time retrieval of aerial images collected by the unmanned aerial vehicle; wherein all aerial images collected by the unmanned aerial vehicle are used to generate remote sensing images; Extracting remote sensing images collected by each adjacent two aerial images as a set of aerial image data; Collecting the aerial image overlap area corresponding to each set of aerial image data; Retrieving the aerial image overlap area corresponding to all aerial image data to obtain the area change standard deviation corresponding to the area of all sets of aerial image data; Extracting the flight speed of the unmanned aerial vehicle corresponding to each aerial image collection; For each set of aerial image data, obtain the corresponding flight speed of the unmanned aerial vehicle; According to the flight speed of the unmanned aerial vehicle corresponding to each set of aerial image data, obtain the flight speed difference value corresponding to each set of aerial image data; According to the flight speed difference value corresponding to each set of aerial image data, obtain the flight speed difference value standard deviation corresponding to each set of aerial image data; Normalizing the area change standard deviation to obtain the first standard deviation parameter after normalization; Normalizing the flight speed difference value standard deviation to obtain the second standard deviation parameter after normalization; Adjusting the flight speed of the unmanned aerial vehicle according to the first standard deviation parameter and the second standard deviation parameter. 3.The dam landslide risk area identification system based on UAV and remote sensing images of claim 2, wherein, Adjusting the flight speed of the unmanned aerial vehicle according to the first standard deviation parameter and the second standard deviation parameter, comprising: Retrieving the weight coefficient corresponding to the first standard deviation parameter and the second standard deviation parameter from the database; Retrieving the global sensitivity coefficient; Wherein, the global sensitivity coefficient is used to limit the speed adjustment amplitude to prevent over-adjustment; and the value range of the global sensitivity coefficient is 0.18-0.32; Retrieving the remote sensing image to generate the required minimum aerial image overlap area; Performing ratio processing on the area change standard deviation and the minimum aerial image overlap area to obtain an area ratio parameter; Adjusting the flight speed of the unmanned aerial vehicle using the area ratio parameter in combination with the first standard deviation parameter and the second standard deviation parameter. 4.The dam landslide risk area identification system based on a UAV and remote sensing images of claim 3, wherein, The region division module further comprises: The landslide degree evaluation unit is configured to select physical parameters and characteristic parameters of the landslide area as evaluation indexes, the physical parameters including area, volume, slope change amount and rock-soil displacement speed, and the characteristic parameters including intensity and characteristic change rate of landslide characteristics; The specific values of the evaluation indexes are obtained through sensor data and image processing results; According to the importance of the evaluation indexes on the landslide degree, the index weight of each evaluation index is calculated, the evaluation index values are combined with the corresponding index weights, and the quantitative evaluation value of the landslide degree is calculated, and the degree level of the current landslide is determined according to the preset degree level standard; The risk level division unit is configured to construct a risk evaluation model, and the landslide degree level, combined with the geographical position, surrounding geological conditions and historical disaster data factors of the landslide area, is used as the input parameter of the risk evaluation model; The landslide area is divided into sub-areas of different risk levels according to the risk degree corresponding to different parameter combinations through comprehensive analysis of the input parameters, and the sub-areas are divided into low-risk areas, medium-risk areas and high-risk areas. 5.The dam landslide risk area identification system based on UAV and remote sensing images of claim 1, wherein, The risk situation release module includes: The risk situation report generation unit is configured to construct a risk situation report template library, and the templates in the risk situation report template library include text description modules, data statistics modules, chart display modules and image data modules; The risk situation report generation unit is configured to construct a risk situation report template library, and the templates in the risk situation report template library include text description modules, data statistics modules, chart display modules and image data modules; The spatial distribution information of the landslide area is visualized on a two-dimensional map or a three-dimensional terrain model to generate a landslide area distribution map, the change trend of the landslide characteristics is analyzed based on multiple monitoring data to generate a landslide development trend curve, and the landslide area distribution map and the landslide development trend curve are filled into the report template to generate a complete landslide risk situation report; The graded early warning release unit is configured to divide different warning levels according to the severity of the landslide risk situation, the warning levels including blue warning, yellow warning, orange warning and red warning, each warning level corresponding to a different risk situation severity interval and corresponding countermeasure suggestion; The warning information is sent through the release channel corresponding to the warning level, and the warning information includes the specific position of the landslide risk situation, the risk level and the countermeasure suggestion. 6.The dam landslide risk area identification system based on a UAV and remote sensing images of claim 1, wherein, When constructing a multi-dimensional environment perception model, the task planning module integrates the dam geographical position, historical landslide data and meteorological condition information, dynamically plans the flight height, speed and shooting angle of the unmanned aerial vehicle when planning the flight path of the unmanned aerial vehicle, so that the unmanned aerial vehicle can collect high-resolution remote sensing images and terrain data containing the overall situation of the dam in the best posture. 7.The dam landslide risk area identification system based on UAV and remote sensing images of claim 1, wherein, The sensors integrated in the data acquisition module include optical cameras, laser radars and thermal imagers, and the multi-modal data collected by the sensors include optical images, three-dimensional terrain point cloud data and thermal radiation data, and the fusion data set generated by the data acquisition module contains visual, spatial and temperature information.

8. The dam landslide danger area identification method based on unmanned aerial vehicle and remote sensing image, applied to the dam landslide danger area identification system based on unmanned aerial vehicle and remote sensing image as claimed in claim 1, characterized in that, The method includes the following steps: A multi-dimensional environment perception model is constructed to plan the flight path of the UAV, collect remote sensing images and multi-modal data, and adjust the flight speed of the UAV according to the standard deviation of the area change of the overlapping area of the aerial images and the standard deviation of the flight speed difference of the UAV; The shallow features of the dam area in the remote sensing image are extracted, including basic vision, spatial geometry and temperature distribution; the shallow features are encoded into a graph structure to establish a feature correlation topology graph, and the deep semantic features containing global semantic information are generated through the iteration of the graph convolution network; A multi-scale feature analysis model is constructed to identify the landslide feature mode, generate a landslide feature vector and input it into a probability classification model to determine the probability value of a pixel point belonging to a landslide area, dynamically determine the segmentation threshold value by calculating the probability distribution statistical characteristics, and determine the boundary of the landslide area; The area, volume and feature strength of the landslide area are selected as evaluation indexes, the index values are obtained, the weights are calculated, and the quantitative evaluation value and grade of the landslide degree are obtained; A risk assessment model is constructed by combining the geographical location of the landslide area, the surrounding geological conditions and the historical disaster data, and the landslide area is divided into low, medium and high risk sub-areas; According to the results of the area division, a landslide danger report is generated, and the warning information containing the danger location, risk level and response suggestions is published through the corresponding channel according to the severity of the landslide danger.

Images