Geological disaster automatic identification system and method based on multi-source remote sensing data

Abstract

The invention discloses a geological disaster automatic identification system based on multi-source remote sensing data, which comprises a data storage unit, a core control unit and a user interaction terminal, and further comprises a data acquisition module used for acquiring multi-source remote sensing data, and the multi-source remote sensing data comprises optical remote sensing data and synthetic aperture radar data. The invention discloses a geological disaster automatic identification method based on multi-source remote sensing data. The method comprises the following steps: S1, task instruction submission and data acquisition; s2, data preprocessing; s3, feature extraction; s4, performing pattern recognition; and S5, outputting and pushing a result. According to the invention, optical and synthetic aperture radar data and hyperspectral and thermal infrared multi-source remote sensing data are synchronously acquired through the data acquisition module, and physical association and complementary information of each data source is mined in combination with a fusion algorithm based on Bayesian estimation, so that the problem that traditional single-source data is limited by weather and spectral information deficiency is effectively solved; and the identification result is more comprehensive and reliable.

Description

Technical Field

[0001] This invention relates to the field of geological hazard identification technology, and in particular to an automatic geological hazard identification system and method based on multi-source remote sensing data. Background Technology

[0002] Geological disasters such as landslides, debris flows, collapses, and ground subsidence are characterized by their suddenness, destructive power, and wide range of impact, seriously threatening people's lives and property, ecological environment stability, and the safety of major engineering projects. Therefore, achieving efficient and accurate identification and monitoring of geological disasters is one of the core requirements of disaster prevention and mitigation work.

[0003] Traditional geological hazard identification primarily relies on a combination of manual field surveys and single-source remote sensing image interpretation. Manual field surveys require personnel to go to the site and assess potential hazards through on-site measurements and geological logging. This method is not only inefficient and costly, but also limited by terrain conditions and adverse weather, making it difficult to achieve large-scale, routine monitoring coverage, especially in remote mountainous areas and complex terrain where monitoring blind spots are likely to occur. While single-source remote sensing image interpretation can expand the monitoring range, the single data source has significant limitations: optical remote sensing images are easily affected by cloud cover and lighting conditions, making all-weather monitoring difficult; SAR radar data, while capable of all-weather operation, lacks spectral information and has insufficient ability to finely distinguish hazard types; and while lidar point cloud data can provide high-precision three-dimensional terrain information, data acquisition costs are high and coverage cycles are long, making it unsuitable for large-scale dynamic monitoring.

[0004] To address these issues, existing technologies have gradually developed identification schemes based on multi-source remote sensing data fusion and machine learning algorithms. However, these schemes still have several key shortcomings in practical applications: First, their cross-regional generalization ability is limited. Existing models are mostly built based on training data from specific regions. When the geological structure, topography, climate background, and other characteristics of the target region differ significantly from the training dataset, the model's recognition accuracy drops drastically, making it difficult to adapt to the monitoring needs of different regions. Second, the depth of multi-source data fusion is insufficient. Most fusion strategies only remain at the level of simple feature-level splicing, failing to deeply explore the physical correlations and constitutive relationships between different remote sensing data sources such as multispectral and lidar. The fusion effect depends on manually designed rules, failing to fully leverage the complementary advantages of multi-source data. Third, their ability to capture subtle deformations and precursor features is weak. Existing feature extraction methods lack sensitivity to weak features such as precursory subtle deformations and spectral anomalies before geological disasters occur, easily leading to missed reports and failing to meet the needs of early disaster warning. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by proposing an automatic geological disaster identification system and method based on multi-source remote sensing data.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] An automatic geological hazard identification system based on multi-source remote sensing data includes a data storage unit, a core control unit, and a user interaction terminal, and also includes:

[0008] The data acquisition module is used to collect multi-source remote sensing data, including optical remote sensing data, synthetic aperture radar data, hyperspectral remote sensing data and thermal infrared remote sensing data. The data acquisition module establishes communication connections with satellite remote sensing platforms, UAV remote sensing platforms and ground remote sensing equipment through interfaces.

[0009] The data preprocessing module, connected to the core control unit and the data storage unit, is used to perform data format standardization, coordinate registration, data fusion and noise reduction on the acquired multi-source remote sensing data.

[0010] The feature extraction module, connected to the core control unit and the data storage unit, is used to extract geological hazard features from preprocessed multi-source remote sensing data. The geological hazard features include texture features, spectral features, morphological features, and temporal variation features.

[0011] The pattern recognition module, connected to the core control unit and the data storage unit, is used to classify and identify the extracted geological hazard features. The pattern recognition module has an improved ResNet-50 deep learning model built in.

[0012] The recognition result output module is connected to the pattern recognition module, the core control unit, and the data storage unit. It is used to output the recognition results in the form of visual charts and text reports. The recognition results include the geological disaster type, geographical location, probability of occurrence, and risk level.

[0013] The data storage unit is used to store the collected raw multi-source remote sensing data, preprocessed intermediate data, extracted feature data, trained pattern recognition model parameters, and final recognition results, and supports data indexing by timestamp and regional location; the core control unit is used to schedule the runtime sequence of each module, control the data transmission path between modules, monitor the running status of each module in real time, trigger alarms and record abnormal logs to the data storage unit when a module experiences an abnormality; the user interaction terminal is used by users to submit data collection task instructions, query historical recognition data, and receive real-time recognition result pushes, and also has user permission management functions, with different permissions corresponding to different data operation permissions for different users.

[0014] Preferably, the data acquisition module is connected to the core control unit and the data storage unit, and transmits the collected raw data to the data storage unit under the scheduling of the core control unit.

[0015] Preferably, the coordinate registration in the data preprocessing module adopts a feature point matching method based on the SIFT algorithm to achieve spatial alignment of different data sources, and the feature point descriptor similarity calculation adopts the Euclidean distance formula:

[0016]

[0017] in , These are 128-dimensional descriptor components for two feature points. The spatial distance between two feature points in 128-dimensional descriptors.

[0018] Preferably, the data fusion in the data preprocessing module adopts a fusion algorithm based on Bayesian estimation, the core formula of which is:

[0019]

[0020] in For posterior probability, Let be the likelihood function. This represents the prior probability.

[0021] Preferably, the noise reduction process in the data preprocessing module employs a wavelet threshold noise reduction algorithm, and the threshold calculation formula is as follows: ,in For wavelet threshold, The standard deviation of noise. This represents the data length.

[0022] Preferably, the texture features in the feature extraction module are extracted using a gray-level co-occurrence matrix, and the contrast calculation formula is as follows:

[0023]

[0024] in For image contrast, For elements of the gray-level co-occurrence matrix, These represent the gray values ​​of two pixels in the image that are at a fixed distance from each other; spectral features are extracted using continuous wavelet transform, and morphological features are extracted using morphological operators; temporal variation features are extracted by analyzing preprocessed remote sensing data from different periods in the same monitoring area.

[0025] Preferably, the improved ResNet-50 deep learning model has the following cross-entropy loss function formula:

[0026]

[0027] in, For the sample size, For the number of categories, For the sample Category The true label, To predict probabilities, the improved ResNet-50 model optimizes feature weight allocation by introducing an attention mechanism. The formula for calculating attention weights is as follows: ,in As input features, , Let be the weight matrix. It is the ReLU activation function. This is the Sigmoid function.

[0028] An automatic geological hazard identification method based on multi-source remote sensing data, employing an automatic geological hazard identification system based on multi-source remote sensing data, includes the following steps:

[0029] S1. Task instruction submission and data acquisition: The user submits the monitoring area and task parameter instructions through the user interaction terminal. After receiving the instructions, the core control unit schedules the data acquisition module to collect optical remote sensing data, synthetic aperture radar data, hyperspectral remote sensing data and thermal infrared remote sensing data. After the data acquisition is completed, the raw data is stored in the data storage unit.

[0030] S2. Data preprocessing: The core control unit calls the data preprocessing module to read the raw data from the data storage unit. First, the data of different formats are converted into a unified GeoTIFF format. Then, coordinate registration is performed by the SIFT algorithm. The Bayesian estimation fusion algorithm combined with the formula is used to achieve multi-source data fusion. Finally, the wavelet threshold noise reduction algorithm combined with the formula is used to remove noise from the fused data. The preprocessed intermediate data is stored in the data storage unit.

[0031] S3. Feature extraction: The core control unit calls the feature extraction module to read preprocessed data from the data storage unit and extract multi-dimensional geological disaster features. Among them, texture features are extracted using gray-level co-occurrence matrix combined with formula, spectral features are extracted using continuous wavelet transform, morphological features are extracted using morphological dilation and erosion operators, and temporal variation features are extracted using temporal data. The extracted feature data is stored in the data storage unit.

[0032] S4. Pattern Recognition: The core control unit calls the pattern recognition module to read feature data from the data storage unit and input it into the improved ResNet-50 model. The model strengthens the weights of key features through an attention mechanism combined with a formula, calculates the loss value using the cross-entropy loss function formula, and employs the Adam algorithm combined with the formula. Optimize parameters, output classification results and occurrence probabilities, and store the identified data in the data storage unit;

[0033] S5. Results Output and Push: The core control unit calls the recognition result output module to read the recognition data from the data storage unit. After determining the risk level, it generates a visual chart and a text report. On the one hand, the results are pushed to the user through the user interaction terminal, and on the other hand, the results are stored in the data storage unit for the user to query later.

[0034] Preferably, the risk level classification standard in step S5 is as follows: when the probability of disaster occurrence is ≥85%, the risk level is Level 1; when 60%≤probability of disaster occurrence is <85%, the risk level is Level 2; when 30%≤probability of disaster occurrence is <60%, the risk level is Level 3; and when the probability of disaster occurrence is <30%, the risk level is Level 4.

[0035] The present invention has the following beneficial effects:

[0036] 1. This invention simultaneously acquires optical, synthetic aperture radar, hyperspectral, and thermal infrared multi-source remote sensing data through a data acquisition module. It combines a Bayesian estimation-based fusion algorithm to mine the physical correlation and complementary information of each data source. Optical data ensures the accuracy of ground feature identification, synthetic aperture radar data enables all-weather monitoring, hyperspectral data enhances the differentiation of ground feature components, and thermal infrared data captures temperature anomalies that are precursors to disasters. This effectively solves the problem of traditional single-source data being limited by weather and lack of spectral information, making the identification results more comprehensive and reliable.

[0037] 2. This invention introduces an attention mechanism module on the basis of the ResNet-50 model, which strengthens key features such as landslide morphology and debris flow spectrum through adaptive weight allocation and suppresses irrelevant interference. It combines texture, spectrum, morphology and time series multi-dimensional feature extraction to build a comprehensive feature system, which solves the defects of existing models in capturing weak features and weak cross-regional generalization ability. After training with samples from multiple regions, the model can maintain high recognition accuracy in different geological structures and meet the actual monitoring needs.

[0038] 3. This invention constructs a fully automated processing and closed-loop management system. From data acquisition, preprocessing, feature extraction to pattern recognition and result output, all processes are uniformly scheduled by the core control unit. Combined with standardized data format conversion, accurate coordinate registration, and efficient noise reduction, it significantly reduces the cost of manual intervention and achieves large-scale, routine geological disaster monitoring coverage. At the same time, the recognition results are presented intuitively in the form of visual charts and text reports, including disaster type, geographical location, probability of occurrence, and four-level risk classification. With the real-time push and historical data indexing functions of the user interactive terminal, it provides timely and accurate technical support for disaster prevention and mitigation decision-making, effectively improving the efficiency of disaster early warning response and the scientific nature of emergency response. Attached Figure Description

[0039] Figure 1 This is a system block diagram of an automatic geological hazard identification system based on multi-source remote sensing data proposed in this invention;

[0040] Figure 2 This is a flowchart of an automatic geological disaster identification method based on multi-source remote sensing data proposed in this invention. Detailed Implementation

[0041] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0042] Reference Figure 1 An automatic geological hazard identification system based on multi-source remote sensing data includes a data storage unit, a core control unit, and a user interaction terminal, and also includes:

[0043] The data acquisition module is connected to the core control unit and the data storage unit. Under the scheduling of the core control unit, the acquired raw data is transmitted to the data storage unit for acquiring multi-source remote sensing data, including optical remote sensing data, synthetic aperture radar data, hyperspectral remote sensing data and thermal infrared remote sensing data. The data acquisition module establishes communication connections with satellite remote sensing platforms, UAV remote sensing platforms and ground remote sensing equipment through interfaces.

[0044] The data preprocessing module, connected to the core control unit and the data storage unit, is used to perform data format standardization, coordinate registration, data fusion, and noise reduction on the collected multi-source remote sensing data.

[0045] The implementation method for data format standardization is as follows: using a geographic data abstraction library, raw data in formats such as TIFF and HDF are converted into TIFF format with geospatial information. The converted data is then embedded with WGS84 geographic coordinates to ensure spatial reference consistency among multi-source data.

[0046] Coordinate registration employs a feature point matching method based on the SIFT algorithm to align spatial locations from different data sources. Feature point descriptor similarity is calculated using the Euclidean distance formula.

[0047]

[0048] in , These are 128-dimensional descriptor components for two feature points. The spatial distance between two feature points in 128-dimensional descriptors.

[0049] The data fusion employs a Bayesian estimation-based fusion algorithm, the core formula of which is:

[0050]

[0051] in For posterior probability, Let be the likelihood function. This represents the prior probability.

[0052] The noise reduction process uses a wavelet thresholding algorithm, and the threshold calculation formula is as follows: ,in For wavelet threshold, The standard deviation of noise. This represents the data length.

[0053] The feature extraction module, connected to the core control unit and the data storage unit, is used to extract geological hazard features from preprocessed multi-source remote sensing data. The geological hazard features include texture features, spectral features, morphological features, and temporal variation features.

[0054] Texture features are extracted using the gray-level co-occurrence matrix, and the contrast calculation formula is as follows:

[0055]

[0056] in For image contrast, For elements of the gray-level co-occurrence matrix, These represent the gray values ​​of two pixels in the image that are at a fixed distance from each other; spectral features are extracted using continuous wavelet transform, and morphological features are extracted using morphological operators; temporal variation features are extracted by analyzing preprocessed remote sensing data from different periods in the same monitoring area.

[0057] The pattern recognition module, connected to the core control unit and data storage unit, is used to classify and identify the extracted geological hazard features. The pattern recognition module incorporates an improved ResNet-50 deep learning model; the cross-entropy loss function formula for the improved ResNet-50 deep learning model is as follows:

[0058]

[0059] in, For the sample size, For the number of categories, For the sample Category The true label, To predict probabilities, the improved ResNet-50 model optimizes feature weight allocation by introducing an attention mechanism. The formula for calculating attention weights is as follows: ,in As input features, , Let be the weight matrix. It is the ReLU activation function. This is the Sigmoid function.

[0060] The recognition result output module is connected to the pattern recognition module, the core control unit, and the data storage unit. It is used to output the recognition results in the form of visual charts and text reports. The recognition results include the geological disaster type, geographical location, probability of occurrence, and risk level.

[0061] The data storage unit is used to store the collected raw multi-source remote sensing data, preprocessed intermediate data, extracted feature data, trained pattern recognition model parameters, and final recognition results, and supports data indexing by timestamp and regional location; the core control unit is used to schedule the runtime sequence of each module, control the data transmission path between modules, monitor the running status of each module in real time, trigger alarms and record abnormal logs to the data storage unit when a module experiences an abnormality; the user interaction terminal is used by users to submit data collection task instructions, query historical recognition data, and receive real-time recognition result pushes, and also has user permission management functions, with different permissions corresponding to different data operation permissions for different users.

[0062] Reference Figure 2 An automatic geological hazard identification method based on multi-source remote sensing data, employing the aforementioned automatic geological hazard identification system based on multi-source remote sensing data, includes the following steps:

[0063] S1. Task instruction submission and data acquisition: The user submits the monitoring area and task parameter instructions through the user interaction terminal. After receiving the instructions, the core control unit schedules the data acquisition module to collect optical remote sensing data, synthetic aperture radar data, hyperspectral remote sensing data and thermal infrared remote sensing data. After the data acquisition is completed, the raw data is stored in the data storage unit.

[0064] S2. Data preprocessing: The core control unit calls the data preprocessing module to read the raw data from the data storage unit. First, the data of different formats are converted into a unified GeoTIFF format. Then, coordinate registration is performed by the SIFT algorithm. The Bayesian estimation fusion algorithm combined with the formula is used to achieve multi-source data fusion. Finally, the wavelet threshold noise reduction algorithm combined with the formula is used to remove noise from the fused data. The preprocessed intermediate data is stored in the data storage unit.

[0065] S3. Feature extraction: The core control unit calls the feature extraction module to read preprocessed data from the data storage unit and extract multi-dimensional geological disaster features. Among them, texture features are extracted using gray-level co-occurrence matrix combined with formula, spectral features are extracted using continuous wavelet transform, morphological features are extracted using morphological dilation and erosion operators, and temporal variation features are extracted using temporal data. The extracted feature data is stored in the data storage unit.

[0066] S4. Pattern Recognition: The core control unit calls the pattern recognition module to read feature data from the data storage unit and input it into the improved ResNet-50 model. The model strengthens the weights of key features through an attention mechanism combined with a formula, calculates the loss value using the cross-entropy loss function formula, and employs the Adam algorithm combined with the formula. Optimize parameters, output classification results and occurrence probabilities, and store the identified data in the data storage unit;

[0067] S5. Results Output and Push: The core control unit calls the recognition result output module to read the recognition data from the data storage unit. After determining the risk level, it generates a visual chart and a text report. On the one hand, the results are pushed to the user through the user interaction terminal, and on the other hand, the results are stored in the data storage unit for the user to query later.

[0068] The risk level classification criteria in step S5 are as follows: when the probability of disaster occurrence is ≥85%, the risk level is Level 1; when 60%≤probability of disaster occurrence is <85%, the risk level is Level 2; when 30%≤probability of disaster occurrence is <60%, the risk level is Level 3; and when the probability of disaster occurrence is <30%, the risk level is Level 4.

[0069] It should also be noted that the Adam algorithm in step S4 is combined with the formula middle, For model parameters in the first... Updated values ​​after the next iteration, such as weights and biases; For model parameters in the first... The value after -1 iterations; The learning rate controls the step size of parameter updates and needs to be preset, typically with values ​​such as 0.001 or 0.0001. The second-order momentum term is the exponential moving average of the squared gradient of the parameters, used to adaptively adjust the learning rate. To be the minimum value, it is usually taken as To prevent the denominator A value of 0 would result in a calculation error; The first-order momentum term is the exponential moving average of the parameter gradient, used to alleviate gradient oscillations and accelerate convergence.

[0070] Example 1

[0071] An automatic geological hazard identification system based on multi-source remote sensing data includes a data acquisition module, a data preprocessing module, a feature extraction module, a pattern recognition module, and an identification result output module.

[0072] The data acquisition module uses the Gaofen-6 satellite to acquire optical remote sensing data with a spatial resolution of 10m, the Sentinel-1 satellite to acquire SAR data in the C-band with VV / VH polarization, the Gaofen-5 satellite to acquire hyperspectral remote sensing data with 330 bands and a spectral resolution of 5nm, and a ground-based thermal infrared sensor to acquire thermal infrared remote sensing data with a temperature resolution of 0.1℃. Real-time data acquisition is achieved through an API interface.

[0073] The data preprocessing module uses Python to write the processing program. Data format standardization uses a geographic data abstraction library to convert the data into TIFF format with geospatial information. Coordinate registration uses the SIFT algorithm implemented in the OpenCV library, and feature point matching uses the K-nearest neighbor algorithm (k=2). Data fusion uses a Bayesian estimation fusion algorithm, implemented in Matlab. Noise reduction uses a wavelet thresholding algorithm with a wavelet basis of db4, a decomposition level of 3, and a threshold set to 1.5 times the noise standard deviation.

[0074] The feature extraction module is implemented using Python's Scikit-learn library. The gray-level co-occurrence matrix parameters are set as follows: distance 1, angles 0°, 45°, 90°, and 135°. The mean of texture features in the four directions is calculated as the final texture features, where contrast is determined by the formula:

[0075]

[0076] The calculations were performed using the PyWavelets library for continuous wavelet transform, with the wavelet basis being morlet and the decomposition scale being 1-5. Morphological operations were performed using the dilation and erosion functions from the OpenCV library, with the structuring element being a 3×3 rectangle.

[0077] The pattern recognition module builds an improved ResNet-50 model based on the TensorFlow framework, while the attention mechanism module combines channel attention and spatial attention, with attention weights expressed by a formula. The model training unit is trained using a labeled sample set, including 1000 samples each of landslides, debris flows, and collapses, and 2000 samples of normal areas. The batch size is 32, the initial learning rate is 0.001, and the Adam optimization algorithm is used in conjunction with the formula:

[0078]

[0079] Adjust the parameters, train for 100 epochs, and store the trained model parameters in a MySQL database.

[0080] The identification results output module uses ArcGIS to create a visual map, with different colors used to label different disaster types and different brightness levels to represent risk levels. The text report is in Word format and includes identification result statistics, risk analysis, and prevention and control recommendations. Users access the user interaction terminal via a web page or mobile app, submit tasks, query data, and receive result push notifications after logging in. Administrators configure user permissions through the backend management interface.

[0081] Example 2

[0082] An automatic geological hazard identification method based on multi-source remote sensing data, using the system of Example 1, includes the following steps:

[0083] S1. Task Command Submission and Data Acquisition: Users log in to the system via a mobile app, submit monitoring tasks, set the monitoring area to a designated mountainous area, the data collection period to January-December 2024, and select Gaofen-6, Sentinel-1, Gaofen-5, and a ground-based thermal infrared sensor as the data source. After receiving the command, the core control unit schedules the data acquisition module to collect the corresponding data and store it in the raw data table of the MySQL database.

[0084] S2. Data Preprocessing: The data preprocessing module reads raw data from the database and converts it to TIFF format with WGS84 coordinates using a geographic data abstraction library; it then implements the SIFT algorithm using OpenCV to extract feature points and applies the Euclidean distance formula.

[0085]

[0086] Registration is completed after matching and removing erroneous matches; a normal distribution probability model is established for each data source, the likelihood function is calculated, and the prior probability is determined by combining historical data and substituting it into Bayes' theorem.

[0087]

[0088] Data fusion; using db4 wavelet 3-level decomposition to calculate noise standard deviation. Through formula Calculate the threshold and perform soft thresholding. The reconstructed data is then stored in a preprocessed data table.

[0089] S3. Feature Extraction: Extract multi-dimensional features. Texture Features: Convert the fused data into a grayscale image, construct a grayscale co-occurrence matrix with a distance of 1 and angles of 0° / 45° / 90° / 135°, and then use the formula...

[0090]

[0091] Calculate the mean values ​​of indicators such as contrast; Spectral features: Perform Morlet wavelet 1-5 scale decomposition on hyperspectral data and extract wavelet coefficients; Morphological features: Perform expansion and erosion processing on 3×3 rectangular structural elements to calculate the area, perimeter, and shape factor of the disaster area; Temporal variation features: Calculate the changes in grayscale and spectral values ​​at the same location from January to December, analyze the trend of change, and store the feature data in the feature data table.

[0092] S4. Pattern recognition: After feature normalization, the data is input into an improved ResNet-50 model. The model extracts deep features through convolutional and pooling layers, and the attention mechanism module uses a formula... Adjust feature weights, and output classification probabilities from fully connected layers. Model training: Use labeled sample sets (1000 sets each of landslides, debris flows, and collapses, and 2000 sets of normal regions) for training, batch size 32, initial learning rate 0.001, and employ the Adam optimization algorithm combined with the formula:

[0093]

[0094] After adjusting the parameters and training for 80 epochs, the accuracy on the validation set no longer improves, so training is stopped; the model parameters are stored in the model parameter table, and the inference results are stored in the recognition result table.

[0095] S5. Result Output and Push: The result output module reads data from the recognition result table, determines the risk level, and generates a visual map and a text report. The visual map displays the types and specific numbers of geological hazards in the area; the text report summarizes the number and location of hazards, analyzes the causes of risk, and proposes key monitoring areas and prevention measures. The results are then pushed to the user's app, where the user receives a pop-up notification to view the report, which is then stored in the database.

[0096] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. An automatic geological hazard identification system based on multi-source remote sensing data, comprising a data storage unit, a core control unit, and a user interaction terminal, characterized in that, Also includes: The data acquisition module is used to collect multi-source remote sensing data, including optical remote sensing data, synthetic aperture radar data, hyperspectral remote sensing data and thermal infrared remote sensing data. The data acquisition module establishes communication connections with satellite remote sensing platforms, UAV remote sensing platforms and ground remote sensing equipment through interfaces. The data preprocessing module, connected to the core control unit and the data storage unit, is used to perform data format standardization, coordinate registration, data fusion and noise reduction on the acquired multi-source remote sensing data. The feature extraction module, connected to the core control unit and the data storage unit, is used to extract geological hazard features from preprocessed multi-source remote sensing data. The geological hazard features include texture features, spectral features, morphological features, and temporal variation features. The pattern recognition module, connected to the core control unit and the data storage unit, is used to classify and identify the extracted geological hazard features. The pattern recognition module has an improved ResNet-50 deep learning model built in. The recognition result output module is connected to the pattern recognition module, the core control unit, and the data storage unit. It is used to output the recognition results in the form of visual charts and text reports. The recognition results include the geological disaster type, geographical location, probability of occurrence, and risk level. The data storage unit is used to store the collected raw multi-source remote sensing data, preprocessed intermediate data, extracted feature data, trained pattern recognition model parameters, and final recognition results, and supports data indexing by timestamp and regional location; the core control unit is used to schedule the runtime sequence of each module, control the data transmission path between modules, monitor the running status of each module in real time, trigger alarms and record abnormal logs to the data storage unit when a module experiences an abnormality; the user interaction terminal is used by users to submit data collection task instructions, query historical recognition data, and receive real-time recognition result pushes, and also has user permission management functions, with different permissions corresponding to different data operation permissions for different users.

2. The automatic geological hazard identification system based on multi-source remote sensing data according to claim 1, characterized in that, The data acquisition module is connected to the core control unit and the data storage unit, and transmits the collected raw data to the data storage unit under the scheduling of the core control unit.

3. The automatic geological hazard identification system based on multi-source remote sensing data according to claim 1, characterized in that, The coordinate registration in the data preprocessing module adopts a feature point matching method based on the SIFT algorithm to achieve spatial alignment of different data sources. The similarity calculation of feature point descriptors uses the Euclidean distance formula. in , These are 128-dimensional descriptor components for two feature points. The spatial distance between two feature points in 128-dimensional descriptors.

4. The automatic geological hazard identification system based on multi-source remote sensing data according to claim 1, characterized in that, The data fusion in the data preprocessing module adopts a Bayesian estimation-based fusion algorithm, the core formula of which is: in For posterior probability, Let be the likelihood function. This represents the prior probability.

5. The automatic geological hazard identification system based on multi-source remote sensing data according to claim 1, characterized in that, The noise reduction process in the data preprocessing module uses a wavelet threshold noise reduction algorithm, and the threshold calculation formula is as follows: ,in For wavelet threshold, The standard deviation of noise. This represents the data length.

6. The automatic geological hazard identification system based on multi-source remote sensing data according to claim 1, characterized in that, The texture features in the feature extraction module are extracted using the gray-level co-occurrence matrix, and the contrast calculation formula is as follows: in For image contrast, For elements of the gray-level co-occurrence matrix, These represent the gray values ​​of two pixels in the image that are at a fixed distance from each other. Spectral features were extracted using continuous wavelet transform, and morphological features were extracted using morphological operators. Temporal variation features were extracted by analyzing preprocessed remote sensing data from different periods in the same monitoring area.

7. The automatic geological hazard identification system based on multi-source remote sensing data according to claim 1, characterized in that, The improved ResNet-50 deep learning model has the following cross-entropy loss function formula: in, For the sample size, For the number of categories, For the sample Category The true label, To predict probabilities, the improved ResNet-50 model optimizes feature weight allocation by introducing an attention mechanism. The formula for calculating attention weights is as follows: ,in As input features, , Let be the weight matrix. It is the ReLU activation function. This is the Sigmoid function.

8. A method for automatic identification of geological hazards based on multi-source remote sensing data, employing an automatic geological hazard identification system based on multi-source remote sensing data as described in any one of claims 1-7, characterized in that, Includes the following steps: S1. Task instruction submission and data acquisition: The user submits the monitoring area and task parameter instructions through the user interaction terminal. After receiving the instructions, the core control unit schedules the data acquisition module to collect optical remote sensing data, synthetic aperture radar data, hyperspectral remote sensing data and thermal infrared remote sensing data. After the data acquisition is completed, the raw data is stored in the data storage unit. S2. Data preprocessing: The core control unit calls the data preprocessing module to read the raw data from the data storage unit. First, the data of different formats are converted into a unified GeoTIFF format. Then, coordinate registration is performed by the SIFT algorithm. The Bayesian estimation fusion algorithm combined with the formula is used to achieve multi-source data fusion. Finally, the wavelet threshold noise reduction algorithm combined with the formula is used to remove noise from the fused data. The preprocessed intermediate data is stored in the data storage unit. S3. Feature extraction: The core control unit calls the feature extraction module to read preprocessed data from the data storage unit and extract multi-dimensional geological disaster features. Among them, texture features are extracted using gray-level co-occurrence matrix combined with formula, spectral features are extracted using continuous wavelet transform, morphological features are extracted using morphological dilation and erosion operators, and temporal variation features are extracted using temporal data. The extracted feature data is stored in the data storage unit. S4. Pattern Recognition: The core control unit calls the pattern recognition module to read feature data from the data storage unit and input it into the improved ResNet-50 model. The model strengthens the weights of key features through an attention mechanism combined with a formula, calculates the loss value using the cross-entropy loss function formula, and employs the Adam algorithm combined with the formula... Optimize parameters, output classification results and occurrence probabilities, and store the identified data in the data storage unit; S5. Results Output and Push: The core control unit calls the recognition result output module to read the recognition data from the data storage unit. After determining the risk level, it generates a visual chart and a text report. On the one hand, the results are pushed to the user through the user interaction terminal, and on the other hand, the results are stored in the data storage unit for the user to query later.

9. The method for automatic identification of geological hazards based on multi-source remote sensing data according to claim 8, characterized in that, The risk level classification criteria in step S5 are as follows: when the probability of disaster occurrence is ≥85%, the risk level is Level 1; when 60%≤probability of disaster occurrence is <85%, the risk level is Level 2; when 30%≤probability of disaster occurrence is <60%, the risk level is Level 3; and when the probability of disaster occurrence is <30%, the risk level is Level 4.

Images