Field geological exploration monitoring system based on remote sensing data

Abstract

The invention relates to the technical field of geological disaster prediction, in particular to a field geological exploration monitoring system based on remote sensing data, which comprises a data acquisition module, a data preprocessing module, a multi-source data fusion module, a deep learning model module and an early warning module. The data acquisition module, the data preprocessing module, the multi-source data fusion module, the deep learning model module and the early warning module are connected in sequence. Through sequential connection of the modules, the system realizes closed loop of data from acquisition, analysis and early warning, and ensures integrity, real-time performance and accuracy of the data. The data preprocessing module eliminates the interference of source data while ensuring the data quality, and the fusion module improves the synergistic effect of multi-source data through an intelligent algorithm. The high-precision prediction of the deep learning model module ensures the timeliness and effectiveness of disaster early warning. The early warning module responds in time to help related departments to do disaster prevention and reduction work, and risks caused by geological disasters are effectively reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of geological disaster prediction, and in particular to a field geological survey and monitoring system based on remote sensing data. Background Art

[0002] With the rapid development of geological disaster monitoring and early warning technology, various monitoring and notification units have been widely used around the world and have improved disaster response capabilities to a certain extent. However, although these systems have basic early warning functions, they still face many challenges and problems in actual use. A major problem is the inadequacy of existing systems in multi-source data fusion and comprehensive utilization, which directly affects the accuracy and timeliness of early warning and cannot fully meet the needs of modern geological disaster prevention. The current geological disaster monitoring system mainly relies on remote sensing data to obtain geological information. Although remote sensing technology can provide high-resolution images and relatively comprehensive surface change data, it can only provide one data source and cannot fully cover the entire process of geological disasters.

[0003] In addition, the integration and utilization of other types of multi-source data (such as ground sensor data, meteorological data, historical disaster data, etc.) are relatively weak. These data are often of extremely high value in complex geological environments in the wild, but due to technical limitations, it is impossible to achieve effective data integration. Especially when factors such as meteorological changes, soil moisture, and ground vibrations have an increasing impact on geological disasters, relying solely on remote sensing data for early warning appears to be single and not comprehensive enough.

[0004] Therefore, the early warning capability of the existing monitoring system is still limited and cannot fully respond to sudden complex geological disaster events. Summary of the invention

[0005] Based on the above purpose, the present invention provides a field geological exploration and monitoring system based on remote sensing data, including a data acquisition module, a data preprocessing module, a multi-source data fusion module, a deep learning model module and an early warning module; The data acquisition module, data preprocessing module, multi-source data fusion module, deep learning model module and early warning module are connected in sequence.

[0006] Preferably, the data acquisition module includes a remote sensing data acquisition unit, a ground sensor data acquisition unit and a meteorological data acquisition unit; The remote sensing data acquisition unit obtains high-resolution image data of geological disaster areas through satellite remote sensing technology, including visual interpretation data, spectral data and terrain data; The ground sensor data acquisition unit collects ground vibration data, displacement data, temperature data and humidity data in real time through various ground sensors installed in the geological disaster area, including vibration sensors, displacement sensors, temperature sensors, humidity sensors, etc.; the meteorological data acquisition unit obtains meteorological data in the area through meteorological stations or meteorological satellites, specifically including air temperature data, air humidity data, rainfall and wind speed in the area.

[0007] Preferably, the data preprocessing module preprocesses the collected multi-source data, including data cleaning, correction, alignment, format conversion and normalization; Multi-source data include ground vibration data, displacement data, temperature data, humidity data, air temperature data, air humidity data, rainfall, and wind speed in the area; The preprocessing module removes invalid data and outliers through a data cleaning algorithm, corrects data errors through a data correction algorithm, ensures the temporal and spatial consistency of different data sources through a data alignment algorithm, converts data in different formats into a unified format through a format conversion algorithm, and normalizes the eigenvalues ​​to between 0 and 1 through a normalization algorithm.

[0008] Preferably, the multi-source data fusion module includes a feature extraction unit, a data alignment and normalization unit and an adaptive data fusion unit; The feature extraction unit performs feature extraction on the preprocessed remote sensing data, ground sensor data and meteorological data, including edge detection, texture analysis and spectral feature extraction of remote sensing images; vibration spectrum analysis, displacement change rate calculation, temperature and humidity time series feature analysis of ground sensors; and feature extraction of temperature change rate, humidity change rate and rainfall trend of meteorological data.

[0009] Preferably, the data alignment and normalization unit aligns the extracted features to ensure the temporal and spatial consistency of different source data, and then performs normalization processing to normalize the feature values ​​to between 0 and 1; The adaptive data fusion unit adopts an adaptive data fusion algorithm to fuse the normalized features, inputs the feature vectors of remote sensing data, ground sensor data and meteorological data into the adaptive data fusion algorithm, and dynamically adjusts the weights of different data sources.

[0010] Preferably, the deep learning model module includes a data preparation unit, a model selection unit, a model training unit and a model evaluation unit; The data preparation unit takes multi-source data of historical geological disaster events as the input of the model, takes the type and probability of occurrence of geological disasters as the output, and constructs training data sets and test data sets; The model selection unit adopts a Transformer-based multimodal deep learning model, and the model training unit uses a training dataset to train the model, including cross-validation and hyperparameter optimization. During the training process, loss functions and optimization algorithms are used to optimize model parameters.

[0011] Preferably, the loss function and optimization algorithm are used to optimize the model parameters, and the specific training steps are as follows: S1: Input the multi-source data feature vectors into the Transformer model, with each feature vector serving as a token of the input; S2: Through the multi-head self-attention mechanism, the correlation between each token and other tokens is calculated to generate an attention matrix; S3: Perform feedforward neural network transformation on the feature vector of each token to enhance the nonlinear expression ability of the feature; S4: Use the cross entropy loss function to calculate the difference between the model prediction value and the true value; S5: Adopt Adam optimization algorithm to optimize model parameters through gradient descent method; The model evaluation unit uses the test data set to evaluate the trained model and evaluates the performance of the model through accuracy, recall, and F1 score indicators.

[0012] Preferably, the early warning module includes a prediction result generation unit and an early warning signal generation unit, and the prediction result generation unit inputs the multi-source data collected in real time into the trained deep learning model to generate the type and probability prediction results of the occurrence of geological disasters; The warning signal generation unit generates geological disaster warning information according to the prediction results, and sends out warning signals through the notification unit; Notification units include SMS notification, email notification, sound and light alarm and automatic notification to emergency management departments. The content of the early warning signal includes the type of disaster, probability of occurrence, expected time of occurrence, affected area and recommended response measures.

[0013] Preferably, the mathematical expression of the vibration spectrum analysis is: ; in, It is a vibration signal. is the frequency, j is an imaginary unit, specifically .

[0014] Preferably, the mathematical expression for calculating the displacement change rate is: ;in, is the change in displacement, is the change in time, and are the displacement values ​​respectively.

[0015] Beneficial effects of the present invention: 1. The data preprocessing module eliminates the interference of source data while ensuring data quality, and the fusion module improves the synergy of multi-source data through intelligent algorithms. The high-precision prediction of the deep learning model module ensures the timeliness and effectiveness of disaster warning.

[0016] 2. Remote sensing data provides a wide range of surface conditions, dynamic changes captured by ground sensors help further refine predictions, and meteorological data provides the climate background in which disasters may intensify. By combining these three types of data, the system can conduct a full range of monitoring of geological disasters, covering a wide area while providing fine-grained local monitoring.

[0017] 3. The data preprocessing module provides data quality assurance for the entire geological exploration and monitoring system, improves the stability and accuracy of the system, and helps relevant departments take timely response measures to reduce the loss of personnel and property caused by geological disasters.

[0018] 4. The close cooperation of feature extraction, data alignment and normalization, and adaptive data fusion has greatly improved the data processing capability and early warning accuracy of the geological exploration and monitoring system, providing a strong guarantee for geological disaster prevention and resource exploration.

[0019] 5. The organic combination of data alignment and normalization as well as adaptive data fusion improves the system's ability to integrate multi-source data, ensures the accuracy and real-time nature of geological monitoring, helps to effectively predict and prevent geological disasters, and supports scientific decision-making in exploration activities. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only for the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0021] Figure 1 is a system block diagram of the system of the present invention; Figure 2 A flowchart of the steps of training a model for the system of the present invention; Figure 3 It is a schematic block diagram of the early warning module in the system of the present invention. DETAILED DESCRIPTION

[0022] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. At the same time, it is explained here that in order to make the embodiments more detailed, the following embodiments are the best and preferred embodiments, and those skilled in the art may also adopt other alternatives to implement some known technologies; and the accompanying drawings are only for more specific description of the embodiments, and are not intended to specifically limit the present invention.

[0023] See also Figure 1-Figure 3 , an embodiment of the present invention provides a field geological exploration and monitoring system based on remote sensing data, including a data acquisition module, a data preprocessing module, a multi-source data fusion module, a deep learning model module and an early warning module; The data acquisition module, data preprocessing module, multi-source data fusion module, deep learning model module and early warning module are connected in sequence.

[0024] In a possible implementation, first, the data acquisition module collects multiple data through multiple subunits, including remote sensing data, ground sensor data, and meteorological data. Remote sensing data is obtained through satellites and drones to provide information such as terrain, soil, and vegetation coverage over a large area; ground sensors obtain real-time displacement, temperature, humidity, and other data through sensors deployed in the field; meteorological data is obtained through weather stations or satellites, including information such as rainfall and wind speed. These data serve as the basis for subsequent analysis.

[0025] Furthermore, after data collection, it needs to be cleaned, corrected, aligned, and normalized through the data preprocessing module. Data cleaning removes noise and outliers to ensure data quality; the correction process corrects equipment errors; the alignment operation ensures the consistency of data from different sources in time and space; and normalization maps the data to a unified scale so that various data sources can work together on the model to avoid errors caused by dimensional differences.

[0026] Furthermore, after data preprocessing, it enters the multi-source data fusion module. In this module, features from remote sensing, sensors and meteorology will be extracted, and the weights of different data sources will be adaptively adjusted according to the algorithm. Adaptive fusion can dynamically adjust the weight of each data source in the comprehensive decision-making according to the real-time changes of data and environmental influences, so that the fused features are more representative and accurate.

[0027] Furthermore, the fused features enter the deep learning model module. The model is trained based on historical data and can identify the potential risks of geological disasters. The model uses a Transformer-based multimodal deep learning method to find potential associations in multi-dimensional input data and continuously optimizes prediction accuracy through a large number of historical cases. The output of this module is the probability of occurrence and possible impact area of ​​each potential disaster type.

[0028] Finally, the output of the deep learning model will be passed to the early warning module to generate disaster warning information. Based on the predicted geological disaster type, probability and occurrence time, the early warning module generates timely and accurate warning signals and communicates them to relevant personnel and departments through text messages, emails, emergency notifications, etc. This link provides a scientific basis for emergency management of geological disasters.

[0029] Through the sequential connection of these modules, the system realizes a closed loop from data collection to analysis and then to early warning, ensuring the integrity, real-time and accuracy of the data. The data preprocessing module eliminates the interference of source data while ensuring data quality, and the fusion module improves the synergy of multi-source data through intelligent algorithms. The high-precision prediction of the deep learning model module ensures the timeliness and effectiveness of disaster warning. Finally, the timely response of the early warning module helps relevant departments to do a good job in disaster prevention and mitigation, effectively reducing the risks brought by geological disasters.

[0030] In an embodiment of the present invention, the data acquisition module includes a remote sensing data acquisition unit, a ground sensor data acquisition unit and a meteorological data acquisition unit; The remote sensing data acquisition unit obtains high-resolution image data of geological disaster areas through satellite remote sensing technology, including visual interpretation data, spectral data and terrain data; The ground sensor data acquisition unit collects ground vibration data, displacement data, temperature data and humidity data in real time through various ground sensors installed in the geological disaster area, including vibration sensors, displacement sensors, temperature sensors, humidity sensors, etc.; the meteorological data acquisition unit obtains meteorological data in the area through meteorological stations or meteorological satellites, specifically including air temperature data, air humidity data, rainfall and wind speed in the area.

[0031] In a possible implementation, the remote sensing data acquisition unit obtains high-resolution image data of the target geological disaster area through satellite remote sensing technology. Specifically, it includes visual interpretation data, spectral data, and terrain data, which can accurately provide the geological characteristics, vegetation distribution, and terrain changes of the region, and help the system preliminarily determine the potential area of ​​geological disasters. These remote sensing data provide large-scale spatial information for subsequent analysis, and can be quickly monitored in a wide area.

[0032] Furthermore, the ground sensor data acquisition unit collects data such as vibration, displacement, temperature and humidity in real time through various sensors installed in the geological disaster area. Specific sensors such as vibration sensors are used to monitor the vibration changes in the area, displacement sensors can sense the tiny displacement of soil or rock, and temperature and humidity sensors can monitor the changes in ambient temperature and humidity, which are key factors in assessing the occurrence of geological disasters. Through real-time data collection, the ground sensor unit can provide more fine-grained information to help discover subtle changes that cannot be detected by remote sensing data, especially the precursors of local disasters.

[0033] The further meteorological data collection unit obtains meteorological data in the area through meteorological stations or satellites, including air temperature, air humidity, rainfall and wind speed. These data are crucial for the prediction of geological disasters, especially rainfall and wind speed, which directly affect the frequency and intensity of disasters such as landslides and mud-rock flows. Meteorological data provides a dimension of environmental change, allowing the monitoring system to combine real-time weather changes for more accurate disaster risk assessment.

[0034] Through real-time, multi-dimensional data integration, these three data acquisition units enable the system to comprehensively and efficiently capture the multi-faceted signals of geological disasters. Remote sensing data provides a wide range of surface conditions, dynamic changes captured by ground sensors help further refine forecasts, and meteorological data provides the climate background in which disasters may intensify. Through the combination of these three types of data, the system can conduct all-round monitoring of geological disasters, covering a wide area while providing fine-grained local monitoring. This connection mode enables the system to have high accuracy and real-time performance, and can promptly detect potential geological disaster risks, provide decision makers with accurate early warning information and scientific basis, and reduce losses when disasters occur.

[0035] In the embodiment of the present invention, the data preprocessing module preprocesses the collected multi-source data, including data cleaning, correction, alignment, format conversion and normalization; Multi-source data include ground vibration data, displacement data, temperature data, humidity data, air temperature data, air humidity data, rainfall, and wind speed in the area; The preprocessing module removes invalid data and outliers through a data cleaning algorithm, corrects data errors through a data correction algorithm, ensures the temporal and spatial consistency of different data sources through a data alignment algorithm, converts data in different formats into a unified format through a format conversion algorithm, and normalizes the eigenvalues ​​to between 0 and 1 through a normalization algorithm.

[0036] In one possible implementation, the collected data usually includes noise, errors, and missing values. These incomplete or inaccurate data may affect the results of subsequent analysis. Therefore, the data cleaning algorithm first screens all the raw data, identifies and removes invalid data (such as duplicate data, missing values, etc.) and outliers (such as data generated by sudden changes in physical quantities or sensor failures). The cleaned data is more reliable and lays the foundation for subsequent processing.

[0037] In the actual data collection process, due to sensor errors, environmental changes and other factors, the data may have a certain degree of deviation. The data correction algorithm corrects the errors in the data by comparing it with known standard values ​​or historical data to ensure the accuracy of the data. For example, cross-validation and correction can be performed by comparing multiple sensor data at the same location. The corrected data is more accurate and can better reflect the actual situation.

[0038] Furthermore, different data sources (such as vibration sensors, meteorological data, remote sensing data, etc.) may be out of sync in time and space. The data alignment algorithm ensures the consistency of all data in time and space. For example, vibration data and meteorological data may come from different devices and locations. The data alignment algorithm will ensure the consistency of these data at the same time and in the same spatial area through timestamps and geographic coordinates. Data alignment can effectively avoid analysis errors caused by time and space differences and improve monitoring accuracy.

[0039] Different data acquisition devices may use different data formats, making it difficult to directly compare and integrate data. The format conversion algorithm converts all data in different formats into a unified standard format. For example, vibration data may be in CSV format, while meteorological data may be in JSON format. Format conversion allows these data to be processed and analyzed on the same platform.

[0040] Different eigenvalues ​​of data may have different magnitudes (for example, the magnitudes of temperature and humidity vary greatly), which will affect the calculation and analysis of subsequent algorithms. Through the normalization algorithm, all data features are converted to a unified scale, usually by normalizing the eigenvalues ​​to between 0 and 1, eliminating the impact of magnitude differences on the analysis model. After normalization, the data is easier to process by the model, especially in machine learning algorithms, where normalization can improve the convergence speed and accuracy of the model.

[0041] In the embodiment of the present invention, the multi-source data fusion module includes a feature extraction unit, a data alignment and normalization unit and an adaptive data fusion unit; The feature extraction unit performs feature extraction on the preprocessed remote sensing data, ground sensor data and meteorological data, including edge detection, texture analysis and spectral feature extraction of remote sensing images; vibration spectrum analysis, displacement change rate calculation, temperature and humidity time series feature analysis of ground sensors; and feature extraction of temperature change rate, humidity change rate and rainfall trend of meteorological data.

[0042] In one possible implementation, the feature extraction unit is responsible for extracting useful features from different data sources (remote sensing data, ground sensor data, and meteorological data) to facilitate subsequent data fusion and analysis. For remote sensing image data, edge detection is first performed. This process can identify mutation points in the image and help detect areas where geological disasters occur (such as landslides or collapses). Then, texture analysis is performed. By analyzing the texture features of the geological structure in the image, information about the geological structure can be obtained. Spectral feature extraction is also used to obtain the reflectance characteristics of different objects, thereby identifying mineral resources and other important geological features.

[0043] For ground sensor data, vibration spectrum analysis is mainly performed to identify different frequency components in the vibration signal to predict abnormal changes in geological activities. By calculating the displacement change rate, changes in ground displacement can be tracked to provide a basis for the prediction of geological disasters such as landslides. At the same time, time series characteristic analysis of temperature and humidity is performed to help monitor the impact of long-term climate change on the geological environment.

[0044] For meteorological data, we first extract the temperature change rate and humidity change rate to understand the impact of climate change on geological stability. By analyzing the rainfall trend, we can identify the risk of geological disasters caused by precipitation (such as mudslides and landslides).

[0045] Furthermore, the data alignment and normalization unit is mainly responsible for synchronizing the time and space of data from different sources. Since remote sensing data, ground sensor data, and meteorological data may have time differences, the data needs to be aligned through timestamps to ensure that they are analyzed in the same time window. In addition, since different data sources may have different scales or units, the normalization unit will uniformly process all feature values ​​to eliminate magnitude differences, thereby ensuring that the data can be compared and comprehensively analyzed.

[0046] Furthermore, after data alignment and normalization, the adaptive data fusion unit will select the appropriate fusion strategy based on the relevance and importance of different data. The unit can dynamically adjust the fusion method according to the quality and reliability of the data to achieve the optimal combination of data. For example, when monitoring the signs of geological disasters, the unit may give priority to remote sensing image data and vibration spectrum data, while in the case of meteorological anomalies, it may rely more on the changing trend of meteorological data. Adaptive fusion can self-adjust according to real-time conditions, maximize the fusion effect, and improve the accuracy and timeliness of monitoring.

[0047] In the embodiment of the present invention, the data alignment and normalization unit aligns the extracted features to ensure the temporal and spatial consistency of different source data, and then performs normalization processing to normalize the feature values ​​to between 0 and 1; The adaptive data fusion unit adopts an adaptive data fusion algorithm to fuse the normalized features, inputs the feature vectors of remote sensing data, ground sensor data and meteorological data into the adaptive data fusion algorithm, and dynamically adjusts the weights of different data sources.

[0048] In a possible implementation, in the process of multi-source data fusion, it is first necessary to process data from different sources, which may have different timestamps, spatial resolutions and dimensions, so data alignment and normalization must be performed. Specifically, data alignment is performed first, which ensures the consistency of data from different sources in time and space. For example, the acquisition frequency of remote sensing image data may be inconsistent with that of ground sensor or meteorological data, so it is necessary to interpolate and synchronize the data according to the timestamp so that they are in the same time window. In addition, spatial alignment is also very important, especially since remote sensing data and ground sensor data may be collected in different areas, it is necessary to unify the spatial information from different sources through the coordinate system to ensure the spatial consistency of the data. Further, data normalization is performed, which is to convert the values ​​of different features to a unified range for comparison and fusion. Usually, the feature values ​​are normalized to the interval [0,1]. This step eliminates the differences in dimensions between different data sources, so that the feature values ​​of remote sensing data, sensor data and meteorological data have the same magnitude, avoiding the unbalanced impact of the feature of a certain data source on the fusion result due to excessive or too small values.

[0049] Furthermore, the data after data alignment and normalization will enter the adaptive data fusion unit, which dynamically adjusts the weights of different data sources through an adaptive algorithm.

[0050] Specifically, first input the feature vector: the adaptive data fusion unit takes the feature vectors from remote sensing data, ground sensor data and meteorological data as input. Each feature vector contains extracted and normalized feature information, such as spectral features of remote sensing data, vibration data of ground sensors, and temperature and humidity information in meteorological data.

[0051] Furthermore, dynamically adjust the weights: The adaptive data fusion algorithm dynamically adjusts the weights of each data source according to its importance and reliability. For example, in severe weather conditions, meteorological data may become an important basis for disaster prediction, while ground sensor data may be more sensitive when monitoring geological activities. The adaptive algorithm automatically adjusts the weights of different data sources through real-time feedback and data quality assessment, so that the final fused feature vector can more accurately reflect the geological status of the target area.

[0052] Finally, the fusion result is output: the fusion result will provide a comprehensive geological feature vector, covering the key information of remote sensing data, ground sensor data and meteorological data, for subsequent geological disaster warning and resource exploration.

[0053] In an embodiment of the present invention, the deep learning model module includes a data preparation unit, a model selection unit, a model training unit and a model evaluation unit; The data preparation unit takes multi-source data of historical geological disaster events as the input of the model, takes the type and probability of occurrence of geological disasters as the output, and constructs training data sets and test data sets; The model selection unit adopts a Transformer-based multimodal deep learning model, and the model training unit uses a training dataset to train the model, including cross-validation and hyperparameter optimization. During the training process, loss functions and optimization algorithms are used to optimize model parameters.

[0054] In one possible implementation, the data preparation unit is the first step of the entire deep learning process, and its main task is to prepare multi-source data of historical geological disaster events as input data for the deep learning model. Specific operations include: multi-source data collection and data annotation and processing, wherein multi-source data collection involves collecting various types of remote sensing data, meteorological data, ground sensor data, etc., to form a historical geological disaster data set with geospatial information. Furthermore, to ensure the validity of the data, the data preparation unit needs to annotate the historical data, and use the type of geological disaster event (such as landslide, mudslide, etc.) and the probability of occurrence (or severity level) as output labels to construct a training data set. At the same time, the input data is cleaned and preprocessed to eliminate noise and fill in missing values ​​to ensure data quality.

[0055] Furthermore, after the data preparation is completed, the model selection unit is responsible for selecting a suitable deep learning model. In this system, a multimodal deep learning model based on Transformer is selected. The Transformer model has a strong sequence modeling capability and can effectively process time series and spatial data, especially when processing complex structured data such as remote sensing images and sensor data. The advantage of the multimodal model is that it can process multiple types of data (such as images, time series data, text, etc.) at the same time, so that it can integrate the characteristics of different data sources and improve the accuracy of geological disaster prediction.

[0056] Furthermore, after the model is selected, the training phase begins. The model training unit uses the training data set generated by the data preparation unit to train the selected Transformer model. The specific steps include: Cross-validation: Use cross-validation technology to divide the training set into multiple subsets, perform training and validation on different subsets, evaluate the generalization ability of the model, and avoid overfitting.

[0057] Hyperparameter optimization: During the training process, the model's hyperparameters (such as learning rate, batch size, network depth, etc.) are adjusted to find the best parameter combination through methods such as grid search or Bayesian optimization.

[0058] Loss function and optimization algorithm: When training a model, a loss function (such as mean square error or cross entropy) is used to quantify the gap between the model prediction and the true label, and an optimization algorithm (such as Adam, SGD, etc.) is used to optimize the model parameters and reduce the loss value, thereby making the model's prediction effect more and more accurate.

[0059] After the model training is completed, the model evaluation unit is used to evaluate the performance of the trained model. By using the test data set, various evaluation indicators (such as precision, recall rate, F1 score, etc.) are calculated to determine the accuracy and stability of the model. At the same time, based on the evaluation results, the training phase may be returned for further adjustment and optimization.

[0060] In the embodiment of the present invention, the loss function and optimization algorithm are used to optimize the model parameters, and the specific training steps are as follows: S1: Input the multi-source data feature vectors into the Transformer model, with each feature vector serving as an input token; S2: Through the multi-head self-attention mechanism, the correlation between each token and other tokens is calculated to generate an attention matrix; S3: Perform feedforward neural network transformation on the feature vector of each token to enhance the nonlinear expression ability of the feature; S4: Use the cross entropy loss function to calculate the difference between the model prediction value and the true value; S5: Adopt Adam optimization algorithm to optimize model parameters through gradient descent method; The model evaluation unit uses the test data set to evaluate the trained model and evaluates the performance of the model through accuracy, recall, and F1 score indicators.

[0061] In one possible implementation, multi-source data (such as remote sensing image data, geographic information, meteorological data, etc.) is preprocessed and converted into feature vectors. These feature vectors represent the key features of different types of data and are fed into the self-attention mechanism of the Transformer model as input. Each feature vector is regarded as a token, representing the basic unit of model input.

[0062] The multi-head self-attention mechanism is a core component of the Transformer model. It calculates the correlation (i.e., "attention") between each token and other tokens and captures different aspects of the relationship through different "heads". This mechanism helps the model learn the complex dependencies between input features and helps extract more representative features. The generated attention matrix is ​​the result of the model's global dependency modeling of the input data.

[0063] Furthermore, Transformer transforms the feature vector of each token through a feedforward neural network. A feedforward neural network usually consists of multiple fully connected layers. After passing through an activation function (such as ReLU), it can enhance the nonlinear expression ability of the feature vector. This process helps the model capture more complex data patterns and improve the expressiveness of features.

[0064] Furthermore, the gap between the model prediction and the actual value (label) is measured by the cross entropy loss function. The cross entropy loss function is widely used in classification tasks. It can quantify the difference between the probability distribution of the model output and the true label. A smaller cross entropy loss means that the model prediction is closer to the true value, thus driving the model to optimize towards higher accuracy.

[0065] The Adam optimization algorithm is a commonly used gradient descent algorithm with high efficiency and robustness. It dynamically adjusts the learning rate by calculating the average value and square mean of the gradient to optimize the model parameters. During the training process, the algorithm can automatically adapt to the changes of different features and gradients, improve the convergence speed, and avoid the local minimum problem that may occur in traditional gradient descent.

[0066] In an embodiment of the present invention, the early warning module includes a prediction result generation unit and an early warning signal generation unit. The prediction result generation unit inputs multi-source data collected in real time into a trained deep learning model to generate a prediction result of the type and probability of occurrence of geological disasters; The warning signal generation unit generates geological disaster warning information according to the prediction results, and sends out warning signals through the notification unit; Notification units include SMS notification, email notification, sound and light alarm and automatic notification to emergency management departments. The content of the early warning signal includes the type of disaster, probability of occurrence, expected time of occurrence, affected area and recommended response measures.

[0067] In one possible implementation, first, multi-source data (including remote sensing images, meteorological data, geological data, etc.) collected in real time are transmitted to a trained deep learning model. The deep learning model is fully trained to identify and predict the probability and type of various geological disasters. Through this process, the system can process and analyze information from different sensors and data sources in real time to predict the type of disaster (such as landslides, earthquakes, mudslides, etc.) and the probability of occurrence.

[0068] Furthermore, the prediction result generation unit obtains the type of disaster, probability of occurrence, expected time of occurrence, and affected area based on the output of the deep learning model. This information is obtained after analyzing multi-dimensional data (such as soil moisture, terrain changes, weather conditions, etc.), helping to predict the risk of geological disasters and their impact range.

[0069] Furthermore, after obtaining the prediction results, the warning signal generation unit generates specific warning information based on the disaster type, probability, expected time of occurrence and affected area, combined with the predetermined warning rules (such as probability threshold, time window, etc.). The warning information includes the disaster type, probability of occurrence, expected time of occurrence, affected area and the recommended response measures according to the specific situation (such as evacuation route, emergency response steps, etc.).

[0070] Furthermore, the warning signal is transmitted to relevant personnel and departments through various means, including SMS notification, email notification, sound and light alarm, and automatic notification to emergency management departments. The content of the warning signal includes specific information about the disaster (such as disaster type, probability of occurrence, expected time of occurrence, affected area) and response measures (such as emergency avoidance, evacuation instructions, etc.), ensuring that relevant personnel receive effective warnings and guidance in a timely manner.

[0071] In the embodiment of the present invention, the mathematical expression of the vibration spectrum analysis is: ; in, It is a vibration signal. is the frequency, j is an imaginary unit, specifically .

[0072] In a possible implementation, during geological survey, the original vibration signal is collected by remote sensing data acquisition equipment or ground vibration sensors (such as accelerometers, seismographs, etc.) These signals represent geological activity in a particular area and can be caused by natural disasters (such as earthquakes) or vibration waves caused by human activities (such as blasting operations).

[0073] The collected time domain vibration signal It is usually a complex signal that may contain noise and multiple frequency components. Therefore, before spectrum analysis, the signal is usually denoised, such as using a filter or noise reduction algorithm to ensure the accuracy of the signal.

[0074] Furthermore, after preprocessing, the vibration signal is converted into a frequency domain representation through Fourier transform (the mathematical expression given in the formula). Fourier transform converts the time domain signal Mapping to the frequency domain allows the amplitude and phase information of each frequency component to be clearly presented. In the formula, represents the frequency spectrum of the vibration signal, where is the frequency, and It includes the strength and changes of the signal at each frequency point.

[0075] Furthermore, the amplitude and phase information of different frequency components can be extracted from the spectrum results. By analyzing these spectrum characteristics, it can help determine the characteristics of the vibration source in the geological exploration area, such as whether there are abnormal geological activities (such as earthquake sources, earthquake precursors before landslides, etc.). High-frequency components usually represent small-scale, rapidly changing vibration sources, while low-frequency components may represent large-scale, slowly changing geological activities.

[0076] During geological surveys, the system can predict potential geological disasters by analyzing different frequency components. If abnormal frequency components appear in the spectrum, or the amplitude of certain frequency components increases abnormally, this may be a sign that certain disasters (such as earthquakes or landslides) are about to occur. The system can combine this information and notify relevant personnel through the early warning module to respond and prepare for emergencies.

[0077] In the embodiment of the present invention, the mathematical expression for calculating the displacement change rate is: ;in, is the change in displacement, is the change in time, and are the displacement values ​​respectively.

[0078] In a possible implementation, the mathematical expression for calculating the displacement change rate is: , in this expression, and represent the displacement values ​​at different time points, and is the time interval between these two time points, is the change in displacement during this time period.

[0079] In geological survey and monitoring, remote sensing technology (such as satellite images, drone monitoring, etc.) is used to obtain the displacement information of ground objects at different times. Geological activities, such as earthquakes, landslides or land subsidence, usually cause changes in the position of the ground surface. Through continuous remote sensing data collection, ground displacement data at a series of time points can be obtained.

[0080] Get displacement data at different time points and Then, use the above formula to calculate the displacement change. Here, is the displacement value at the initial moment, is the displacement value at a later time, and the difference between the two is equal to the time interval Combined with the above data, the displacement change rate per unit time is calculated. This change rate directly reflects the dynamic changes of ground objects or geological structures.

[0081] In practical applications, remote sensing data is often preprocessed (such as geometric correction, radiation correction, etc.) to ensure data accuracy. After that, remote sensing images or LiDAR technology are used to obtain displacement information at each moment. These data are input into the displacement change rate calculation module, and the displacement change rate in different time periods is calculated through the formula.

[0082] The calculated displacement change rate can be used to reflect the dynamic changes in the geological exploration area. If the value increases significantly, it may indicate some geological activities (such as crustal deformation before an earthquake or soil movement before a landslide). By tracking and analyzing the continuous displacement change rate data, the trend and potential risks of geological changes can be determined, helping surveyors to take preventive measures in a timely manner.

[0083] The present invention covers any substitution, modification, equivalent method and scheme made on the essence and scope of the present invention. In order to make the public have a thorough understanding of the present invention, specific details are described in detail in the following preferred embodiments of the present invention, but those skilled in the art can fully understand the present invention without the description of these details. In addition, in order to avoid unnecessary confusion about the essence of the present invention, well-known methods, processes, procedures, components and circuits are not described in detail.

[0084] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A field geological survey and monitoring system based on remote sensing data, characterized in that: It includes data acquisition module, data preprocessing module, multi-source data fusion module, deep learning model module and early warning module; The data acquisition module, data preprocessing module, multi-source data fusion module, deep learning model module and early warning module are connected in sequence.

2. A field geological survey monitoring system based on remote sensing data according to claim 1, characterized in that: The data acquisition module includes a remote sensing data acquisition unit, a ground sensor data acquisition unit and a meteorological data acquisition unit; The remote sensing data acquisition unit obtains high-resolution image data of geological disaster areas through satellite remote sensing technology, including visual interpretation data, spectral data and terrain data; The ground sensor data acquisition unit collects ground vibration data, displacement data, temperature data and humidity data in real time through various ground sensors installed in the geological disaster area, including vibration sensors, displacement sensors, temperature sensors and humidity sensors; the meteorological data acquisition unit obtains meteorological data in the area through meteorological stations or meteorological satellites, including air temperature data, air humidity data, rainfall and wind speed in the area.

3. A field geological survey monitoring system based on remote sensing data according to claim 2, characterized in that: The data preprocessing module preprocesses the collected multi-source data, including data cleaning, correction, alignment, format conversion and normalization; Multi-source data include ground vibration data, displacement data, temperature data, humidity data, air temperature data, air humidity data, rainfall, and wind speed in the area; The preprocessing module removes invalid data and outliers through a data cleaning algorithm, corrects data errors through a data correction algorithm, ensures the temporal and spatial consistency of different data sources through a data alignment algorithm, converts data in different formats into a unified format through a format conversion algorithm, and normalizes the eigenvalues ​​to between 0 and 1 through a normalization algorithm.

4. A field geological survey monitoring system based on remote sensing data according to claim 3, characterized in that: The multi-source data fusion module includes a feature extraction unit, a data alignment and normalization unit and an adaptive data fusion unit; The feature extraction unit performs feature extraction on the preprocessed remote sensing data, ground sensor data and meteorological data, including edge detection, texture analysis and spectral feature extraction of remote sensing images; vibration spectrum analysis, displacement change rate calculation, temperature and humidity time series feature analysis of ground sensors; temperature change rate, humidity change rate and rainfall trend feature extraction of meteorological data.

5. A field geological survey and monitoring system based on remote sensing data according to claim 4, characterized in that: The data alignment and normalization unit aligns the extracted features to ensure the temporal and spatial consistency of different source data, and then performs normalization processing to normalize the feature values ​​to between 0 and 1; The adaptive data fusion unit adopts an adaptive data fusion algorithm to fuse the normalized features, inputs the feature vectors of remote sensing data, ground sensor data and meteorological data into the adaptive data fusion algorithm, and dynamically adjusts the weights of different data sources.

6. A field geological survey monitoring system based on remote sensing data according to claim 1, characterized in that: The deep learning model module includes a data preparation unit, a model selection unit, a model training unit and a model evaluation unit; The data preparation unit takes multi-source data of historical geological disaster events as the input of the model, takes the type and probability of occurrence of geological disasters as the output, and constructs training data sets and test data sets; The model selection unit adopts a Transformer-based multimodal deep learning model, and the model training unit uses a training dataset to train the model, including cross-validation and hyperparameter optimization. During the training process, loss functions and optimization algorithms are used to optimize model parameters.

7. A field geological survey monitoring system based on remote sensing data according to claim 1, characterized in that: The loss function and optimization algorithm are used to optimize the model parameters. The specific training steps are as follows: S1: Input the multi-source data feature vectors into the Transformer model, with each feature vector serving as a token of the input; S2: Through the multi-head self-attention mechanism, the correlation between each token and other tokens is calculated to generate an attention matrix; S3: Perform feedforward neural network transformation on the feature vector of each token to enhance the nonlinear expression ability of the feature; S4: Use the cross entropy loss function to calculate the difference between the model prediction value and the true value; S5: Adopt Adam optimization algorithm to optimize model parameters through gradient descent method; The model evaluation unit uses the test data set to evaluate the trained model and evaluates the performance of the model through accuracy, recall, and F1 score indicators.

8. A field geological survey monitoring system based on remote sensing data according to claim 1, characterized in that: The early warning module includes a prediction result generation unit and an early warning signal generation unit. The prediction result generation unit inputs multi-source data collected in real time into a trained deep learning model to generate a prediction result of the type and probability of geological disasters. The early warning module also includes a notification unit, and the early warning signal generation unit generates geological disaster early warning information according to the prediction results and sends out early warning signals through the notification unit; Notification units include SMS notification, email notification, sound and light alarm and automatic notification to emergency management departments. The content of the early warning signal includes the type of disaster, probability of occurrence, expected time of occurrence, affected area and recommended response measures.

9. A field geological survey and monitoring system based on remote sensing data according to claim 4, characterized in that: The mathematical expression of the vibration spectrum analysis is: ； in, It is a vibration signal. is the frequency, j is an imaginary unit, specifically .

10. A field geological survey and monitoring system based on remote sensing data according to claim 4, characterized in that: The mathematical expression for calculating the displacement change rate is: ;in, is the change in displacement, is the change in time, and are the displacement values ​​respectively.

Images