Multi-source time-series remote sensing soil water content prediction technology assisted geological disaster hidden danger investigation method and system

Abstract

The application provides a geohazard hidden danger investigation method and system assisted by a multi-source time sequence remote sensing soil water content prediction technology, and the method comprises the following steps: pre-processing multi-source time sequence remote sensing data associated with soil water content to obtain model basic data and construct a model sample database; using the sample database to optimize and train a pre-constructed soil water content prediction deep learning model to obtain model parameter initialization model with optimal weight; using the model parameter initialization model with optimal weight to predict soil water content at a next time step of the multi-source time sequence remote sensing data to obtain a prediction result of the soil water content, and then using the prediction result to assist in geohazard hidden danger investigation. By using the application, the defects of high measurement cost and low efficiency of the traditional soil water content monitoring scheme can be overcome, and comprehensive analysis can be performed in combination with other related information to improve the accuracy of geohazard hidden danger investigation based on relatively single remote sensing data.

Description

Technical Field

[0001] This invention relates to the fields of soil moisture content prediction and geological hazard investigation, and more specifically, to a method and system for geological hazard investigation assisted by multi-source time-series remote sensing soil moisture content prediction technology. Background Technology

[0002] Frequent geological disasters, such as landslides, collapses, and mudslides, severely disrupt and threaten the daily operation of infrastructure in various sectors, and also pose a serious threat to the lives and property of people living along the affected areas. Therefore, it is of great significance to conduct high-precision and regular advance investigations of potential geological disaster hazard points using technological means.

[0003] Remote sensing can be used to conduct large-scale, timely investigations of potential geological hazards. For example, high-resolution optical remote sensing imagery can interpret changes in the spectral texture and other characteristics of land features before a geological disaster to identify potential hazard points. Furthermore, InSAR technology can acquire deformation information within a region, which can be used to further determine the specific circumstances of potential hazard points. Currently, remote sensing technology can be used to investigate potential geological hazards; however, the methods mentioned above all neglect the relationship between soil moisture content and the formation of geological hazards, relying solely on the spectral or deformation characteristics of potential hazard points. This means that the accuracy and reliability of remote sensing-based investigations of potential geological hazards have room for improvement.

[0004] Soil moisture, also known as soil humidity or soil water content, is a crucial parameter in the Earth's biochemical cycles and energy exchange processes. Increased soil moisture content leads to decreased inter-soil friction and increased pore water pressure, thus increasing the risk of geological disasters such as landslides, collapses, and ground subsidence, as exemplified by geological disasters triggered by continuous heavy rainfall. Therefore, information on changes in soil moisture content is of significant value for geological disaster investigation and prevention.

[0005] However, current traditional soil moisture content monitoring methods often rely on field sampling combined with laboratory measurements. The advantage of this method is that it can obtain accurate soil moisture content information. The disadvantage is that the obtained soil moisture content can only reflect the soil moisture content of a single point at the sampling location. This method is not only small in scale but also requires a long laboratory measurement time. It also requires a large amount of manpower and physical resources, making it difficult to meet the requirements for large-scale and rapid updates of soil moisture content in geological disaster hazard investigation scenarios. Summary of the Invention

[0006] In view of the above problems, the purpose of this invention is to provide a geological disaster hazard investigation method assisted by multi-source time-series remote sensing soil water content prediction technology, so as to solve the problems of low investigation accuracy and poor reliability in the current geological disaster hazard investigation schemes.

[0007] This invention provides a method for identifying potential geological hazards assisted by multi-source time-series remote sensing soil moisture content prediction technology, comprising:

[0008] The acquired multi-source time-series remote sensing data associated with soil water content are preprocessed to obtain the basic data for the model.

[0009] A model sample database is constructed based on the aforementioned model foundation data;

[0010] The pre-constructed deep learning model for predicting soil water content is optimized and trained using the sample database to obtain the model parameter initialization model with optimal weights. During the training process, the model is trained with deep supervision using sample category information and supervised training using sample time series information. The randomly initialized parameters are optimized and adjusted, and the model weights that perform best on the validation dataset are saved.

[0011] The model is initialized using the optimal weighted model parameters, and the soil water content of the next time step of the input multi-source time-series remote sensing data of the area to be predicted is predicted to obtain the predicted soil water content of the area.

[0012] Based on the predicted soil water content in the region, the rate of change and the second-order rate of change of soil water content are determined; based on the rate of change and the second-order rate of change, the magnitude of geological disaster risk in the region is determined.

[0013] Alternatively, the multi-source time-series remote sensing data may include optical remote sensing time-series data, synthetic aperture radar time-series data, land cover time-series data for the corresponding region, and meteorological precipitation data.

[0014] Alternatively, the preprocessing may include geometric correction, radiometric correction, exponential analysis, sampling interpolation, normalization, and multi-source fusion.

[0015] Alternatively, the index analysis can be conducted using the normalized vegetation index (NVI) to analyze vegetation growth status, thereby indirectly reflecting soil water content; the calculation formula for the NVI is as follows:

[0016]

[0017] NDVI represents the Normalized Difference Vegetation Index, and NIR and R represent the near-infrared and red bands of optical remote sensing images, respectively.

[0018] Alternatively, the normalization can be achieved through Z-Score normalization to unify the value range and data distribution from different data sources; wherein, Z-Score normalization is shown in the following formula:

[0019]

[0020] Where Z represents the normalized data, X represents the data to be normalized, and the data to be normalized includes optical remote sensing data, radar remote sensing data, land cover data and derived normalized vegetation index after maximum-minimum normalization processing; μ and σ represent the mean and standard deviation of the data to be normalized, respectively.

[0021] Alternatively, the multi-source fusion method can be used to fuse the normalized data by weighted fusion and addition, as shown in the following formula:

[0022]

[0023] Where X1, X2, X3, X index These represent optical remote sensing data, SAR radar data, land cover type data, and remote sensing indices obtained through index analysis, respectively; a, b, c, and d are X1, X2, X3, and X... index Each has its own proportional weight.

[0024] In addition, an alternative approach is to construct a model sample database based on the model's basic data, including: performing time-series block pruning on the model's basic data, and mapping the time-series block pruning data one-to-one with soil water content label samples, so that each time-series input sample has a time-series label and a category label; and performing data augmentation on the time-series input samples with time-series labels and category labels.

[0025] Alternatively, the input samples in the model sample database are processed multi-source time-series remote sensing data, and the corresponding ground truth labels are soil water content data at the next time step above and below the time-series node; wherein, the method for creating the ground truth labels includes:

[0026] S201: Based on the latitude and longitude information in the image, perform regional matching between the temporal input data and the soil moisture dataset representing the corresponding ground area;

[0027] S202: Based on the time-series input data (T0, T1, T2…Tn-1), obtain the soil moisture data of the next time node (Tn) of the corresponding region as the time-series label;

[0028] S203: Based on the average precipitation meteorological data at the time nodes (T0-Tn-1) of the time series input data area, the corresponding time series input samples are divided into arid area samples, semi-arid samples, humid samples and humid area samples, and regional category label information is constructed.

[0029] Following the data processing methods of S201 to S203, all sample inputs and labels are matched and aligned sequentially. Finally, each time-series input sample has a category label reflecting the average dryness and wetness at the time-series node of the sample, and a time-series label reflecting the soil water content at the next time step.

[0030] On the other hand, the present invention also provides a geological disaster hazard investigation system assisted by multi-source time-series remote sensing soil moisture content prediction technology, used for geological disaster hazard investigation using the aforementioned multi-source time-series remote sensing soil moisture content prediction technology-assisted geological disaster hazard investigation method, the system comprising:

[0031] The basic data acquisition unit is used to preprocess the acquired multi-source time-series remote sensing data associated with soil water content to obtain the basic data for the model.

[0032] A database construction unit is used to construct a model sample database based on the model's basic data.

[0033] The model training unit is used to optimize and train the pre-constructed deep learning model for predicting soil water content using the sample database, and to initialize the model with the optimal model parameters. During the training process, the model is trained with deep supervision using sample category information and supervised training using sample time series information. The randomly initialized parameters are optimized and adjusted, and the optimal model weights are saved on the validation dataset.

[0034] The prediction unit is used to initialize the model using the model parameters with the optimal weights, predict the soil water content of the next time step of the input multi-source time-series remote sensing data of the area to be predicted, and obtain the prediction result of the soil water content of the area.

[0035] The hazard investigation unit is used to determine the rate of change and the second-order rate of change of soil water content based on the predicted soil water content of the area; and to determine the magnitude of geological disaster risk in the area based on the rate of change and the second-order rate of change in conjunction with other indicators.

[0036] As can be seen from the above technical solution, the multi-source time-series remote sensing soil moisture content prediction technology-assisted geological disaster hazard investigation method and system provided by the present invention first predicts soil moisture content based on selected multi-source time-series remote sensing data and a pre-constructed deep learning model for soil moisture content prediction, and obtains regional soil moisture content prediction results. Subsequently, based on the obtained regional soil moisture content prediction results and other relevant indicators (such as regional deformation and optical texture features obtained by InSAR technology), the magnitude of geological disaster hazard risk in the region is comprehensively determined. This not only overcomes the shortcomings of high measurement cost and low efficiency of traditional soil moisture content monitoring schemes, but also allows for comprehensive analysis by combining regional soil moisture content prediction results, regional deformation obtained by InSAR technology, and regional spectral texture information provided by optical images, thereby improving the accuracy of geological disaster hazard investigation based on relatively single remote sensing data.

[0037] To achieve the foregoing and related objectives, one or more aspects of the invention include the features that will be described in detail below. The following description and accompanying drawings illustrate certain exemplary aspects of the invention. However, these aspects indicate only a few of the various ways in which the principles of the invention can be used. Furthermore, the invention is intended to encompass all such aspects and their equivalents. Attached Figure Description

[0038] Other objects and results of the invention will become more apparent and readily understood with reference to the following description taken in conjunction with the accompanying drawings. In the drawings:

[0039] Figure 1 A flowchart illustrating a method for investigating potential geological hazards assisted by multi-source temporal remote sensing soil moisture content prediction technology according to an embodiment of the present invention.

[0040] Figure 2 This is the overall technical approach for a geological disaster hazard investigation method assisted by multi-source temporal remote sensing soil water content prediction technology according to embodiments of the present invention;

[0041] Figure 3 This is a schematic diagram illustrating the adaptive weighted fusion process of multi-source data according to an embodiment of the present invention;

[0042] Figure 4 This is a schematic diagram illustrating the correspondence between time-series sample data X and two types of labels according to an embodiment of the present invention;

[0043] Figure 5 This is a schematic diagram of a convolutional neural network structure with an attention fusion mechanism according to an embodiment of the present invention;

[0044] Figure 6 This is a schematic diagram illustrating the process of obtaining deformation results of a target area using InSAR technology according to an embodiment of the present invention;

[0045] Figure 7 This is a schematic diagram of the overall process for investigating potential geological disaster sites according to an embodiment of the present invention.

[0046] In all the accompanying drawings, the same reference numerals indicate similar or corresponding features or functions. Detailed Implementation

[0047] In the following description, numerous specific details are set forth for illustrative purposes and to provide a thorough understanding of one or more embodiments. However, it will be apparent that these embodiments may also be implemented without these specific details. In other instances, well-known structures and devices are shown in block diagram form for ease of description of one or more embodiments.

[0048] To address the aforementioned problems of traditional soil moisture content monitoring being small-scale, requiring lengthy laboratory measurements, and incurring significant human and physical costs, which makes it difficult to meet the requirements of large-scale and rapid updates of soil moisture content in geological disaster hazard investigation scenarios, this invention provides a method and system for geological disaster hazard investigation assisted by multi-source time-series remote sensing soil moisture content prediction technology. This method predicts soil moisture content and further processes the obtained soil moisture content factors, incorporating them into the geological disaster hazard investigation process to achieve multi-source, high-precision investigation of geological disaster hazards. The method first proposes a model for predicting soil moisture content changes based on multi-source time-series remote sensing data. This model obtains predicted soil moisture content values ​​for future time steps in a region. Subsequently, this information is processed and comprehensively analyzed with regional deformation information obtained through InSAR technology and regional spectral texture information provided by optical imagery to improve the accuracy of geological disaster hazard investigation compared to using single remote sensing data.

[0049] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0050] To illustrate the geological disaster hazard investigation method assisted by the multi-source time-series remote sensing soil moisture content prediction technology provided by this invention. Figure 1 The flowchart of a method for investigating potential geological hazards assisted by multi-source temporal remote sensing soil water content prediction technology according to an embodiment of the present invention is shown.

[0051] like Figure 1 As shown, the geological disaster hazard investigation method assisted by multi-source time-series remote sensing soil water content prediction technology provided by the present invention includes:

[0052] S1: Preprocess the acquired multi-source time-series remote sensing data associated with soil water content to obtain the basic data for the model;

[0053] S2: Construct a model sample database based on the aforementioned model foundation data;

[0054] S3: The pre-constructed deep learning model for predicting soil water content is optimized and trained using the sample database to obtain the model parameter initialization model with optimal weights; wherein, during the training process, the model is trained with deep supervision using sample category information and supervised training using sample time series information, the randomly initialized parameters are optimized and adjusted, and the model weights that perform best on the validation dataset are saved.

[0055] S4: Initialize the model using the optimal weighted model parameters, and predict the soil water content of the next time step of the input multi-source time-series remote sensing data of the area to be predicted, to obtain the predicted soil water content of the area.

[0056] Based on this, after predicting the soil water content in the area to be predicted, the following step S5 is further included: determining the rate of change and the second-order rate of change of soil water content based on the predicted soil water content of the area; and determining the magnitude of the geological disaster risk in the area based on the rate of change and the second-order rate of change.

[0057] Based on multi-source time-series remote sensing data, this invention designs and constructs a hybrid deep learning model structure for time-series prediction of regional soil water content. The predicted soil water content is then applied to the investigation and screening of regional geological disaster hazard points, thereby improving the accuracy of geological disaster hazard point screening by combining multi-source remote sensing data. Figure 2 The overall technical route for the geological disaster hazard investigation method assisted by multi-source time-series remote sensing soil water content prediction technology provided by this invention.

[0058] like Figure 2 As shown, soil moisture content prediction is a prerequisite for conducting geological disaster hazard investigation. During the soil moisture content prediction process:

[0059] Step 1: Acquire multi-source time-series remote sensing data and perform necessary preprocessing steps, including geometric radiometric correction, exponential analysis, sampling interpolation, and multi-source fusion, to prepare the data for subsequent steps.

[0060] Step 2: The data obtained and processed in Step 1 is segmented and pruned, data augmented, and mapped one-to-one with soil water content label samples to build a sample library for training deep learning models.

[0061] Step 3: Construct a deep learning model for predicting soil water content. In a specific embodiment of the present invention, the model is composed of a hybrid convolutional neural network (CNN) and an attention network (Transform), and its model parameters are randomly initialized.

[0062] Step 4: Using the constructed sample library, optimize and train the model designed in Step 3. During the training process, use sample category information to perform deep supervised training and sample time series information to perform supervised training. Optimize and adjust the randomly initialized parameters and save the model weights that perform best on the validation dataset.

[0063] Step 5: Initialize the model based on the optimal weighted model parameters, input multi-source time-series remote sensing data, and use the model to predict the soil water content for the next time step to obtain the predicted soil water content.

[0064] Step 6: After obtaining the predicted results of regional soil water content, the deformation analysis of the region is first carried out by combining multi-source remote sensing data, including optical and SAR remote sensing images. Then, the geological disaster characteristics of key areas are further analyzed and extracted. In this process, the predicted soil water content information is combined to improve the accuracy of the investigation of potential geological disaster points in the region.

[0065] The following section will provide a more detailed technical explanation of each of the above steps.

[0066] This invention utilizes multi-source temporal remote sensing data and a hybrid deep learning model to perform temporal prediction of soil moisture content. Remote sensing technology receives and analyzes the electromagnetic wave reflection curves of different ground features, enabling the differentiation of different ground features and the refined inversion of relevant parameters.

[0067] In acquiring multi-source data using remote sensing technology, soils with different water contents exhibit different reflection curves for the same electromagnetic wave band, and soils with the same water content also show different reflection or scattering curves for different electromagnetic wave bands. Therefore, this invention employs comprehensive modeling and analysis of the relationship between the electromagnetic spectrum and soil moisture content using multi-source remote sensing data (different electromagnetic wave bands), which is crucial for obtaining high-precision soil moisture content information. Soil samples with different water contents possess differentiated optical characteristics; arid soil samples often appear sandy and yellowish, while moist soil samples are darker in color and lower in brightness, reflected in optical images as either brighter or darker. The near-infrared band is sensitive to changes in moisture content. Therefore, optical remote sensing images using the visible-near-infrared band are also of significant value. Furthermore, the entire physical process of obtaining soil scattering characteristics from SAR radar images is well-defined, with a highly deterministic electromagnetic transmission model, and the different polarization modes of radar waves are also important for describing soil moisture content information from multiple angles and orientations.

[0068] Furthermore, considering the significant correlation between soil water content and soil surface cover type—for example, plant roots help retain water, while desert areas have poor water retention—land cover use data is also used as an important auxiliary data for soil moisture content inversion and prediction in the embodiments provided by this invention.

[0069] In addition, in one specific embodiment of the present invention, considering that the growth status of vegetation is highly correlated with the moisture content in the soil and there is a certain causal relationship, the growth status of vegetation can also indirectly reflect the soil moisture content. The prediction of soil moisture content can be assisted by calculating the remote sensing index that reflects the vegetation status. The data reflecting the soil moisture content includes meteorological precipitation data, etc.

[0070] Therefore, in a specific embodiment provided by this invention, the multi-source remote sensing raw data and related auxiliary data used in training the model include visible light remote sensing time-series data in the (red-green-blue-near-infrared) band, synthetic aperture radar (SAR) time-series data in fully polarized modes (HH, VV, VH, HV), land cover time-series data of the corresponding region, and meteorological precipitation data, etc. Based on the direct correlation between the above raw remote sensing data and soil water content, the relationship between them can be directly modeled. Therefore, when preprocessing the multi-source time-series remote sensing data, it can be preprocessed and used as part of the subsequent model input.

[0071] After acquiring the raw remote sensing data, geometric and radiometric corrections are first performed. Then, through remote sensing index analysis, indirect indicators reflecting soil moisture content, such as vegetation indices, are calculated, serving as one of the data sources for soil moisture content prediction. In the embodiments provided by this invention, index analysis is performed on optical remote sensing images to calculate the Normalized Difference Vegetation Index (NDVI), which is derived from the raw remote sensing data to assist in the inversion and prediction of soil moisture content.

[0072] Subsequently, a unified sampling and interpolation operation was performed on the acquired data from different data sources (optical remote sensing data, radar remote sensing data, land cover data, etc.), raw remote sensing data of various resolutions, and remote sensing index data obtained from index calculations, to unify the spatial size and resolution of different data. Data normalization was then performed, specifically including Z-score normalization to adjust all data to a standard normal distribution with a mean of 0 and a standard deviation of 1.

[0073] Specifically, the raw optical and radar remote sensing data need to be processed as follows:

[0074] 1. Radiometric correction: Eliminate radiometric distortion caused by factors such as sensor characteristics, atmospheric conditions, and changes in solar altitude angle, and ensure that images acquired by different sensors and at different acquisition times have consistent radiometric characteristics, which facilitates subsequent analysis of multi-source and multi-temporal data.

[0075] 2. Geometric correction: Corrects geometric distortions caused by factors such as sensor platform movement, terrain undulation, Earth curvature, and atmospheric refraction during image acquisition. It accurately locates the image pixels in the geographic coordinate system, which is of great significance for the matching and accurate overlay of subsequent multi-source and multi-temporal remote sensing images.

[0076] 3. Remote sensing index calculation:

[0077] After radiometric and geometric corrections, further feature analysis is performed on the optical imagery to calculate the corresponding remote sensing indices. In this embodiment, the remote sensing index used for soil water content inversion and prediction is the Normalized Difference Vegetation Index (NDVI), calculated using the following formula:

[0078]

[0079] As shown in formula (1) above, the formula for calculating the Normalized Difference Vegetation Index (NDVI) is as follows: NIR and R represent the near-infrared band and red band of the optical remote sensing image, respectively. The Normalized Difference Vegetation Index (NDVI) can be calculated by subtracting the red band from the near-infrared band and the ratio of the near-infrared band plus the red band. This index reflects the current state of vegetation.

[0080] The NDVI (Normalized Difference Vegetation Index) used in this embodiment reflects the current growth status of vegetation and is related to soil water content. The NDVI can indirectly reflect the soil water content by analyzing the vegetation growth status, which is of great help in the inversion and prediction of soil water content.

[0081] 4. Sampling interpolation and normalization processing:

[0082] To perform meaningful multi-source data fusion processing on the aforementioned data, it is necessary to first perform uniform sampling interpolation on the data of different sizes and resolutions. The interpolation method used in this embodiment is bilinear interpolation, which achieves consistency in spatial size and resolution among optical remote sensing data, radar remote sensing data, land cover data, and remote sensing index data through sampling interpolation.

[0083] The data from different data sources are then normalized to unify the value range and data distribution of different sources so that a meaningful weighted fusion process can be carried out later.

[0084] Z-Score normalization is used to unify the data distribution from different data sources. The Z-Score normalization formula is shown below:

[0085]

[0086] Where X represents the data to be normalized. In this embodiment, the data to be normalized includes optical remote sensing data, radar remote sensing data, land cover data, and the derived NDVI index after max-min normalization. μ and σ are the mean and standard deviation of the above data, respectively. After processing with the above formula, all different types of source data are standardized by Z-scores to unify the data distribution.

[0087] Through the above processing, the value range and data distribution of different source data have been unified.

[0088] After completing the data spatial size unification and data distribution normalization processing, the consistency of spatial size and data distribution of different data can be achieved. Then, multi-source data fusion is performed. In a specific embodiment of the present invention, the method used for multi-source data fusion is as follows: by assigning different weights to different data sources, multi-source data fusion is achieved by using the band weighted summation method, as shown in the following formula (3):

[0089]

[0090] Where X1, X2, X3, X index These represent optical remote sensing data, SAR radar data, land cover type data, and remote sensing indices obtained through index analysis, respectively. a, b, c, and d are their respective proportional weights. The specific weight parameters of different data sources are adaptively determined by a learnable model. During training, this learnable model determines the importance and contribution of each data source to the soil moisture content prediction task based on the magnitude of the loss function and the optimization of the optimizer, and finally assigns the weights of a, b, c, and d.

[0091] Furthermore, in this embodiment, regularization with prior knowledge is applied to the weight parameters a, b, c, and d of different source data to artificially limit the range of contribution ratios of different data sources, thereby highlighting or limiting the importance of a certain type of data in soil water content prediction. As shown in equations (b) and (c) above, the regularization rules in this embodiment are:

[0092] First, we need to ensure that a+b+c+d=1, so that the sum of the weights of each source data is equal to 1;

[0093] Secondly, consider remote sensing index data (X indexX is calculated from optical remote sensing data (X1), therefore X is derived from X1. index The weight d of the remote sensing index should not exceed the weight a of X1 (optical remote sensing data) to ensure that the information in the original data makes a greater contribution.

[0094] Figure 3 A schematic diagram illustrating the adaptive weighted fusion process for multi-source data according to an embodiment of the present invention is shown. Figure 3 The diagram shows the data fusion process of one time-series node in the above embodiment (taking time T0 as an example). The same steps are then followed to process the data of other time-series nodes in the same way, and finally the overall time-series sample data X is obtained.

[0095] After preprocessing and fusing multi-source data, the processed data is further used to build a high-quality deep learning training sample library.

[0096] In the optimization process of deep learning-based supervised models, high-quality training samples are needed to train the model to obtain high-precision soil moisture content prediction results. The goal of this embodiment is to predict soil moisture content information at the next time step of a time series node based on temporal multi-source remote sensing data. Based on this task, the input samples in the sample library are processed multi-source temporal remote sensing data X, and the corresponding ground truth labels are the soil moisture content data at the next time step before and after the time series node. In creating the ground truth labels, considering the difficulty of collecting real soil moisture content data, a publicly available high-quality Chinese soil moisture / soil water dataset is used as the time series label data Y. series In addition, in this embodiment, each time-series input sample X, besides its time-series label, also has a category sample label Y. class In this embodiment, the correspondence between the input time-series sample data X and the two types of labels is as follows: Figure 4 As shown.

[0097] In constructing the deep learning training sample library, the temporal input data is first matched with soil moisture datasets representing the corresponding ground areas based on the latitude and longitude information in the images. Secondly, data from T0 to T... n The preprocessed data is arranged in a time series to form the time series sample input X; based on the time series input data (T0, T1, T2…T…), the time series sample input X is generated. n-1 The time sequence node information of the corresponding region is used to obtain the next time node (T) of the corresponding region. n Soil moisture data is used as the time series label. Additionally, the time series input data region corresponds to the time series time node (T0-T...). n-1The average precipitation meteorological data on the time series is used to construct category labels by setting a precipitation threshold. The corresponding time series sample data X is then divided into arid region samples, semi-arid samples, wet region samples, and humid region samples, thus constructing regional category label information. The precipitation threshold settings in this embodiment are shown in the table below:

[0098] Sample Threshold Classification Table

[0099] Classification mm / month drought sample <10 semi-arid samples 10–25 Moist sample 25–100 Moist sample 100>

[0100] Then, the Tth sample was obtained from the soil moisture content product corresponding to the sample area in the manner described above. n+1 The soil moisture content data at each time step is used as a time series label. All input samples are sequentially matched and aligned with the labels. The final time series input sample has a category label reflecting the average dryness / wetness at that time step, and a time series label reflecting the soil moisture content at the next time step. As shown in the diagram of input X and label Y, in this embodiment, the final training sample should include the time series input data X and the corresponding category label Y. class With time tag Y series Furthermore, a training sample library for the deep learning model is further constructed from several sample pairs.

[0101] After the sample library is constructed, data augmentation is performed based on the category label and time series label of the samples to further enrich the diversity of the samples in the sample library.

[0102] Since this embodiment has two types of labels: category labels and time-series labels, different data augmentation methods are required for different types of sample and label data. Specifically, for input data and category label data, this embodiment uses data augmentation methods commonly used in computer vision tasks, including random simulated illumination intensity, rotation, cropping, and contrast enhancement to expand the samples. For input sample and time-series label data pairs, methods such as time shifting, noise injection, and amplitude perturbation are used for data augmentation.

[0103] To achieve high-precision soil moisture content prediction based on multi-source remote sensing imagery, a hybrid deep learning model structure integrating a convolutional neural network (CNN) and an attention mechanism model (Transformer) is specifically designed and constructed in one specific embodiment of this invention for soil moisture content prediction. Specifically, the CNN extracts features from the input data, demonstrating its advantage in acquiring features from raster remote sensing images. The attention mechanism model then processes these features, capturing their temporal changes. Finally, the predicted soil moisture content time series is output via a feedforward neural network and a fully connected layer. Figure 5A fused convolutional neural network structure according to an embodiment of the present invention is shown.

[0104] like Figure 5 As shown, the input of the hybrid deep learning model for soil moisture content prediction is temporal multi-source remote sensing raster data (T0-Tn), and the final task is to perform temporal prediction of soil water content at time Tn+1. It includes a convolutional module responsible for raster feature extraction and a multi-head attention module responsible for temporal feature extraction.

[0105] The convolutional module responsible for raster feature extraction includes multiple weight-shared basic convolutional modules, consisting of convolutional layers, batch normalization layers, downsampling layers, and activation function layers. In specific applications, the configuration of different modules can be modified according to the application scenario. Modifiable aspects include the number of basic modules used, the number of convolutional channels in each module, and the downsampling ratio. In this embodiment, the convolutional module is responsible for extracting basic features of the input data, such as optical, texture, and shape features. Since the basic features of input samples from different time-series nodes have significant commonalities, a weight-sharing strategy is adopted to save training costs. Furthermore, for the differentiated features in the time-series input data, a multi-path feature processing branch module is constructed to capture the differentiated feature patterns. Finally, the feature fusion module achieves the fusion of multi-branch features to obtain the features extracted by the convolutional module. And using this feature and the sample's class loss label, the sample class loss loss1 is calculated using the cross-entropy loss function.

[0106] The features are then flattened and positionally encoded as input to the temporal attention module. The temporal attention module in this embodiment consists of multiple multi-head attention modules, and a fully connected layer maps the features to the output space.

[0107] In this hybrid deep learning model, the multi-head attention module is responsible for temporal feature analysis. The attention mechanism has a strong ability to capture global features and can handle long-distance feature dependencies in temporal input data, which is significant for temporal tasks with long input sequences. Furthermore, the multi-head attention module can capture long-distance feature dependencies under various patterns by adjusting the number of attention heads. Finally, a fully connected layer maps the features to the output space, and the temporal loss loss (loss2) between the output value and the sample temporal label is calculated using the mean squared error (MSE) function. Finally, the total model loss (loss) is calculated by summing loss1 and loss2.

[0108] After constructing the overall structure of the hybrid deep learning model, the model training process is carried out based on the constructed sample library. The Stochastic Gradient Descent (SGD) algorithm is used to calculate the gradient information of the overall model loss with respect to the network weight parameters. Mini-batch data randomly selected from the sample library is used to gradually optimize and update the model parameters through backpropagation, ultimately completing the training process of the deep learning model. Using the method of randomly selecting mini-batches can reduce the computational load and accelerate the training speed.

[0109] In this embodiment, when using the stochastic gradient descent (SGD) algorithm, considering that the learning rate scheduling strategy is one of the key factors affecting model convergence during the optimization training process, a multinomial decay learning rate scheduling strategy is adopted to adjust the learning rate. The specific formula is as follows:

[0110]

[0111] In the above formula, p is the degree of the polynomial, t represents the current training epoch, and t max For the maximum number of training rounds, η t η0 and η1 represent the learning rate for the current round and the initial learning rate, respectively. In this embodiment, a polynomial decay learning rate scheduling strategy helps the model converge quickly in the early stages of training and stabilizes at a better solution in the later stages of training.

[0112] In this embodiment, the design model is trained based on the constructed sample library using the above method. After the total loss of the model converges and stabilizes and no longer decreases, the optimal model weights are saved for subsequent prediction.

[0113] After the hybrid deep learning model is trained, the model weight parameters that achieve the best accuracy on the validation set during training are saved. This allows the model to be used to predict soil water content in unknown areas later.

[0114] In the subsequent soil water content prediction, the overall structure of the model is first initialized and the optimal model parameters obtained through model training are loaded, and the forward propagation mode of the model is started. The data to be predicted is first preprocessed and used as input to the hybrid deep learning model. After layer-by-layer processing by the hybrid deep learning model, the predicted soil water content is output.

[0115] Specifically, as an example, the specific steps for predicting soil water content in unknown areas based on the above hybrid deep learning model are as follows:

[0116] First, the original data of the area to be predicted is obtained, specifically the same as the input data during training, and preprocessed in the same way as during training. Second, a model structure consistent with that used in training is constructed, and the optimal model weight parameters obtained through training optimization are loaded. Finally, the preprocessed input to be predicted is used as the model input, fed into the model, and the forward propagation process is initiated. After model processing, the predicted soil water content of the area is obtained.

[0117] After obtaining information on regional soil moisture content through the above steps, this information can be used as an important indicator in conjunction with multi-source remote sensing data for the verification of potential geological hazards. First, InSAR technology is used to obtain regional surface deformation, and this surface deformation information is used to conduct a preliminary analysis of potential geological hazard points. Then, based on this, information such as spectral texture from optical imagery and corresponding soil moisture content information is used to further verify the risk level of the geological hazard points.

[0118] Specifically, as an example, the obtained regional soil moisture content information is used as an important indicator in conjunction with multi-source remote sensing data to verify potential geological hazards, including:

[0119] S501: Regional surface deformation is acquired using InSAR technology. Based on this surface deformation information, a preliminary analysis of potential soil moisture content is performed. Firstly, the soil moisture content of the target area is predicted using the above steps, and the predicted value of the soil moisture content at the next time step is denoted as S(t). i Since this embodiment takes into account not only the absolute value of soil water content, but also the strong correlation between the rate of change of soil water content and geological disaster risks, the soil water content change index at a certain time step is calculated by the following formula (5), namely the rate of change of soil water content and the second-order rate of change, which respectively reflect the speed of change of soil water content and the rate of change of the change rate at that time step.

[0120]

[0121] In Formula 5(a), V i S(t) is the rate of change of soil water content within a time step. i ) and S(t i-1 ) are respectively the tth i With t i-1 The soil moisture content at time t. This can be obtained using formula 5(a). i-1 , t i The rate of change of soil water content V over a period of time i Then, [t] is further calculated according to formula 5(b). i-1 , t i Indicator A: Rate of change of soil water content over a time period i .

[0122] Where V i With A i These two indicators respectively reflect the rate of change of soil moisture content over a given period and the speed of that change. They provide a more comprehensive description of the changes in soil moisture content within the region. When the rate of change of soil moisture content V... i When the value is large, or the rate of change changes drastically (i.e., A) i A larger value usually indicates a greater risk of geological disasters in the area. In subsequent parts of this disclosure, the absolute value of soil water content S(t) at time i, predicted by the model, will be fully utilized. i The rate of change of soil water content V i The change in the rate of change of soil water content, A i Further integrate with other remote sensing data to identify potential geological disaster sites.

[0123] After determining the rate of change and the second-order rate of change of soil moisture content, the magnitude of the geological disaster risk in the area is further determined based on these rates. Considering that the formation of geological disaster hazards is often accompanied by certain ground deformation, after determining the rate of change and the second-order rate of change of soil moisture content, it is also necessary to perform a secondary risk zone division between risky and low-risk areas within the area. Specifically, the deformation results of the target area can be obtained using InSAR technology, and the secondary risk zone division can be performed based on the deformation results information. Figure 6 The flowchart illustrating the process of obtaining deformation results of a target region using InSAR technology according to an embodiment of the present invention is shown, such as... Figure 6 As shown, the process of obtaining deformation results of a target area using InSAR technology includes:

[0124] First, SAR image data for the region is registered, and interferometric pairs are generated. Then, interferometric processing is performed on these pairs to obtain differential interferograms. These differential interferograms contain phase information of surface deformation. Subsequently, noise is filtered out using phase filtering, and then the wrapped phases are unwrapped. After unwrapping, the unwrapped phases are converted into true surface deformation. Finally, geocoding is performed to obtain the regional deformation map. The regional deformation map reflects the basic deformation situation within the region. Areas with a certain degree of deformation often have a greater risk of geological disasters, while areas with less deformation have a lower risk.

[0125] After dividing the potential hazard points in the region into risky and low-risk areas based on the deformation information obtained through InSAR technology, the potential hazard points in the region are further divided into high-risk, medium-high-risk, medium-risk, low-risk, and risk-free areas through secondary risk area division.

[0126] Furthermore, in one specific embodiment of the present invention, considering that the formation of geological disaster hazards often leads to certain changes in the vegetation around the hazard point, and that human engineering activities and regional soil types also affect the formation of geological disaster hazard points, a normalized vegetation index is further calculated using optical images to indicate regional vegetation changes. Additionally, regional human engineering activities and soil types are interpreted using optical images.

[0127] Ultimately, by analyzing the regional deformation map using decision tree analysis, the soil moisture content information and spectral texture characteristics of multi-source areas are used to identify potential geological disaster sites within the region.

[0128] Figure 7 The overall process for investigating potential geological disaster sites according to an embodiment of the present invention is illustrated. Figure 7 As shown, the overall process for geological disaster hazard investigation based on soil moisture content information is as follows:

[0129] First, regional deformation information is acquired. Since the formation of geological disaster hazards is often accompanied by ground deformation, hazard points within the region can be divided into high-risk and low-risk areas based on threshold values. Subsequently, considering the role of soil moisture content changes in the formation of geological disaster hazards, two different sets of threshold indicators are set for high-risk and low-risk areas to further classify them. Finally, through regional optical imagery, the Normalized Difference Vibration Index (NDVI) is calculated to identify soil types and human engineering activities, further investigating the regional geological disaster hazard level and obtaining the final results of the regional hazard point investigation.

[0130] In practical applications, it is first necessary to set thresholds a, (b1, b2, b3), and (c1, c2, c3) sequentially based on regional historical geology and hydrology, as well as experience. Threshold a is the regional deformation threshold; (b1, b2, b3) is a set of soil water content-related thresholds for risky areas, namely, the threshold for the absolute value of soil water content, the threshold for the rate of change of soil water content, and the threshold for the speed of change of the rate of change of soil water content, respectively. (c1, c2, c3) is a set of soil water content thresholds for risk-free areas, similarly, the threshold for the absolute value of soil water content, the threshold for the rate of change of soil water content, and the threshold for the speed of change of the rate of change of soil water content, respectively.

[0131] When the deformation value of a region exceeds the threshold 'a', it is divided into risky and low-risk regions. Further verification is then conducted based on the soil moisture content thresholds (b1, b2, b3) for risky regions and (c1, c2, c3) for low-risk regions. When judging risky region samples against the soil moisture content thresholds (b1, b2, b3), if all three indicators exceed the threshold, it is classified as a high-risk hazard point; if two indicators exceed the threshold, it is classified as a medium-high risk region; if any one indicator exceeds the threshold, it is classified as a medium-risk region; and if no indicator exceeds the threshold, it is classified as a medium-low risk region. Additionally, in low-risk regions, the same verification is performed using the soil moisture content thresholds (c1, c2, c3) for low-risk regions, successively classifying the region samples into medium-risk, medium-low-risk, low-risk, and no-risk regions.

[0132] Furthermore, this invention also considers that areas prone to geological disasters are often accompanied by changes in surface vegetation, and that the risk of geological disasters varies across different soil types. Moreover, human engineering activities can exacerbate the risk of geological disaster formation. Based on these factors, the invention uses spectral texture information provided by high-resolution optical images to calculate the normalized vegetation index, identifying factors related to the formation of geological disaster risks, such as soil type and traces of human engineering activities.

[0133] The above factors were used to further comprehensively verify and assess the medium- and high-risk areas identified in the classification. Specifically, if the NDVI index calculated multiple times showed significant changes, it indicated a greater risk of geological disasters. Similarly, if more than one instance of human engineering activity was identified, the risk level of geological disasters was upgraded accordingly. Finally, regional soil information was combined to achieve a comprehensive investigation of potential geological disaster sites in the region.

[0134] Corresponding to the above-mentioned method for investigating geological disaster hazards assisted by multi-source time-series remote sensing soil moisture content prediction technology, this invention also provides a geological disaster hazard investigation system assisted by multi-source time-series remote sensing soil moisture content prediction technology, used for investigating geological disaster hazards using the above method. The system includes:

[0135] The basic data acquisition unit is used to preprocess the acquired multi-source time-series remote sensing data associated with soil water content to obtain the basic data for the model.

[0136] A database construction unit is used to construct a model sample database based on the model's basic data.

[0137] The model training unit is used to optimize and train the pre-constructed deep learning model for predicting soil water content using the sample database, and to initialize the model with the optimal model parameters. During the training process, the model is trained with deep supervision using sample category information and supervised training using sample time series information. The randomly initialized parameters are optimized and adjusted, and the optimal model weights are saved on the validation dataset.

[0138] The prediction unit is used to initialize the model using the model parameters with the optimal weights, predict the soil water content of the next time step of the input multi-source time-series remote sensing data of the area to be predicted, and obtain the prediction result of the soil water content of the area.

[0139] The hazard investigation unit is used to determine the rate of change and the second-order rate of change of soil water content based on the predicted soil water content of the area; and to determine the magnitude of the geological disaster hazard risk in the area based on the rate of change and the second-order rate of change.

[0140] For the embodiments of the geological disaster hazard investigation system assisted by the multi-source time-series remote sensing soil moisture content prediction technology provided by the present invention, since they are basically similar to the embodiments of the geological disaster hazard investigation method assisted by the multi-source time-series remote sensing soil moisture content prediction technology, the relevant parts can be referred to the description of the method embodiment, and will not be repeated here.

[0141] As can be seen from the specific embodiments provided by the present invention, the method and system for investigating geological disaster hazards assisted by multi-source time-series remote sensing soil moisture content prediction technology provided by the present invention predicts soil moisture content by selecting multi-source time-series remote sensing data and pre-constructed deep learning models for soil moisture content prediction, obtains regional soil moisture content prediction results, and then determines the magnitude of geological disaster hazard risk in the region based on the obtained regional soil moisture content prediction results. This not only overcomes the shortcomings of high measurement cost and low efficiency of traditional soil moisture content monitoring schemes, but also allows for comprehensive analysis by combining regional soil moisture content prediction results, regional deformation obtained through InSAR technology, and regional spectral texture information provided by optical images, thereby improving the accuracy of geological disaster hazard investigation relative to single remote sensing data and realizing rapid investigation of regional geological disaster hazard points by using multi-source remote sensing data and deep learning technology.

[0142] The method and system for investigating potential geological hazards assisted by multi-source time-series remote sensing soil moisture content prediction technology according to the present invention have been described above by way of example with reference to the accompanying drawings. However, those skilled in the art should understand that various modifications can be made to the method and system for investigating potential geological hazards assisted by multi-source time-series remote sensing soil moisture content prediction technology proposed in the present invention without departing from the scope of the present invention. Therefore, the scope of protection of the present invention should be determined by the content of the appended claims.

Claims

1. A method for investigating potential geological hazards assisted by multi-source time-series remote sensing soil water content prediction technology, characterized in that, include: The acquired multi-source time-series remote sensing data associated with soil water content are preprocessed to obtain the basic data for the model. A model sample database is constructed based on the aforementioned model foundation data; This includes: performing time-series block pruning on the model's basic data, and mapping the pruned data to soil water content label samples one-to-one, so that each time-series input sample has a time-series label and a category label; and performing data augmentation on the time-series input samples with time-series labels and category labels. The pre-constructed deep learning model for predicting soil water content is optimized and trained using the sample database to obtain the model parameter initialization model with optimal weights. The model is composed of a hybrid convolutional neural network and an attention network. During the training process, the model is trained with deep supervision using sample category information and supervised training using sample time series information. The randomly initialized parameters are optimized and adjusted, and the model weights that perform best on the validation dataset are saved. The model is initialized using the optimal weighted model parameters, and the soil water content of the next time step of the input multi-source time-series remote sensing data of the area to be predicted is predicted to obtain the predicted soil water content of the area. Based on the predicted soil water content in the region, the rate of change and the second-order rate of change of soil water content are determined; based on the rate of change and the second-order rate of change, the magnitude of geological disaster risk in the region is determined. The input samples in the model sample database are processed multi-source time-series remote sensing data, and the corresponding ground truth labels are soil water content data at the next time step above and below the time-series node; the method for creating the ground truth labels includes: S201: Based on the latitude and longitude information in the image, perform regional matching between the temporal input data and the soil moisture dataset representing the corresponding ground area; S202: Based on the time-series input data (T0, T1, T2…Tn-1), obtain the soil moisture data of the next time node (Tn) of the corresponding area as the time-series label; S203: Based on the timing input data area timing time nodes (T0-T... n-1 The average precipitation meteorological data on the ) is used to divide the corresponding time series input samples into arid region samples, semi-arid samples, humid samples and humid region samples, and to construct regional category label information; Following the data processing methods of S201 to S203, all sample inputs and labels are matched and aligned sequentially. Finally, a time-series input sample has a category label information reflecting the average dryness and wetness at the time-series node of the sample, and a time-series label reflecting the soil water content at the next time step. During the training process of the deep learning model for predicting soil water content, The cross-entropy category loss Loss1 is calculated using the category label of the input sample and the output features of the convolutional neural network; and the temporal loss Loss2 is calculated using the temporal label of the input sample and the temporal features output by the attention network. The sum of Loss1 and Loss2 is used as the overall loss of the deep learning model for predicting soil water content. The gradient information of the overall loss of the deep learning model for predicting soil moisture content with respect to the network weight parameters is calculated using the stochastic gradient descent method. The network parameters are then gradually optimized using the backpropagation algorithm, and the training of the deep learning model for predicting soil moisture content is finally completed.

2. The method for investigating potential geological hazards assisted by multi-source time-series remote sensing soil water content prediction technology as described in claim 1, characterized in that, The multi-source time-series remote sensing data includes optical remote sensing time-series data, synthetic aperture radar time-series data, land cover time-series data for the corresponding area, and data reflecting soil moisture content.

3. The method for investigating geological disaster risks assisted by multi-source time-series remote sensing soil water content prediction technology as described in claim 2, characterized in that, The preprocessing includes geometric correction, radiometric correction, exponential analysis, sampling interpolation, normalization, and multi-source fusion.

4. The method for investigating geological disaster risks assisted by multi-source time-series remote sensing soil water content prediction technology as described in claim 3, characterized in that, The index analysis uses the normalized vegetation index (NVI) to analyze vegetation growth status, indirectly reflecting soil water content; the calculation formula for the NVI is as follows: in, NDVI The normalized vegetation index (NIR) represents the near-infrared band and the red band of the optical remote sensing image, respectively.

5. The method for investigating potential geological hazards assisted by multi-source time-series remote sensing soil water content prediction technology as described in claim 3, characterized in that, The normalization is achieved through Z-Score normalization, which unifies the value range and data distribution from different data sources; the Z-Score normalization formula is shown below: in, Z This represents normalized data. X The data to be normalized includes optical remote sensing data, radar remote sensing data, land cover data, and derived normalized vegetation index after maximum-minimum normalization. , These represent the mean and standard deviation of the data to be normalized, respectively.

6. The method for investigating potential geological hazards assisted by multi-source time-series remote sensing soil water content prediction technology as described in claim 5, characterized in that, The multi-source fusion is used to fuse the normalized data by weighted fusion and addition, as shown in the following formula: in, , , , These represent optical remote sensing data, SAR radar data, land cover type data, and remote sensing indices obtained through index analysis, respectively. , , , for , , , Each has its own proportional weight.

7. A geological disaster hazard investigation system assisted by multi-source time-series remote sensing soil water content prediction technology, characterized in that, A method for investigating geological disaster risks using the multi-source time-series remote sensing soil moisture content prediction technology as described in any one of claims 1 to 6, the system comprising: The basic data acquisition unit is used to preprocess the acquired multi-source time-series remote sensing data associated with soil water content to obtain the basic data for the model. A database construction unit is used to construct a model sample database based on the model's basic data. The model training unit is used to optimize and train the pre-constructed deep learning model for predicting soil water content using the sample database, and to initialize the model with the optimal model parameters. During the training process, the model is trained with deep supervision using sample category information and supervised training using sample time series information. The randomly initialized parameters are optimized and adjusted, and the optimal model weights are saved on the validation dataset. The prediction unit is used to initialize the model using the model parameters with the optimal weights, predict the soil water content of the next time step of the input multi-source time-series remote sensing data of the area to be predicted, and obtain the prediction result of the soil water content of the area. The hazard investigation unit is used to determine the rate of change and the second-order rate of change of soil water content based on the predicted soil water content of the area; and to determine the magnitude of the geological disaster hazard risk in the area based on the rate of change and the second-order rate of change.

Images