Geological disaster remote sensing image processing system based on multi-source data
By constructing a multi-source data fusion processing system and employing deep learning and ensemble learning models, the problems of collaborative fusion and identification accuracy in multi-source data processing were solved, generating high-precision geological disaster distribution maps that meet the needs of professional applications.
Patent Information
- Application Number
- CN202511542295.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-26
- Publication Date
- 2026-01-16
Smart Images
Figure CN121350992A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of geological disaster monitoring and remote sensing image processing technology, and more specifically, to a geological disaster remote sensing image processing system based on multi-source data. Background Technology
[0002] The field of remote sensing monitoring of geological hazards has evolved rapidly in recent years along with the advancements in Earth observation technology. Early technologies primarily relied on single-source optical remote sensing imagery, with hazard identification achieved through manual visual interpretation—a process that was inefficient and highly dependent on expert experience. Subsequently, technological development entered a phase of collaborative application of multi-source data, introducing new types of remote sensing data such as synthetic aperture radar (SAR) and lidar to compensate for blind spots in optical imagery under vegetation cover and severe weather conditions. Simultaneously, artificial intelligence algorithms such as machine learning and deep learning have been widely applied to image classification and target recognition, driving the field towards automation and intelligence. Achieving high-precision and high-efficiency geological hazard identification using massive amounts of multi-source data has become a clear technological trend.
[0003] However, existing technologies still have significant shortcomings. First, most systems process multi-source data in a relatively isolated manner, lacking effective collaborative fusion mechanisms. They typically only perform simple data stacking or result-level fusion, failing to fully leverage the complementary advantages between different data sources, resulting in insufficient information utilization. Second, in the intelligent identification stage, the generalization ability and adaptability of algorithm models are insufficient. When facing complex and ever-changing geographical environments, false alarms and missed alarms are common. Furthermore, feature engineering largely relies on manual design, and the level of automation and intelligence needs to be deepened. Finally, post-processing of identification results is often neglected, lacking a specific optimization process for the morphological characteristics of geological hazards. This results in rough boundaries and a large amount of noise in the final output, making it difficult to meet the accuracy requirements of practical applications.
[0004] Therefore, this paper proposes a geological hazard remote sensing image processing system based on multi-source data to address the above-mentioned problems. The system aims to solve the following core issues: how to systematically integrate and fuse multi-source heterogeneous remote sensing geographic data to overcome the limitations of a single data source; how to improve the accuracy, robustness, and automation level of machine learning models in intelligent geological hazard identification, reducing misjudgments; and how to effectively optimize and refine the preliminary identification results to produce high-precision hazard distribution maps and standardized reports that can be directly used for professional assessment. Summary of the Invention
[0005] In order to overcome the above-mentioned defects of the prior art, embodiments of the present invention provide a geological disaster remote sensing image processing system based on multi-source data to solve the problems mentioned in the background art.
[0006] To achieve the above objectives, the present invention provides the following technical solution: a geological disaster remote sensing image processing system based on multi-source data, the system comprising:
[0007] The multi-source data input module is used to receive remote sensing image data and auxiliary geographic information data from different remote sensing platforms;
[0008] The data processing and fusion module is connected to the multi-source data input module and is used to preprocess, register, and perform fusion processing based on the feature layer or decision layer on various types of input data to generate fused data.
[0009] The intelligent geological disaster identification module is connected to the data processing and fusion module. It is used to extract features and classify the fused data using a built-in machine learning model and output information on suspected geological disaster areas.
[0010] The post-processing and optimization module is connected to the intelligent geological disaster identification module and is used to perform morphological processing and false judgment elimination on the identified suspected geological disaster areas to generate optimized identification results.
[0011] The results output and visualization module is connected to the post-processing and optimization module, which is used to overlay and integrate the final recognition results with the original remote sensing images or basic geographic base maps, and generate visualized thematic maps and data reports.
[0012] Preferably, the remote sensing image data received by the multi-source data input module includes two or more of optical satellite imagery, synthetic aperture radar imagery, and lidar point cloud data, and the received auxiliary geographic information data includes one or more of digital elevation model data, slope and aspect data, geological lithology data, and precipitation monitoring data.
[0013] Preferably, the preprocessing operations performed by the data processing and fusion module include one or more of radiometric calibration, atmospheric correction, geometric correction, image enhancement, and noise suppression. The data processing and fusion module uses an image registration algorithm to unify data from different sources into the same geographic coordinate system, and uses pixel-level fusion, feature-level fusion, or decision-level fusion methods to generate the fused data.
[0014] Preferably, the machine learning model built into the intelligent geological disaster identification module is a trained deep learning neural network model or an ensemble learning model. The module automatically extracts spectral features, texture features, shape features and elevation change features from the fused data through the model, and completes the classification and preliminary identification of landslide, collapse and debris flow geological disaster types based on the extracted features.
[0015] Preferably, the post-processing and optimization module filters the preliminary identification results by applying set area thresholds, shape index ranges, or spatial overlay analysis with known geological conditions to remove obviously misjudged non-geological hazard patches, and uses morphological opening or closing operations to smooth and optimize the boundaries of the retained suspected hazard areas.
[0016] Preferably, the output results and the visualization thematic map generated by the visualization module simultaneously display the original remote sensing image background and the geological hazard identification and zoning results superimposed on it in the form of layers. The data report includes the identified geological hazard locations, areas, confidence levels, and geographical location information.
[0017] Preferably, the system also includes a user interaction and control module, which provides a graphical user interface that allows users to select data sources, set data processing parameters, start processing flows, interactively check intermediate and final results, and export the thematic maps and data reports.
[0018] Preferably, the data connection and call between the modules of the system are implemented through a predefined application programming interface to ensure the data flow and execution of control logic between modules.
[0019] The technical effects and advantages of this invention are as follows:
[0020] Compared to existing technologies, this invention constructs a systematic multi-source data fusion processing flow. First, it utilizes standardized interfaces to access heterogeneous data. Then, it employs preprocessing algorithms tailored to different data characteristics and a fusion method based on feature point registration, achieving efficient integration and complementary advantages of multi-source information. This approach effectively overcomes the limitations of single data sources being susceptible to environmental interference, improving data quality and information integrity in subsequent identification stages. Its advantage lies in providing a richer and more reliable data foundation for accurate identification.
[0021] Compared to existing technologies, this invention achieves automated intelligent identification and classification of geological hazard bodies by introducing advanced machine learning models (such as U-Net and XGBoost) and combining them with multi-dimensional feature combinations (spectral, texture, shape, and topography) specifically designed for geological hazards. This approach reduces reliance on human experience, enhances the algorithm's generalization ability and identification accuracy in different geographical environments, and its advantage lies in its ability to quickly and accurately identify hazard targets such as landslides, collapses, and debris flows from complex scenes, greatly improving monitoring efficiency.
[0022] Compared to existing technologies, this invention adds a dedicated post-processing and optimization module, comprehensively utilizing morphological processing, spatial relationship filtering, and boundary optimization algorithms based on active contour models to refine the preliminary identification results. This approach effectively eliminates obvious misclassified patches and smoothly optimizes disaster boundaries. Its advantages lie in producing more realistic disaster boundaries and cleaner results, generating high-quality thematic outputs that can be directly used for disaster assessment and decision-making. Attached Figure Description
[0023] Figure 1 This is a system framework diagram of the present invention. Detailed Implementation
[0024] 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. Embodiment 1:
[0025] As attached Figure 1 The system shown is a remote sensing image processing system for geological hazards based on multi-source data. The system includes:
[0026] The multi-source data input module is used to receive raw remote sensing image data and auxiliary geographic information data from various remote sensing platforms through a standardized data interface. This module includes a data format parsing unit, a data quality check unit, and a metadata extraction unit. The data format parsing unit supports parsing standard raster data formats such as GeoTIFF, HDF5, and NetCDF, as well as vector data formats such as Shapefile and GeoJSON. The data quality check unit performs usability screening on the input data based on preset signal-to-noise ratio thresholds, cloud coverage ratio thresholds, and spatial resolution thresholds. The metadata extraction unit automatically reads the data acquisition time, sensor type, coordinate system information, and projection parameters and generates a standardized metadata table.
[0027] The data processing and fusion module is connected to the multi-source data input module via an internal data bus. This module includes a preprocessing submodule, a registration submodule, and a data fusion submodule. The preprocessing submodule uses the dark pixel method for atmospheric correction of optical images, an adaptive filtering algorithm for speckle noise suppression of synthetic aperture radar images, and an interpolation algorithm based on irregular triangular meshes to generate a digital elevation model for lidar point cloud data. The registration submodule uses an automatic registration algorithm based on SIFT feature point detection and RANSAC mismatch removal to unify multi-source data to the same geographic coordinate system. The data fusion submodule selects pixel-level fusion algorithms such as Brovey transform or Gram-Schmidt spectral sharpening, feature-level fusion algorithms such as coefficient fusion based on wavelet transform, or decision-level fusion algorithms such as DS evidence theory to integrate multi-source information and generate a high-precision fused dataset with a unified geographic reference.
[0028] The registration submodule employs a feature point detection and description algorithm based on Scale Invariant Feature Transform (SIFT). Its core lies in constructing a Gaussian difference pyramid and detecting extreme points to determine the location and scale of key points. The orientation θ of a key point is determined through its neighborhood gradient histogram.
[0029] m(x,y)=√((L(x+1,y)-L(x-1,y)) 2 +(L(x,y+1)-L(x,y-1)) 2 )
[0030] θ(x,y)=arctan((L(x,y+1)-L(x,y-1)) / (L(x+1,y)-L(x-1,y)))
[0031] Where m(x,y) represents the gradient magnitude at point (x,y), θ(x,y) represents the gradient direction at that point, and L represents the Gaussian scale spatial image. Subsequently, the Random Sample Consensus (RANSAC) algorithm is used to remove mismatches from the initially matched feature point pairs. The optimal homography transformation matrix H is then calculated iteratively.
[0032] H = argmin h ∑ i ρ(||x i '-Hx i || 2 )
[0033] Where, x i and x i Let ' be the homogeneous coordinates of a pair of matching feature points, and ρ be a loss function (such as Huber loss) used to reduce the influence of outliers. This process ultimately achieves sub-pixel level registration accuracy.
[0034] The intelligent geological disaster identification module calls the output of the data processing and fusion module through the data interface. This module includes a feature extraction unit and a classification and identification unit. The feature extraction unit uses a convolutional neural network to automatically extract multi-level spectral features, gray-level co-occurrence matrix texture features, shape factors and elevation variation coefficients from the fused data. The classification and identification unit uses a support vector machine or random forest classifier to perform supervised classification on the extracted feature vectors, so as to achieve preliminary identification of geological disaster targets such as landslides, collapses, and debris flows and output a binary suspected area distribution map.
[0035] The feature extraction unit extracts texture features using the gray-level co-occurrence matrix. For a given image window and pixel pair offset (Δx, Δy), it calculates the gray-level co-occurrence matrix P(i,j) and uses it to calculate features such as the second angular moment (energy), contrast, and correlation. The formula for calculating the contrast Con is as follows:
[0036] Con=∑ i ∑ j (ij) 2 P(i,j)
[0037] The formula for calculating the correlation Cor is:
[0038] Cor = ∑ i ∑ j [(i-μ i )(j-μ j )P(i,j)] / (σ i σ j )
[0039] Where i and j are the row and column indices (i.e., gray levels) in the co-occurrence matrix, respectively, P(i,j) represents the probability estimate of gray levels i and j co-occurring under a specific spatial relationship, and μ i ,μ j and σ i ,σ j Let $\mathbf$ be the mean and standard deviation of the rows and columns of matrix $P$, respectively.
[0040] These two formulas quantitatively define the key texture features (contrast and correlation) used for geological hazard identification, clarify how these features are derived from the original image data through mathematical calculations, reflect the scientific nature and interpretability of the feature extraction process, and form the basis of the intelligent identification module.
[0041] The post-processing and optimization module receives the output results from the intelligent geological disaster identification module. This module includes a misjudgment removal unit and a boundary optimization unit. The misjudgment removal unit uses a morphological filtering algorithm combined with area threshold and shape index threshold to remove sporadic noise points and misjudged areas that do not conform to the morphological characteristics of geological disasters. The boundary optimization unit uses a region growing algorithm based on the level set method to refine the boundaries of the remaining suspected disaster areas and generate an optimized geological disaster distribution vector layer.
[0042] The results output and visualization module integrates the output vector data and original multi-source image data from the post-processing and optimization module. This module includes a thematic map mapping unit and a report generation unit. The thematic map mapping unit uses a layered color method to overlay and render geological disaster distribution data with a geographic base map and generate thematic maps that conform to mapping specifications. The report generation unit automatically extracts the location coordinates of disaster points, disaster area, confidence index, and geological environmental parameters and generates a structured data report.
[0043] The remote sensing image data received by the multi-source data input module includes, but is not limited to, two or more combinations of multispectral optical satellite imagery with a spatial resolution between 0.5 meters and 30 meters, synthetic aperture radar imagery with polarization modes of HH / HV / VH / VV, and lidar point cloud data with a point density greater than 100 points per second. The received auxiliary geographic information data includes one or more of the following: digital elevation model data, slope and aspect data calculated by the Sobel operator based on the digital elevation model, vector geological map data with geological age and lithology as attributes, and precipitation monitoring grid data with a time resolution of not less than 1 hour. This module automatically downloads online data sources by implementing the OGC standard network data service interface, and also supports batch import of local data files. During the data reception process, integrity verification and spatial reference consistency checks are implemented, and a re-acquisition mechanism is triggered for data that does not meet the requirements.
[0044] The data processing and fusion module performs the following preprocessing operations: for optical images, it uses the 6S radiative transfer model for atmospheric correction and the MODTRAN model for aerosol optical thickness compensation; for synthetic aperture radar images, it uses the Gamma MAP filtering algorithm for speckle suppression and implements topographic radiative correction to eliminate geometric distortion; for lidar point cloud data, it uses a digital elevation model generation algorithm based on Kriging interpolation and separates ground points from non-ground points through moving surface filtering.
[0045] The registration submodule adopts an automatic registration method based on SURF feature descriptors. It completes the initial matching by calculating the main direction of the feature points and the description vector, and then uses the RANSAC algorithm to remove mismatched point pairs and solve the affine transformation parameters to achieve sub-pixel level registration accuracy.
[0046] The data fusion submodule selects a fusion strategy based on the application scenario. For applications with high requirements for spectral information, principal component analysis fusion is used to preserve spectral characteristics. For applications with high requirements for spatial details, high-pass filtering fusion is used to enhance texture information. For classification and recognition applications, a feature-level fusion method based on neural networks is used to extract the optimal feature combination.
[0047] The machine learning model built into the geological disaster intelligent identification module is a deep learning neural network model based on the U-Net architecture or an ensemble learning model based on XGBoost. The U-Net model adopts an encoder-decoder structure to fuse shallow detail features and deep semantic features through skip connections. The encoder part uses a VGG16 network to extract multi-level feature maps, and the decoder part gradually restores the spatial resolution and outputs pixel-level classification results through deconvolution layers.
[0048] The XGBoost model employs a gradient boosting decision tree algorithm, which iteratively generates weak classifiers and combines them in a weighted manner to form a strong classifier. During training, cross-validation is used to determine the optimal tree depth and learning rate parameters. The feature extraction unit comprehensively utilizes the gray-level co-occurrence matrix to calculate texture feature parameters such as angular second moment, contrast, and correlation, extracts shape features through the Hu invariant moment descriptor, and calculates terrain features such as elevation standard deviation and surface roughness based on digital elevation model data.
[0049] The classification and recognition unit adopts a one-to-one multi-class classification strategy. By constructing multiple binary classifiers, it achieves accurate identification of three types of geological hazards: landslides, collapses, and debris flows, and outputs the category probability distribution map for each pixel.
[0050] When the classification and recognition unit uses a support vector machine for classification, its core algorithm lies in finding an optimal hyperplane that maximizes the margin between the two classes of samples. For the linearly separable case, this hyperplane is determined by the equation w. T Let x + b = 0 be a hyperplane, where w is the normal vector of the hyperplane and b is the offset term. Solving for the optimal hyperplane reduces to solving the following convex quadratic programming problem:
[0051] minimize1 / 2||w|| 2 +C∑ i ξ i
[0052] subject to y i (w T x i +b)≥1-ξ i ,andξ i ≥0, for all i
[0053] Where, x i y is the feature vector of the i-th training sample. i∈{-1,+1} is its corresponding category label, ξ i It is a slack variable used to handle non-linearly separable data, allowing a certain degree of classification error. C>0 is a penalty parameter used to control the tolerance for errors.
[0054] This formula clearly expresses the optimization objective and constraints of the support vector machine, clarifying its core idea of maximizing the classification margin. This indicates that the classifier used by the system has a solid mathematical theoretical foundation, and its decision-making process is explicit and optimizable, rather than a "black box" operation, thus enhancing the reliability of the technical solution.
[0055] The misjudgment elimination unit of the post-processing and optimization module calculates morphological parameters such as the area of the connected region and the ratio of the perimeter area, sets the area threshold range to 100-10000 pixels and the shape index range to 0.2-0.8, and combines geological lithology vector data to eliminate misjudged areas that are obviously inconsistent with geological conditions, such as those located in hard bedrock areas, through spatial overlay analysis.
[0056] The boundary optimization unit adopts a boundary optimization algorithm based on an active contour model. By defining an energy function, the initial contour is driven to evolve into the real disaster boundary. At the same time, morphological closing operations are applied to fill the internal holes and opening operations are used to eliminate small burrs, ultimately generating a smooth and continuous geological disaster boundary vector polygon.
[0057] The thematic map mapping unit of the output and visualization module adopts layered rendering technology. First, a professional geological disaster color scheme based on CMYK color space is established. Then, the semi-transparent disaster zoning layer is superimposed on the Hillshade topographic shading base map through Alpha channel blending technology. At the same time, map finishing elements such as scale bar, legend, and north arrow are automatically added and output as a GeoTIFF format file with a printing precision of 300dpi.
[0058] The report generation unit automatically generates structured reports based on template engine technology. It extracts the geometric center coordinates of each geological hazard polygon as the location of the hazard point, calculates the projected area of the polygon as the hazard area, uses the maximum class probability output by the classifier as the confidence index, and associates environmental parameters such as slope, lithology, and precipitation in the area to form a multi-dimensional attribute table. The final output is a report document in both PDF and Excel formats.
[0059] The system also includes a user interaction and control module. This module uses the Qt framework to develop a graphical user interface, providing a data source selection panel that displays a list of available data and their spatial coverage and time information; a parameter setting panel that provides sliders and drop-down menus for adjusting preprocessing parameters, classifier parameters, and post-processing thresholds; a workflow control panel that provides a one-click processing start button and a progress bar; a results verification panel that provides a split-screen comparison function to enable synchronous browsing of original images, intermediate results, and final results; and a data export panel that supports outputting thematic maps in multiple formats such as JPEG, PNG, and PDF, and outputting data reports in multiple geographic data formats such as CSV, Shapefile, and KML.
[0060] Data connections and calls between the various modules of the system are implemented through application programming interfaces based on the REST architecture. The interfaces use JSON format for data exchange and define standardized data request and response protocols, including a data query interface that returns a list of available data, a data processing interface that receives parameter settings and returns processing progress, and a result acquisition interface that returns a binary format processing result file. Backward compatibility is ensured through interface version control, and the OAuth2.0 protocol is used for interface access permission authentication to ensure system data security and communication reliability.
[0061] Example 2
[0062] The detailed workflow of the geological disaster remote sensing image processing system based on multi-source data described in this invention is as follows.
[0063] The workflow of this invention begins with the multi-source data input module, which automatically receives or is manually imported by the user from optical satellite images, synthetic aperture radar images, lidar point cloud data, and auxiliary geographic information data such as digital elevation models, geological maps, and slope aspect maps from different remote sensing platforms through a standardized interface. The module then performs format parsing, quality checks, and metadata extraction on these heterogeneous data to form a standardized data queue.
[0064] The data then enters the data processing and fusion module. This module initiates a parallel preprocessing pipeline for different data characteristics, performing atmospheric correction on optical images, suppressing speckle noise on radar images, and filtering and interpolating laser point clouds to generate a high-precision digital elevation model. It further calculates derived terrain factors and then uses a registration method based on feature point matching and random sampling consensus algorithm to unify all data into the same geographic coordinate system, achieving sub-pixel level registration accuracy. Finally, depending on the application scenario, a pixel-level, feature-level, or decision-level fusion strategy is selected to generate a high-dimensional fused dataset with a unified geographic reference.
[0065] Subsequently, the fused data is transmitted to the geological disaster intelligent identification module. This module uses a built-in pre-trained deep learning neural network or ensemble learning model to automatically extract multi-level spectral features, texture features, shape features and terrain features from the fused data. Based on these features, it completes pixel-level classification and preliminary identification of geological disaster targets such as landslides, collapses, and debris flows, and outputs preliminary binary identification results with probability values.
[0066] Next, the post-processing and optimization module refines the preliminary identification results. First, mathematical morphological operations (such as opening and closing operations) are applied to remove sporadic noise points and fill holes. Then, morphological parameters such as area threshold and shape index range are set to filter out misjudged areas that do not conform to the morphological characteristics of geological hazards. Furthermore, spatial overlay analysis can be performed in combination with prior knowledge such as geological lithology to further eliminate false alarms. Finally, a boundary optimization algorithm based on an active contour model is used to perform boundary evolution and smoothing on the retained suspected hazard areas to generate an optimized geological hazard distribution vector layer.
[0067] Finally, the results output and visualization module overlays and integrates the optimized vector results with the original imagery or geographic base map, and generates a visualized thematic map that conforms to professional cartographic standards through layered color rendering. At the same time, it automatically extracts information such as the geometric attributes, geographical location, and classification confidence of each disaster polygon, and generates a structured data report by associating environmental parameters. Users can check and export all final results through the interactive interface, thus completing the entire process of integrated processing from multi-source data input to intelligent recognition and professional results output.
[0068] Finally, the following points should be noted: First, in the description of this application, it should be noted that, unless otherwise specified and limited, the terms "installation", "connection", and "linkage" should be interpreted broadly, and can be mechanical or electrical connections, or internal connections between two components, or direct connections. "Up", "down", "left", "right", etc. are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may change.
[0069] Secondly: The accompanying drawings of the embodiments disclosed in this invention only involve the structures involved in the embodiments disclosed in this invention. Other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of this invention can be combined with each other.
[0070] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1.A geological disaster remote sensing image processing system based on multi-source data, characterized in that, The system comprises: a multi-source data input module for receiving remote sensing image data and auxiliary geographic information data from different remote sensing platforms; a data processing and fusion module connected to the multi-source data input module for pre-processing, registration and fusion processing based on feature layers or decision layers of the input data of various types to generate fusion data; a geological disaster intelligent identification module connected to the data processing and fusion module for feature extraction and classification identification of the fusion data by using an embedded machine learning model to output suspected geological disaster area information; a post-processing and optimization module connected to the geological disaster intelligent identification module for morphological processing and false identification elimination of the identified suspected geological disaster area to generate an optimized identification result; a result output and visualization module connected to the post-processing and optimization module for superimposing and integrating the final identification result with original remote sensing images or basic geographic base maps to generate visual thematic maps and data reports. 2.The geological disaster remote sensing image processing system based on multi-source data according to claim 1, wherein, The remote sensing image data received by the multi-source data input module includes two or more of optical satellite images, synthetic aperture radar images and lidar point cloud data, and the auxiliary geographic information data received includes one or more of digital elevation model data, slope and aspect data, geological lithology data and precipitation monitoring data. 3.The geological disaster remote sensing image processing system based on multi-source data according to claim 1, characterized in that, The pre-processing operations performed by the data processing and fusion module include one or more of radiometric calibration, atmospheric correction, geometric correction, image enhancement and noise suppression, the data processing and fusion module adopts an image registration algorithm to unify data of different sources to the same geographic coordinate system, and adopts pixel-level fusion, feature-level fusion or decision-level fusion methods to generate the fusion data. 4.The geological disaster remote sensing image processing system based on multi-source data according to claim 1, characterized in that, The machine learning model embedded in the geological disaster intelligent identification module is a trained deep learning neural network model or an ensemble learning model, which automatically extracts spectral features, texture features, shape features and elevation change features in the fusion data based on the extracted features to complete classification and preliminary identification of landslide, collapse and debris flow geological disaster types. 5.The geological disaster remote sensing image processing system based on multi-source data according to claim 4, characterized in that, The post-processing and optimization module filters the preliminary identification result by applying a set area threshold, shape index range or spatial overlay analysis with known geological conditions to eliminate obvious false non-geological disaster polygons, and uses morphological opening or closing operations to smooth and optimize the boundaries of the retained suspected disaster areas. 6.The geological disaster remote sensing image processing system based on multi-source data according to claim 1, wherein, The visual thematic maps generated by the result output and visualization module simultaneously display the original remote sensing image background and the superimposed geological disaster identification zoning result in layer form, and the data report includes the identified geological disaster point, area, confidence and geographic location information. 7.The geological disaster remote sensing image processing system based on multi-source data according to claim 1, characterized in that, The system further comprises a user interaction and control module which provides a graphical user interface to enable users to select data sources, set data processing parameters, start processing flow, interactively check intermediate and final results, and export the thematic maps and data reports. 8.The geological disaster remote sensing image processing system based on multi-source data according to claim 1, wherein, The data connection and invocation between the modules of the system are realized through predefined application programming interfaces to ensure the data flow and the execution of control logic between the modules.