Geological disaster identification method and system combined with remote sensing technology

Abstract

The embodiment of the invention provides a geological disaster recognition method and system combined with a remote sensing technology, and the method comprises the steps: firstly obtaining a multi-time-sequence remote sensing image data set of a target region, which comprises remote sensing images collected at different times; then, performing topographic feature extraction processing on the data set to obtain a topographic feature set, covering earth surface coverage features, slope features and elevation change features, then, performing multi-time-sequence change analysis on the topographic feature set to generate a dynamic change feature set, and performing abnormal region identification on the dynamic change feature set by using a geological disaster identification model to obtain an abnormal region identification result; according to the method and the device, the abnormal region set with the geological disaster risk is determined, and finally, the disaster early warning signal is generated according to the abnormal region set and is sent to the disaster monitoring terminal, so that comprehensive and dynamic monitoring and accurate early warning of the geological disaster risk region are realized, and the accuracy and the efficiency of geological disaster monitoring and early warning are improved.

Description

Technical Field

[0001] The present application relates to the field of geological disaster monitoring technology, and more specifically, to a geological disaster identification method and system combined with remote sensing technology. Background Art

[0002] In the field of geological disaster monitoring and early warning, timely and accurate identification of potential geological disaster risk areas is crucial for protecting people's lives and property. Traditional geological disaster identification methods mainly rely on on-site exploration and geological surveys. Although these methods can detect geological disaster risks to a certain extent, they are often time-consuming and labor-intensive. Due to the limitations of exploration scope and personnel experience, it is difficult to achieve large-scale and efficient monitoring.

[0003] With the rapid development of remote sensing technology, the use of remote sensing images for geological hazard identification has become a new research hotspot. However, existing remote sensing geological hazard identification methods are mostly based on single-time series remote sensing images, which are difficult to capture the dynamic changes before geological hazards occur, resulting in limited recognition accuracy and timeliness. Summary of the Invention

[0004] In view of this, the purpose of this application is to provide a geological disaster identification method and system combined with remote sensing technology.

[0005] In conjunction with the first aspect of the present application, a method for identifying geological hazards using remote sensing technology is provided, which is applied to a geological hazard identification system using remote sensing technology. The method comprises: Acquire a multi-time series remote sensing image dataset of a target area, wherein the multi-time series remote sensing image dataset includes remote sensing images acquired at different times; Performing terrain feature extraction processing on the multi-time series remote sensing image dataset to obtain a terrain feature set of the target area, wherein the terrain feature set includes surface cover features, slope features, and elevation change features; Performing multi-time series change analysis on the terrain feature set to generate a dynamic change feature set of the target area; Calling a geological disaster identification model to perform abnormal area identification processing on the dynamic change feature set to determine a set of abnormal areas with geological disaster risks in the target area; A disaster warning signal is generated according to the abnormal area set, and the disaster warning signal is sent to a disaster monitoring terminal.

[0006] In combination with the second aspect of the present application, a geological disaster identification system combined with remote sensing technology is provided. The geological disaster identification system combined with remote sensing technology includes a machine-readable storage medium and a processor. The machine-readable storage medium stores machine-executable instructions. When the processor executes the machine-executable instructions, the geological disaster identification system combined with remote sensing technology implements the aforementioned geological disaster identification method combined with remote sensing technology.

[0007] In conjunction with the third aspect of the present application, a computer-readable storage medium is provided, in which computer-executable instructions are stored. When the computer-executable instructions are executed, the aforementioned geological disaster identification method combined with remote sensing technology is implemented.

[0008] In combination with any of the above aspects, by obtaining a multi-time series remote sensing image dataset of the target area and performing terrain feature extraction and multi-time series change analysis and processing on it, the dynamic change characteristics of the terrain in the target area are fully captured, and the geological disaster identification model is used to identify and process abnormal areas of these dynamic change characteristics. It is possible to accurately and efficiently determine the set of abnormal areas with geological disaster risks in the target area, which not only improves the accuracy and timeliness of geological disaster identification, but also overcomes the limitations of traditional methods that are limited by a single time series remote sensing image, and realizes comprehensive and dynamic monitoring of geological disaster risk areas. At the same time, disaster warning signals are generated based on the set of abnormal areas and sent to the disaster monitoring terminal in a timely manner to facilitate disaster emergency response, which helps to reduce casualties and property losses caused by geological disasters. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained by combining these drawings without paying any creative work.

[0010] Figure 1 A flow chart of a method for identifying geological hazards using remote sensing technology is provided in an embodiment of the present application. DETAILED DESCRIPTION

[0011] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0012] The terms "first," "second," and so on, in the specification and claims of the present invention and the accompanying drawings are used to distinguish between different items, not to describe a specific order. Furthermore, the terms "including," "having," and any variations thereof, are intended to cover non-exclusive inclusions. For example, a process, method, apparatus, product, or end comprising a series of steps or elements is not limited to the listed steps or elements but may optionally include steps or elements not listed therein, or may optionally include other steps or elements inherent to such process, method, product, or end.

[0013] References herein to "embodiments" mean that a particular feature, structure, or characteristic described in connection with the embodiments may be included in at least one embodiment of the present invention. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it constitute a separate or alternative embodiment that is mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0014] Figure 1 The flowchart of the geological hazard identification method combined with remote sensing technology provided by the embodiment of the present application is shown. It should be understood that in other embodiments, the order of some steps in the geological hazard identification method combined with remote sensing technology of this embodiment can be shared based on actual needs, or some steps can be omitted or maintained. The details of the geological hazard identification method combined with remote sensing technology include: Step S110: Acquire a multi-time series remote sensing image dataset of the target area, wherein the multi-time series remote sensing image dataset includes remote sensing images acquired at different times.

[0015] In this example, to identify geological hazards using remote sensing technology, the primary task is to obtain a multi-temporal remote sensing image dataset of the target area. This dataset, which forms the basis for subsequent analysis, contains remote sensing images collected at different times and reflects the changing surface information of the target area over time.

[0016] To acquire this dataset, various remote sensing platforms can be used. Satellite remote sensing is a common method. High-resolution optical satellites, for example, can scan and image the Earth's surface according to predetermined orbits and periods. After identifying the target area, image data of that area is collected at multiple points in time, based on the satellite's revisit period. Assuming the satellite's revisit period is duration A, images can be collected at multiple time points separated by multiples of duration A. These images have the same spatial coverage, forming a continuous time series.

[0017] In addition to satellite remote sensing, aerial remote sensing is also an effective data collection method. Aircraft equipped with remote sensing equipment can be used to collect targeted imagery of target areas as needed. The advantage of aerial remote sensing is the flexibility to adjust the acquisition time and scope based on actual conditions, enabling the acquisition of higher-resolution image data. When conducting aerial remote sensing acquisition, flight routes and altitudes must be carefully planned based on factors such as the size and topography of the target area to ensure high-quality image data and coverage of the entire target area.

[0018] After collecting remote sensing images from different time periods, they must undergo preliminary screening and organization. Image quality, including clarity and completeness, must be checked. Images with issues such as cloud cover and missing data must be processed or removed. Ultimately, remote sensing images from different time periods that meet quality requirements are combined to form a multi-temporal remote sensing image dataset.

[0019] Step S120: performing terrain feature extraction processing on the multi-time series remote sensing image dataset to obtain a terrain feature set of the target area, wherein the terrain feature set includes surface cover features, slope features, and elevation change features.

[0020] After obtaining the multi-time series remote sensing image dataset, the next step is to extract and process the terrain features to obtain the terrain feature set of the target area. The terrain feature set includes surface cover features, slope features, and elevation change features. These features are of great significance for the identification of geological disasters and can reflect the terrain conditions of the target area from different angles.

[0021] Step S121: performing radiation correction processing on each remote sensing image in the multi-time series remote sensing image dataset to obtain a standardized remote sensing image set.

[0022] Each remote sensing image in a multi-temporal remote sensing image dataset may have deviations in its radiation value due to the influence of various factors such as the sensor's own characteristics, atmospheric conditions, and sunlight angle. To eliminate these deviations, each remote sensing image needs to be subjected to radiation correction processing.

[0023] There are various methods for radiometric correction, a common one being based on calibration coefficients. Assuming the sensor records a calibration coefficient B when acquiring an image, and the image's original radiometric value is C, the radiometric value D after radiometric correction can be calculated using the formula D = B × C. By performing this radiometric correction calculation on each image in a multi-temporal remote sensing image dataset, standardized radiometric values ​​can be obtained, thereby forming a standardized remote sensing image set. This process ensures consistent radiometric characteristics in the images, facilitating subsequent analysis and processing.

[0024] Step S122: performing multi-band fusion processing on the standardized remote sensing image set to generate a fused image feature map.

[0025] Each image in a standardized remote sensing image collection typically contains information from multiple bands, and images from different bands reflect different features of the target area. To fully utilize this multi-band information, it is necessary to perform multi-band fusion processing on the standardized remote sensing image collection.

[0026] There are many methods for multi-band fusion, such as weighted fusion. Assume that a standardized remote sensing image collection contains E bands of images, with the weights for each band being F1, F2, ..., FE, and the corresponding band image data being G1, G2, ..., GE. When performing weighted fusion, each band's image data is first multiplied by its corresponding weight, and then the results are concatenated. Specifically, for each pixel in the image, the pixel's value in each band is multiplied by the corresponding weight, and then the results are concatenated in a specific order to generate a fused image feature map. This generated fused image feature map integrates information from multiple bands and can more comprehensively reflect the characteristics of the target area's terrain.

[0027] Step S123: performing surface cover classification processing on the fused image feature map to obtain surface cover features of the target area, wherein the surface cover features include vegetation coverage, water body distribution range, and bare land area identification.

[0028] Step S1231: Calling a pre-trained surface cover classification model to perform pixel-level classification processing on the fused image feature map to generate an initial classification result map.

[0029] In order to obtain the surface cover characteristics of the target area, it is necessary to perform surface cover classification processing on the fused image feature map. Here, a pre-trained surface cover classification model can be called. The surface cover classification model is trained with a large amount of labeled sample data and can identify different types of surface cover.

[0030] The fused image feature map is fed into a pre-trained land cover classification model, which classifies each pixel in the image. Based on the pixel's spectral and textural characteristics, the model determines whether the pixel belongs to vegetation, water, or bare ground. This classification generates an initial classification result map, in which each pixel is labeled with a land cover type.

[0031] Step S1232: performing spatial smoothing processing on the initial classification result map, performing region merging processing on the spatially smoothed classification result map, and generating continuous surface coverage area identifiers.

[0032] The initial classification result map may contain some noise and discontinuous classification areas. To make the classification results more accurate and consistent, spatial smoothing is required. Spatial smoothing can be performed using a filtering algorithm, such as a mean filter. For each pixel in the initial classification result map, a neighborhood of a certain size is selected with that pixel as the center. The mean of the land cover types of all pixels in the neighborhood is calculated and used as the new land cover type for that pixel. This process can reduce the impact of noise and achieve smoother classification results.

[0033] After spatial smoothing, the classification result map needs to be merged. Some adjacent small areas may belong to the same land cover type, but due to classification errors or other reasons, they are labeled separately. Through the region merging process, these adjacent small areas of the same land cover type are combined into a continuous large area, thus generating a continuous land cover area label.

[0034] Step S1233: Count the area proportion corresponding to each surface coverage area identifier, generate vegetation coverage, water body distribution range and bare land area identifier, and obtain the surface coverage characteristics of the target area.

[0035] After obtaining continuous surface cover area identification, it is necessary to count the area proportion corresponding to each surface cover area identification. For each surface cover type, calculate the proportion of the corresponding area in the entire target area. For example, for vegetation coverage areas, count its area proportion, and this proportion is the vegetation coverage; for water distribution areas, count its area proportion to obtain relevant information about the water distribution range; for bare land areas, also count its area proportion and identify the area. Through such statistics and identification, the surface cover characteristics of the target area are obtained, including vegetation coverage, water distribution range, and bare land area identification.

[0036] Step S124: reconstructing the digital elevation model on the fused image feature map to generate an elevation data matrix.

[0037] Digital elevation models (DEMs) can reflect the topographical undulations of a target area and are crucial for extracting slope and elevation change characteristics in geological hazard identification. To generate a DEM of the target area, the fused image feature map must be reconstructed into a DEM.

[0038] There are many methods for reconstructing digital elevation models, such as stereo pair matching. An appropriate stereo pair is selected from the fused image feature map. By matching the same-name points in the stereo pair, the disparity of these points is calculated. Based on this disparity information and photogrammetry principles, the elevation value corresponding to each same-name point can be calculated. By arranging these elevation values ​​according to the pixel positions of the image, an elevation data matrix is ​​generated, in which each element corresponds to the elevation value of a pixel in the target area.

[0039] Step S125: Calculating the slope change gradient based on the elevation data matrix to obtain the slope characteristics of the target area.

[0040] After obtaining the elevation data matrix, the slope change gradient can be calculated based on the matrix to obtain the slope characteristics of the target area. The slope change gradient reflects the degree of inclination and change of the terrain.

[0041] The method for calculating the slope gradient can be used to use the difference method. For each pixel in the elevation data matrix, select its adjacent pixels and calculate the elevation difference and horizontal distance between the adjacent pixels to obtain the slope gradient of the pixel. For example, for pixel H, select its adjacent pixels I and J and calculate the elevation differences K1 and K2 between H and I and J, as well as the horizontal distances L1 and L2. Based on these differences and distances, the slope gradient of the pixel in different directions can be calculated. After calculating the slope gradient of all pixels, the slope characteristics of the target area are obtained.

[0042] Step S126: performing difference calculation on the elevation data matrices corresponding to the remote sensing images collected at different times in the multi-time series remote sensing image dataset to obtain elevation change characteristics of the target area.

[0043] Remote sensing images collected at different times in a multi-temporal remote sensing image dataset can all be reconstructed using the digital elevation model to obtain corresponding elevation data matrices. In order to obtain the elevation change characteristics of the target area, it is necessary to perform difference calculations on these elevation data matrices at different times.

[0044] Assume that the elevation data matrices corresponding to the remote sensing images collected at time M1 and time M2 are N1 and N2 respectively. For each corresponding pixel in the elevation data matrix, calculate its elevation difference at the two time points. For example, for pixel P, its elevation value in N1 is Q1, and its elevation value in N2 is Q2, then the elevation difference of the pixel is R=Q2-Q1. After calculating the elevation differences of all pixels, the elevation change characteristics of the target area between the two time points are obtained. By performing such difference calculations on the elevation data matrices of different time combinations in the multi-time series remote sensing image dataset, the elevation change characteristics of the target area in multiple time periods can be obtained.

[0045] Step S130: performing multi-time series change analysis on the terrain feature set to generate a dynamic change feature set of the target area.

[0046] After obtaining the target area's terrain feature set, the next step is to perform a multi-time series analysis on this set to generate a dynamic feature set for the target area. By analyzing how terrain features change over time, we can identify potential geological hazards in the target area.

[0047] Step S131: performing time series change detection processing on the surface cover features to extract the vegetation coverage change rate, water body expansion and contraction trends, and bare land area migration paths.

[0048] Land cover characteristics such as vegetation coverage, water distribution, and bare land area may change over time. To analyze these changes, it is necessary to perform time series change detection on land cover characteristics.

[0049] For vegetation coverage, we compare vegetation coverage values ​​at different times to calculate its rate of change. Assuming the vegetation coverage at time S1 is T1 and at time S2 is T2, then the vegetation coverage change rate U = (T2 - T1) / T1. By performing this calculation on vegetation coverage at multiple time points, we can generate a sequence of vegetation coverage change rates over time, thereby analyzing the changing trends of vegetation coverage.

[0050] For water body distribution, observe the distribution area of ​​the water body over time to determine whether it is expanding or contracting. The expansion or contraction trend can be determined by comparing the size of the water body distribution area over time. For example, if the water body distribution area increases over a certain period of time, it indicates that the water body is expanding; otherwise, it is contracting.

[0051] For bare land areas, their migration paths can be extracted by analyzing their location and shape changes over time. Spatial analysis methods can be used to compare features such as the center of gravity and boundary shape of bare land areas over time to determine their migration direction and distance.

[0052] Step S132: performing time series superposition analysis on the slope characteristics to extract slope gradient mutation areas and continuous erosion areas.

[0053] Slope characteristics may also change at different times. In order to discover abnormalities in these changes, it is necessary to perform time series superposition analysis on the slope characteristics.

[0054] Slope characteristic maps from different times are overlaid to compare the slope changes at each pixel over time. Areas with significant slope changes, i.e., sudden changes in slope gradient, are marked and extracted. These areas may be caused by sudden changes in topography due to geological activities, human engineering activities, or other factors, and may pose a risk of geological disasters.

[0055] At the same time, we analyze the changing trends of slope characteristics at multiple time points to identify areas where the slope continues to increase, i.e., areas of persistent erosion. These areas of persistent erosion may be due to the gradual erosion of the terrain caused by water erosion, weathering, etc., and they also require special attention.

[0056] Step S133: performing time series difference processing on the elevation change characteristics to extract abnormal elevation subsidence areas and uplift areas.

[0057] Step S1331: Arrange the elevation data matrices at different times in chronological order to generate an elevation time series dataset.

[0058] To analyze elevation change characteristics, we first need to arrange the elevation data matrices from different time periods in chronological order to form an elevation time series dataset. This allows us to clearly observe how the elevation of the target area changes over time. The elevation data matrices for each time point are arranged in the order of their acquisition time, with each elevation data matrix corresponding to a specific time point.

[0059] Step S1332: performing pixel-by-pixel difference calculation on the elevation data matrix of adjacent time points in the elevation time series dataset to generate an elevation change rate map.

[0060] After obtaining the elevation time series dataset, the elevation data matrices of adjacent time points are subjected to pixel-by-pixel differential calculations. For two adjacent elevation data matrices in the elevation time series dataset, such as the elevation data matrices W1 and W2 corresponding to time V1 and time V2, the elevation difference is calculated for each corresponding pixel. Assuming that the elevation value of pixel X in W1 is Y1 and the elevation value in W2 is Y2, then the elevation difference of the pixel is Z=Y2-Y1. After calculating the elevation difference of all pixels, the elevation change between adjacent time points is obtained. By performing such calculations on multiple adjacent time points, an elevation change rate map can be generated, which reflects the elevation change rate of each pixel in the target area during different time periods.

[0061] Step S1333: performing threshold segmentation processing on the elevation change rate map, and marking pixels whose change rate exceeds a preset threshold as abnormal change pixels.

[0062] To identify areas of abnormal elevation change in the elevation rate map, threshold segmentation is required. A preset elevation rate threshold is set. For each pixel in the elevation rate map, if its elevation rate exceeds this threshold, the pixel is marked as abnormally changing. The threshold setting should be based on a comprehensive consideration of factors such as the geological background of the target area and historical elevation changes. Through threshold segmentation, pixels with abnormal elevation changes can be quickly identified.

[0063] Step S1334: performing spatial clustering processing on the abnormally changed pixels to generate spatial boundaries of abnormal elevation subsidence areas and elevation areas.

[0064] After marking the abnormally changing pixels, spatial clustering is required to determine the specific scope of the abnormal elevation subsidence and elevation uplift areas. This spatial clustering can be performed using a density-based clustering algorithm, such as the DBSCAN algorithm. The DBSCAN algorithm clusters adjacent, high-density abnormally changing pixels into a single cluster based on the spatial distribution density of the abnormally changing pixels. Clustering can then divide the abnormally changing pixels into distinct regions, each representing an area of ​​abnormal elevation subsidence or elevation. The spatial boundaries of these regions are then generated based on the clustering results, clarifying their specific scope.

[0065] Step S134: The vegetation coverage change rate, water body expansion and contraction trend, bare land migration path, slope gradient mutation area, continuous erosion area, elevation abnormal settlement area and uplift area are combined into a dynamic change feature set.

[0066] After extracting the vegetation coverage change rate, water body expansion and contraction trends, bare land migration paths, slope gradient abrupt changes, persistent erosion, and elevation abnormal subsidence and uplift, these features are combined to form a dynamic change feature set for the target area. This dynamic change feature set integrates dynamic change information from multiple aspects of the target area. These features can be combined according to specific rules, such as by type or spatial location, to form a unified dynamic change feature set.

[0067] Step S140: calling a geological disaster identification model to perform abnormal area identification processing on the dynamic change feature set, and determining a set of abnormal areas with geological disaster risks in the target area.

[0068] After obtaining the target region's dynamic change feature set, the next step is to use the geological hazard identification model to identify abnormal regions within the target region, thereby determining the set of abnormal regions with geological hazard risks. The geological hazard identification model can comprehensively analyze various information in the dynamic change feature set to identify areas with potential geological hazard risks.

[0069] Step S141: inputting the dynamically changing feature set into a pre-trained geological hazard identification model, wherein the geological hazard identification model includes a feature fusion module and a risk prediction module.

[0070] The pre-trained geological hazard identification model consists of a feature fusion module and a risk prediction module. The dynamically changing feature set is input into the model, first entering the feature fusion module. The feature fusion module integrates and optimizes the various features in the dynamically changing feature set to better inform subsequent risk prediction.

[0071] Step S142: performing weight assignment and feature fusion on the dynamically changing feature set through the feature fusion module to generate a comprehensive risk feature map.

[0072] Step S1421: assigning a dynamic weight coefficient to each dynamic change feature in the dynamic change feature set.

[0073] In the feature fusion module, dynamic weight coefficients must first be assigned to each dynamically changing feature in the dynamically changing feature set. Different dynamically changing features may have varying degrees of impact on geological hazard risk, so different weights need to be assigned based on their importance. For example, areas of abnormal elevation subsidence and areas with sudden changes in slope gradient may have a greater impact on geological hazard risk, so they can be assigned larger weights; while features such as the rate of change in vegetation cover have a relatively smaller impact and can be assigned smaller weights. These weight coefficients are not fixed but can be dynamically adjusted based on actual conditions to adapt to different geological environments and hazard types.

[0074] Step S1422: performing weighted splicing on each dynamically changing feature based on the dynamic weight coefficient to obtain a preliminary fusion feature map.

[0075] After assigning dynamic weight coefficients to each dynamic change feature, a weighted concatenation is performed based on these weight coefficients. For each dynamic change feature, its data is multiplied by its corresponding weight coefficient. These weighted feature data are then concatenated in a specific order. For example, features such as vegetation cover change rate and water body expansion / contraction trends are multiplied by their respective weight coefficients and then concatenated to form a preliminary fused feature map. This preliminary fused feature map integrates information from each dynamic change feature, but may still have some shortcomings.

[0076] Step S1423: performing spatial context enhancement processing on the preliminary fused feature map, performing residual connection between the spatial context enhanced feature map and the preliminary fused feature map, and generating a comprehensive risk feature map.

[0077] Spatial context enhancement processing aims to capture the spatial correlation information between pixels in the preliminary fusion feature map to improve the feature expression capability. One feasible way is to use convolution operation. Assuming that the preliminary fusion feature map is A and the convolution kernel is B, the convolution operation can be regarded as a weighted summation of each pixel and its neighborhood in the preliminary fusion feature map. For each pixel in the preliminary fusion feature map A, a neighborhood of a specific size is selected with the pixel as the center, and the pixel values ​​in the neighborhood are multiplied by the weight value of the corresponding position in the convolution kernel B. These products are then added to obtain the new value of the pixel after convolution processing. By sliding the convolution kernel B across the entire preliminary fusion feature map, such convolution calculation is performed on each pixel to generate a feature map C after spatial context enhancement.

[0078] The size and weight of the convolution kernel B are key factors influencing the effectiveness of spatial context enhancement. The kernel size determines the scope of the neighborhood. Larger kernels capture a wider range of spatial information, but the computational effort also increases. Smaller kernels focus on extracting local information. Weights are set based on different task requirements and data characteristics, aiming to highlight spatial features related to geological hazard risk.

[0079] After obtaining the spatially context-enhanced feature map C, it is residually connected to the initially fused feature map A. The purpose of the residual connection is to prevent information loss during the feature fusion process and ensure that the model can learn subtle changes in features. Specifically, the pixel values ​​at corresponding positions in feature map C and feature map A are added together to obtain the new feature map D. Specifically, for pixels at the same position in feature map C and feature map A, their pixel values ​​are added together to obtain the corresponding pixel value in feature map D.

[0080] When performing residual connections, it is necessary to ensure that the dimensions (number) of feature maps C and A match. If the dimensions are inconsistent, dimensionality adjustment may be required, such as through interpolation or pooling, to bring them into alignment to ensure smooth addition. The feature map D obtained through residual connections is the comprehensive risk feature map. It combines the information from the initial fused feature map and the information enhanced by spatial context, and can more comprehensively and accurately reflect the geological hazard risk characteristics of the target area.

[0081] Step S143: performing binary classification processing on the comprehensive risk feature map through the risk prediction module, and outputting the geological disaster risk probability of each pixel.

[0082] The risk prediction module receives the comprehensive risk characteristic map D as input. Its main task is to perform binary classification on each pixel in the comprehensive risk characteristic map, determine whether the area corresponding to the pixel has geological disaster risk, and output the corresponding risk probability.

[0083] The risk prediction module can employ a neural network model, such as a fully connected neural network. The feature vector of each pixel in the comprehensive risk feature map D is used as input to the fully connected neural network. A fully connected neural network consists of multiple neural layers, each containing multiple neurons. Neurons in adjacent layers are connected by weights. During the forward propagation of the neural network, the input feature vector is processed sequentially by each neural layer. Each neuron performs a weighted summation of the input signal, applies a nonlinear transformation using an activation function, and transmits the transformed result to the next layer of neurons.

[0084] Suppose a fully connected neural network has E layers of neurons, the input layer's eigenvector is F, the weight matrix of the neurons in layer i is Gi, the bias vector is Hi, and the activation function is f. In the first layer, the input eigenvector F is multiplied by the weight matrix G1, then added with the bias vector H1. This is then transformed by the activation function f to obtain the output of layer 1, I1. That is, I1 = f(F*G1+H1). Next, the output of layer 1, I1, serves as the input to layer 2, and the above process is repeated to obtain the output of layer 2, I2. This process continues until the output of the last layer is obtained.

[0085] The output of the last layer is a vector containing two elements, representing the scores of the pixel belonging to the class with no geological hazard risk and the class with geological hazard risk, respectively. In order to convert these two scores into probability values, the Softmax function can be used. The Softmax function normalizes the output of the last layer so that the sum of the two elements is 1, and the value of each element represents the probability that the pixel belongs to the corresponding category. For example, the output vector is (J1, J2). After processing with the Softmax function, the probability vector (K1, K2) is obtained, where K1 represents the probability that the pixel belongs to the class with no geological hazard risk, K2 represents the probability that the pixel belongs to the class with geological hazard risk, and K1+K2=1.

[0086] By performing such processing on each pixel in the comprehensive risk characteristic map D, the geological hazard risk probability of each pixel can be obtained.

[0087] Step S144: performing regional aggregation processing on the geological disaster risk probability to generate the abnormal area set.

[0088] After obtaining the geological disaster risk probability of each pixel, these probabilities need to be regionally aggregated to determine the set of abnormal areas with geological disaster risks.

[0089] Regional aggregation processing can adopt a threshold-based method. First, a geological hazard risk probability threshold L is set. For each pixel in the comprehensive risk feature map, if its geological hazard risk probability is greater than the threshold L, the pixel is marked as a suspected abnormal pixel.

[0090] Next, connectivity analysis is performed on these suspected abnormal pixels. Connected region labeling algorithms, such as the four-connected or eight-connected algorithms, are used to divide adjacent suspected abnormal pixels into different connected regions. The four-connected algorithm assumes that a pixel is connected to its four adjacent pixels above, below, left, and right; the eight-connected algorithm assumes that a pixel is connected to its eight adjacent pixels above, below, left, right, upper left, upper right, lower left, and lower right.

[0091] For each connected region, statistics such as the sum or average of the geological hazard risk probabilities of all pixels in the region are calculated to assess the overall geological hazard risk level of the region. Based on these statistics, connected regions with higher risk levels are screened and identified as abnormal areas.

[0092] Finally, all the identified abnormal areas are integrated to generate an abnormal area set, which contains information about all abnormal areas with geological disaster risks in the target area, such as their location and range.

[0093] Step S150: Generate a disaster warning signal according to the abnormal area set, and send the disaster warning signal to a disaster monitoring terminal.

[0094] After determining the set of abnormal areas with geological disaster risks in the target area, it is necessary to generate a disaster warning signal based on the set and send the signal to the disaster monitoring terminal so that timely response measures can be taken.

[0095] Step S151: performing risk level classification processing on each abnormal area in the abnormal area set to generate high-risk area, medium-risk area and low-risk area identification.

[0096] Step S1511: Obtain the geological disaster risk probability, historical disaster frequency and surrounding population density data corresponding to each abnormal area.

[0097] In order to accurately classify the risk levels of abnormal areas, it is necessary to obtain the geological disaster risk probability, historical disaster frequency and surrounding population density data corresponding to each abnormal area.

[0098] The geological disaster risk probability has been obtained in step S143, that is, the statistical value of the geological disaster risk probability of all pixels in each abnormal area, such as the average value or the sum.

[0099] The frequency of historical disasters can be obtained by consulting historical geological disaster records. For each abnormal area, the number of geological disasters that occurred in the area over the past period of time is counted, and then the frequency of historical disasters is calculated based on the time span.

[0100] Surrounding population density data can be obtained through census data, geographic information system (GIS) data, etc. For each abnormal area, the population within a certain range around it is determined, and the surrounding population density is calculated based on the area of ​​this range.

[0101] Step S1512: Input the geological disaster risk probability, historical disaster frequency and surrounding population density data into the risk level assessment model, and output the risk score of each abnormal area.

[0102] The risk level assessment model is a trained model that can output the risk score of each abnormal area based on the input geological disaster risk probability, historical disaster occurrence frequency, and surrounding population density data.

[0103] The risk level assessment model can adopt a linear regression model or other machine learning models. Assuming a linear regression model is adopted, its expression is the risk score M = a * geological disaster risk probability + b * historical disaster occurrence frequency + c * surrounding population density + d, where a, b, and c are weight coefficients. The geological disaster risk probability, historical disaster occurrence frequency, and surrounding population density data need to be normalized before being input into the risk level assessment model, and d is the bias term. These weight coefficients and bias terms are obtained through learning with a large amount of sample data during the model training process.

[0104] After normalizing and transforming the geological disaster risk probability, historical disaster occurrence frequency, and surrounding population density data of each abnormal area, substitute them into the above expression to calculate the risk score of each abnormal area.

[0105] Step S1513: Divide the risk score into high-risk intervals, medium-risk intervals, and low-risk intervals according to a preset risk score threshold.

[0106] Preset two risk score thresholds N1 and N2 (N1 < N2), and divide the risk score into three intervals according to these two thresholds. If the risk score of an abnormal area is greater than N2, then this area belongs to the high-risk interval; if the risk score is between N1 and N2, then this area belongs to the medium-risk interval; if the risk score is less than N1, then this area belongs to the low-risk interval.

[0107] Step S1514: Assign corresponding risk level identifiers to each abnormal area.

[0108] According to the risk interval to which the abnormal area belongs, assign corresponding risk level identifiers to each abnormal area. For example, for an abnormal area belonging to the high-risk interval, assign a high-risk area identifier; for an abnormal area belonging to the medium-risk interval, assign a medium-risk area identifier; for an abnormal area belonging to the low-risk interval, assign a low-risk area identifier.

[0109] Step S152: Generate corresponding early warning level signals according to the risk level division result.

[0110] Based on the risk level of each abnormal area, a corresponding warning level signal is generated. For example, a high-risk area generates a high-level warning signal; a medium-risk area generates a medium-level warning signal; and a low-risk area generates a low-level warning signal. These warning level signals can be represented by different colors, symbols, or text to ensure clear display on the disaster monitoring terminal.

[0111] Step S153: Associating the warning level signal with the geographic coordinates of the abnormal area to generate a disaster warning signal.

[0112] To accurately locate anomaly areas, the disaster monitoring terminal needs to associate the warning level signal with the geographic coordinates of the anomaly area. For each anomaly area, its geographic coordinates, such as longitude and latitude, are obtained. The warning level signal for that anomaly area is then combined with the geographic coordinates to generate a disaster warning signal containing both the warning level and the geographic coordinates.

[0113] Step S154: The disaster warning signal is transmitted to the disaster monitoring terminal in real time via the wireless communication network, triggering a terminal warning prompt.

[0114] Using wireless communication networks, such as 4G and 5G, the generated disaster warning signals are transmitted in real time to the disaster monitoring terminal. Upon receiving the disaster warning signal, the disaster monitoring terminal triggers corresponding warning prompts, such as sound alarms, vibration prompts, and screen pop-ups, to remind relevant personnel to pay attention to the geological disaster risk in the target area.

[0115] Step S210: Collect remote sensing image datasets of historical geological disaster cases and corresponding disaster annotation data to construct a training sample set.

[0116] In order to train the geological disaster identification model, it is necessary to collect remote sensing image datasets of historical geological disaster cases and corresponding disaster annotation data to construct a training sample set.

[0117] Remote sensing image datasets of historical geological disaster cases can be obtained from multiple channels, such as satellite image databases and aerial remote sensing data centers. Remote sensing images of geological disasters occurring at different times and in different regions should be collected to ensure image diversity and representativeness.

[0118] The corresponding disaster annotation data is the annotation information of geological disaster areas in remote sensing images. Geological disaster areas in remote sensing images can be annotated through manual visual interpretation or other automated annotation methods, such as the boundaries and scope of disaster areas such as landslides and mudslides.

[0119] The collected remote sensing image datasets and the corresponding disaster annotated data are sorted and paired. Each pair of remote sensing image and disaster annotated data constitutes a training sample. All training samples are combined together to construct a training sample set.

[0120] Step S220: performing terrain feature extraction and dynamic change analysis processing on the remote sensing image dataset in the training sample set to generate a sample dynamic change feature set.

[0121] The process of extracting terrain features and performing dynamic change analysis on the remote sensing image dataset in the training sample set is similar to steps S120 - S130 .

[0122] First, radiometric correction is performed on the remote sensing image dataset to generate a standardized remote sensing image set. Multi-band fusion is then performed on this standardized remote sensing image set to generate a fused image feature map. The fused image feature map is then subjected to land cover classification and digital elevation model reconstruction to determine land cover characteristics, slope characteristics, and elevation change characteristics.

[0123] Afterwards, these terrain features were subjected to multi-time series change analysis and processing, including time series change detection processing of surface cover characteristics, time series superposition analysis processing of slope characteristics, time series difference processing of elevation change characteristics, etc., to extract dynamic change characteristics such as vegetation coverage change rate, water body expansion and contraction trend, bare land area migration path, slope gradient mutation area, continuous erosion area, elevation abnormal subsidence area and uplift area.

[0124] Finally, these dynamic change features are merged into a sample dynamic change feature set.

[0125] Step S230: input the sample dynamic change feature set into the initial geological disaster identification model, and output the predicted disaster area.

[0126] The generated sample dynamic change feature set is input into the initial geological hazard identification model. The structure of the initial geological hazard identification model is the same as the pre-trained geological hazard identification model used in step S140, including a feature fusion module and a risk prediction module.

[0127] The feature fusion module assigns weights and fuses features from the dynamically changing sample features to generate a comprehensive risk profile. The risk prediction module performs binary classification on the comprehensive risk profile and outputs the geological hazard risk probability for each pixel. Based on these probabilities, regional aggregation is used to determine the predicted disaster area.

[0128] Step S240: Calculate the difference loss between the predicted disaster area and the disaster annotation data, and update the model parameters based on the difference loss.

[0129] In order to evaluate the performance of the initial geological hazard identification model, it is necessary to calculate the difference loss between the predicted hazard area and the hazard annotation data. The cross entropy loss function can be used to calculate the difference loss.

[0130] For each training example, the predicted geological hazard risk probability for each pixel in the disaster area is compared with the corresponding pixel's true label (hazard or no hazard) in the disaster annotated data. The cross-entropy loss function calculates a loss value based on the difference between the two. This loss value reflects the degree of deviation between the model's prediction and the actual situation.

[0131] The sum of the difference losses of all training samples is calculated to obtain the total difference loss. Based on this total difference loss, an optimization algorithm, such as stochastic gradient descent, is used to update the parameters of the initial geological hazard identification model. The stochastic gradient descent algorithm adjusts the model parameters based on the gradient of the difference loss with respect to the model parameters, gradually reducing the difference loss.

[0132] Step S250: Repeat the training until the model converges to obtain a pre-trained geological hazard identification model.

[0133] Repeat steps S230-S240, continuously input the sample dynamic change feature set into the model, calculate the difference loss and update the model parameters. As training progresses, the performance of the model will gradually improve and the difference loss will gradually decrease.

[0134] When the difference loss no longer decreases significantly or reaches the preset number of training rounds, the model is considered to have converged. At this point, the obtained model is a pre-trained geological hazard identification model and can be used for actual geological hazard identification tasks.

[0135] Through the above steps, geological disaster identification combined with remote sensing technology can be realized. It starts with obtaining a multi-time series remote sensing image data set, and then goes through terrain feature extraction, dynamic change analysis, and abnormal area identification. Finally, a disaster warning signal is generated and sent to the disaster monitoring terminal. It also includes the training process of the geological disaster identification model to ensure that the model can accurately identify geological disaster risk areas.

[0136] In the above embodiments, the geological hazard identification system combined with remote sensing technology for executing the above method embodiments has at least one processor, a control module (chip set) coupled to at least one of the (at least one) processors, a memory coupled to the control module, a non-volatile memory (NVM) / storage device coupled to the control module, at least one load / output device coupled to the control module, and a network interface coupled to the control module.

[0137] The processor may include at least one single-core or multi-core processor, and the processor may include any combination of general-purpose processors or special-purpose processors (such as graphics processors, application processors, baseband processors, etc.). For some alternative implementations, the geological disaster identification system combined with remote sensing technology can be based on electronic devices such as the gateway described in the embodiments of this application.

[0138] For some alternative embodiments, a geological hazard identification system combined with remote sensing technology may include at least one computer-readable medium (e.g., a memory or NVM / storage device) having instructions and at least one processor integrated with the at least one computer-readable medium and configured to execute instructions to implement a module to perform the actions described in the present disclosure.

[0139] For one embodiment, the control module may include any suitable interface controller to provide any suitable interface to at least one of the processor(s) and / or any suitable device or component in communication with the control module.

[0140] The control module may include a memory controller module to provide an interface to the memory. The memory controller module may be a hardware module, a software module, and / or a firmware module.

[0141] The memory may be used, for example, to load and store data and / or instructions for a geological hazard identification system combined with remote sensing technology. For one embodiment, the memory may include any suitable volatile memory, for example, a suitable DRAM.

[0142] For one embodiment, the control module may include at least one load / output controller to provide an interface to the NVM / storage device and the (at least one) load / output device.

[0143] For example, NVM / storage devices may be used to store data and / or instructions. The NVM / storage devices may include any suitable non-volatile memory (e.g., flash memory) and / or may include any suitable (at least one) non-volatile storage device (e.g., at least one hard disk drive (HDD), at least one compact disc (CD) drive, and / or at least one digital versatile disc (DVD) drive).

[0144] The NVM / storage device may include a storage resource that is physically part of the device on which the remote sensing-based geohazard identification system is installed, or it may be accessible to the device without being part of the device. For example, the NVM / storage device may be accessed via (at least one) load / output device over a network.

[0145] The (at least one) load-in / output device may provide an interface for the geological hazard identification system combined with remote sensing technology to communicate with any other appropriate device, and the load-in / output device may include a communication component, a phonetic component, a sensor component, etc. The network interface may provide an interface for the geological hazard identification system combined with remote sensing technology to communicate based on at least one network, and the geological hazard identification system combined with remote sensing technology may wirelessly communicate with at least one component of a wireless network based on any of at least one wireless network priors and / or protocols, for example, access a wireless network based on a communication prior.

[0146] For one embodiment, at least one of the (at least one) processors may be loaded together with the logic of at least one controller of a control module (e.g., a memory controller module). For one embodiment, at least one of the (at least one) processors may be loaded together with the logic of at least one controller of a control module to form a system-level load. For one embodiment, at least one of the (at least one) processors may be fused on the same die with the logic of at least one controller of a control module. For one embodiment, at least one of the (at least one) processors may be fused on the same die with the logic of at least one controller of a control module to form a system-on-chip (SoC).

[0147] The above is a detailed introduction to the embodiments of the present application. Specific examples are used herein to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method and core idea of ​​the present application. At the same time, for those skilled in the art, based on the ideas of the present application, there may be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

[0148] An embodiment of the present invention discloses a computer-readable storage medium storing a computer program for electronic data exchange, wherein the computer program enables a computer to execute the steps of the geological hazard identification method combined with remote sensing technology described in the aforementioned embodiment.

[0149] An embodiment of the present invention discloses a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program, and the computer program is operable to enable a computer to execute the steps of the geological disaster identification method combined with remote sensing technology described in the above embodiment.

[0150] The device embodiments described above are merely illustrative, wherein the modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical modules, i.e., they may be located in one place or distributed across multiple modules. Some or all of the modules may be selected based on actual needs to achieve the purpose of the present embodiment. Those skilled in the art can understand and implement the present invention without inventive effort.

[0151] Through the detailed description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, or of course, by means of hardware. Based on this understanding, the above technical solution, in essence, or the portion that contributes to the prior art, can be embodied in the form of a software product, which can be stored in a computer-readable storage medium, including a read-only memory (ROM), a random access memory (RAM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), a one-time programmable read-only memory (OTPROM), an electronically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM), or other optical disk storage, magnetic disk storage, magnetic tape storage, or any other computer-readable medium capable of storing or storing data.

[0152] Finally, it should be noted that what is disclosed above is only a preferred embodiment of the present invention, which is only used to illustrate the technical solution of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments can still be modified, or some of the technical features therein can be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for identifying geological hazards using remote sensing technology, characterized in that: The method comprises: Acquire a multi-time series remote sensing image dataset of a target area, wherein the multi-time series remote sensing image dataset includes remote sensing images acquired at different times; Performing terrain feature extraction processing on the multi-time series remote sensing image dataset to obtain a terrain feature set of the target area, wherein the terrain feature set includes surface cover features, slope features, and elevation change features; Performing multi-time series change analysis on the terrain feature set to generate a dynamic change feature set of the target area; Calling a geological disaster identification model to perform abnormal area identification processing on the dynamic change feature set to determine a set of abnormal areas with geological disaster risks in the target area; A disaster warning signal is generated according to the abnormal area set, and the disaster warning signal is sent to a disaster monitoring terminal.

2. The geological disaster identification method combined with remote sensing technology according to claim 1 is characterized in that: The performing terrain feature extraction processing on the multi-time series remote sensing image dataset to obtain a terrain feature set of the target area includes: Performing radiation correction processing on each remote sensing image in the multi-time series remote sensing image dataset to obtain a standardized remote sensing image set; Performing multi-band fusion processing on the standardized remote sensing image set to generate a fused image feature map; Performing surface cover classification processing on the fused image feature map to obtain surface cover features of the target area, wherein the surface cover features include vegetation coverage, water body distribution range, and bare land area identification; Performing digital elevation model reconstruction processing on the fused image feature map to generate an elevation data matrix; Calculating the slope change gradient based on the elevation data matrix to obtain the slope characteristics of the target area; The elevation data matrix corresponding to the remote sensing images collected at different times in the multi-time series remote sensing image data set is subjected to difference calculation to obtain the elevation change characteristics of the target area.

3. The geological disaster identification method combined with remote sensing technology according to claim 2 is characterized in that: The performing surface cover classification processing on the fused image feature map to obtain the surface cover features of the target area includes: Calling a pre-trained land cover classification model to perform pixel-level classification processing on the fused image feature map to generate an initial classification result map; Performing spatial smoothing processing on the initial classification result map, and performing regional merging processing on the spatially smoothed classification result map to generate continuous surface coverage area identifiers; The area proportion corresponding to each surface coverage area identifier is counted to generate vegetation coverage, water body distribution range and bare land area identifier, and the surface coverage characteristics of the target area are obtained.

4. The geological disaster identification method combined with remote sensing technology according to claim 1 is characterized in that: The performing multi-time series change analysis on the terrain feature set to generate a dynamic change feature set of the target area includes: Performing time series change detection processing on the surface cover characteristics to extract vegetation coverage change rate, water body expansion and contraction trends, and bare land area migration paths; Performing time series superposition analysis on the slope characteristics to extract slope gradient mutation areas and continuous erosion areas; Performing time series difference processing on the elevation change characteristics to extract abnormal elevation subsidence areas and uplift areas; The vegetation coverage change rate, water body expansion and contraction trend, bare land migration path, slope gradient mutation area, continuous erosion area, elevation abnormal subsidence area and uplift area are combined into a dynamic change feature set.

5. The geological disaster identification method combined with remote sensing technology according to claim 4 is characterized in that: The performing time series difference processing on the elevation change characteristics to extract abnormal elevation subsidence areas and uplift areas includes: Arrange the elevation data matrices at different times in chronological order to generate an elevation time series dataset; Performing pixel-by-pixel difference calculation on the elevation data matrix of adjacent time points in the elevation time series dataset to generate an elevation change rate map; Performing threshold segmentation processing on the elevation change rate map, marking pixels whose change rate exceeds a preset threshold as abnormal change pixels; The abnormal change pixels are subjected to spatial clustering processing to generate spatial boundaries of abnormal elevation subsidence areas and uplift areas.

6. The geological disaster identification method combined with remote sensing technology according to claim 1 is characterized in that: The calling of the geological disaster identification model to perform abnormal area identification processing on the dynamic change feature set to determine the abnormal area set with geological disaster risks in the target area includes: Inputting the dynamically changing feature set into a pre-trained geological hazard identification model, wherein the geological hazard identification model includes a feature fusion module and a risk prediction module; The feature fusion module performs weight assignment and feature fusion on the dynamically changing feature set to generate a comprehensive risk feature map; Performing binary classification processing on the comprehensive risk feature map through the risk prediction module, and outputting the geological disaster risk probability of each pixel; The geological disaster risk probability is subjected to regional aggregation processing to generate the abnormal area set.

7. The method for identifying geological hazards using remote sensing technology according to claim 6, wherein: The step of performing weight assignment and feature fusion on the dynamically changing feature set by the feature fusion module to generate a comprehensive risk feature map includes: Assigning a dynamic weight coefficient to each dynamically changing feature in the dynamically changing feature set; Perform weighted splicing on each dynamically changing feature based on the dynamic weight coefficient to obtain a preliminary fusion feature map; The preliminary fusion feature map is subjected to spatial context enhancement processing, and the feature map after spatial context enhancement is residually connected with the preliminary fusion feature map to generate a comprehensive risk feature map.

8. The method for identifying geological hazards using remote sensing technology according to claim 1, wherein: Generating a disaster warning signal according to the abnormal area set and sending the disaster warning signal to a disaster monitoring terminal includes: Performing risk level classification processing on each abnormal area in the abnormal area set to generate high-risk area, medium-risk area and low-risk area identification; Generate a corresponding warning level signal according to the risk level classification result; Associating the warning level signal with the geographical coordinates of the abnormal area to generate a disaster warning signal; Transmitting the disaster warning signal to the disaster monitoring terminal in real time via a wireless communication network to trigger a terminal warning prompt; The step of performing risk classification on each abnormal area in the abnormal area set to generate high-risk area, medium-risk area, and low-risk area identifiers includes: Obtain data on geological disaster risk probability, historical disaster frequency, and surrounding population density corresponding to each abnormal area; Input the geological disaster risk probability, historical disaster frequency and surrounding population density data into the risk level assessment model, and output the risk score for each abnormal area; Dividing the risk score into a high-risk range, a medium-risk range, and a low-risk range according to a preset risk score threshold; Assign a corresponding risk level identification to each abnormal area.

9. The geological disaster identification method combined with remote sensing technology according to claim 1, characterized in that: The method also includes a training process of a disaster identification model, including: Collect remote sensing image datasets of historical geological disaster cases and corresponding disaster annotation data to construct a training sample set; Performing terrain feature extraction and dynamic change analysis on the remote sensing image data set in the training sample set to generate a sample dynamic change feature set; Inputting the sample dynamic change feature set into the initial geological disaster identification model and outputting the predicted disaster area; Calculating the difference loss between the predicted disaster area and the disaster annotated data, and updating the model parameters based on the difference loss; Repeat the training until the model converges to obtain a pre-trained geological hazard identification model.

10. A geological disaster identification system combined with remote sensing technology, characterized in that: The method comprises a processor and a computer-readable storage medium, wherein the computer-readable storage medium stores machine-executable instructions, and when the machine-executable instructions are executed by a computer, the method for identifying geological disasters combined with remote sensing technology as described in any one of claims 1 to 9 is implemented.

Images