Landslide risk assessment method and system
By constructing a multi-source data-driven landslide risk assessment method and optimizing model parameters using risk attribute confidence and loss function, the limitations of data coverage and insufficient model adaptation in existing technologies are solved, achieving a more efficient and accurate landslide risk assessment.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-27
- Publication Date
- 2026-03-27
AI Technical Summary
Existing landslide risk assessment technologies suffer from limitations in data coverage, model adaptation deficiencies, and insufficient feature mining, making it difficult to accurately capture risk characteristics. Furthermore, the lack of cross-model collaborative optimization mechanisms leads to inaccurate assessment results and weak generalization performance, making it difficult to adapt to dynamically changing landslide risks.
By constructing a landslide risk assessment method, a second risk assessment model and a risk result generation model are configured using multi-source risk data. Feature extraction and result generation are performed, and model parameters are optimized by combining risk attribute confidence and loss function values. This achieves efficient model configuration and feature extraction, thereby improving assessment accuracy and generalization ability.
It improves the accuracy of landslide risk assessment and the generalization ability of the model, enabling it to better adapt to complex risk scenarios and dynamic changes, and providing accurate risk assessment support.
Smart Images
Figure CN121745667A_ABST
Abstract
Description
Technical Field
[0001] This disclosure relates to the technical field of landslide risk assessment, and in particular to a landslide risk assessment method and system. Background Technology
[0002] Landslides, a common geological hazard, are widely distributed in mountainous areas, mining areas, and engineering construction zones. Their occurrence is often accompanied by the sliding of large amounts of rock and soil within a short period, easily causing serious consequences such as casualties, infrastructure damage, and ecological destruction, posing a significant threat to regional safety production, residents' lives, and sustainable economic and social development. With the intensification of global climate change and the increase in the intensity of human engineering activities, the frequency, scope, and severity of landslide disasters are showing an increasing trend year by year. Therefore, there is an urgent need for accurate and efficient landslide risk assessment technologies to support disaster early warning, prevention and control decisions, and emergency management.
[0003] Current landslide risk assessment technologies mainly rely on traditional geological surveys, empirical model analysis, and single data-driven methods. However, they face several bottlenecks in practical applications: First, traditional assessment methods are mostly based on local geological data obtained from field surveys, which cannot fully cover the complex environmental conditions that give rise to landslides (such as topography, stratigraphy, hydrogeology, rainfall intensity, etc.). Insufficient data representativeness leads to limitations in assessment results. Second, existing risk assessment models are mostly fixed-structure statistical models or simple machine learning models, lacking the ability to adapt to the global distribution characteristics of landslide risk data. They are unable to capture the personalized risk patterns of different regions and types of landslides, resulting in weak model generalization performance. Third, landslide risk data has characteristics such as high dimensionality, nonlinearity, and dynamic changes. Traditional feature extraction methods are unable to effectively mine the key risk attributes hidden in the data, resulting in insufficient model adaptability to complex risk scenarios and assessment accuracy that fails to meet practical needs. Fourth, some assessment models rely on a large amount of labeled data for training. However, the collection of landslide disaster data is limited by factors such as terrain conditions, monitoring costs, and the suddenness of disasters. Labeled data is scarce and difficult to obtain, further restricting the improvement of model performance.
[0004] Furthermore, existing technologies lack effective cross-model collaborative optimization mechanisms in model configuration, making it difficult to improve the reliability of feature extraction through multi-model interactive verification. The risk result generation process also fails to adequately consider the consistency of global data distribution, resulting in risk assessment samples that do not accurately reflect actual landslide risk characteristics and cannot provide high-quality data support for model optimization. Simultaneously, the lack of fine-tuning mechanisms for specific application scenarios after model training means that assessment accuracy is prone to significant decline when facing new risk data or complex conditions, making it difficult to adapt to dynamically changing landslide risk prevention and control needs.
[0005] Therefore, how to overcome the limitations of traditional technologies in terms of data coverage, model adaptation defects, and insufficient feature mining, and to construct landslide risk assessment methods and models that can fully utilize multi-source risk data, accurately capture risk characteristics, and have strong generalization capabilities, has become a core technical challenge that urgently needs to be solved in the field of landslide disaster prevention and control. It is also the key to promoting the transformation of landslide risk assessment from "experience-driven" to "data intelligence-driven". Summary of the Invention
[0006] To address the technical problems existing in related technologies, this disclosure provides a landslide risk assessment method and system.
[0007] A landslide risk assessment method, the method comprising: The system acquires a first risk data example collected in the landslide assessment instruction, a first risk assessment model to be configured in the model configuration instruction, and a second risk assessment model and a risk result generation model configured based on the first risk data example. The second risk assessment model is used for feature extraction, and the risk result generation model is used to generate new risk assessment results, and the new risk assessment results satisfy the global distribution of the first risk data example. Based on the configured second risk assessment model and the first risk assessment model to be configured, feature extraction is performed on the new risk assessment results generated by the risk result generation model to obtain the first risk attribute features and the second risk attribute features. Based on the first risk attribute characteristics and the second risk attribute characteristics, the first risk assessment model to be configured is configured to obtain the configured first risk assessment model; Configure the risk outcome generation model according to the following steps: Obtain the first risk attribute based on the output of the risk outcome verification model; the risk outcome verification model is used to generate elements that decompose the first risk data example into multiple primitives; The first risk attribute is loaded into the risk outcome generation model to be configured, and the second risk attribute is obtained from the output of the risk outcome generation model. Based on the risk attribute confidence between the second risk attribute and the first risk attribute, the loss function value of the risk outcome generation model to be configured is determined; The risk outcome generation model to be configured is configured based on the loss function value, resulting in a configured risk outcome generation model.
[0008] In one standalone implementation, loading a first risk attribute into a risk outcome generation model to be configured includes: The local risk attribute region in the first risk attribute is hidden to obtain the first risk attribute after hiding. Load the hidden first risk attribute into the risk outcome generation model to be configured.
[0009] In one standalone implementation, the risk outcome verification model includes a compression unit and a derivation unit, and is configured according to the following steps: Repeat the following steps until the confidence level between the risk attribute output by the derivative unit and the first risk data example input into the compression unit is greater than a preset threshold: Input the first risk data example into the compression unit to be configured to obtain the elements output by the compression unit; input the elements output by the compression unit into the derivative unit to be configured to obtain the risk attributes output by the derivative unit.
[0010] In a standalone implementation, the first risk attribute based on the output of the risk outcome verification model is obtained by following these steps: Input the first risk data example into the compression unit included in the risk result verification model to obtain the elements output by the compression unit; The elements output by the compressed unit are input into the derived units included in the risk outcome verification model to obtain the first risk attribute output by the derived unit.
[0011] In one standalone embodiment, the risk outcome generation model includes a first risk outcome generation sub-model for generating elements that decompose a first risk data example into multiple primitives, and a second risk outcome generation sub-model for generating new risk assessment results based on the risk attributes output by the first risk outcome generation sub-model; the risk outcome generation model is configured according to the following steps: Load the first risk data example into the configured first risk result generation sub-model to obtain the first risk attribute output by the first risk result generation sub-model; The first risk attribute is loaded into the second risk result generation sub-model to be configured, and the second risk attribute output by the second risk result generation sub-model is obtained. Based on the confidence level of the first risk attribute between the first risk attribute and the input first risk data example, and the confidence level of the second risk attribute between the second risk attribute and the first risk attribute, the loss function value of the risk outcome generation model to be configured is determined; The risk outcome generation model to be configured is configured based on the loss function value, resulting in a configured risk outcome generation model.
[0012] In a standalone implementation, after obtaining the configured first risk assessment model, the method further includes: Example of obtaining the second risk data collected in the model configuration command; Based on the second risk data example, the configured first risk assessment model is reconfigured to obtain the final configured first risk assessment model.
[0013] In one standalone implementation, the configured first risk assessment model is reconfigured based on a second risk data example to obtain a final configured first risk assessment model, including: The second risk data example is loaded into the first risk assessment model to obtain the model's calculation results; Based on the comparison between the calculation results and the task labeling results for the second risk data example, the loss function value of the first risk assessment model is determined; Based on the loss function value, the first risk assessment model is reconfigured to obtain the final configured first risk assessment model.
[0014] In a standalone implementation, the second risk assessment model is configured according to the following steps: Obtain an initial risk assessment model; the initial risk assessment model should include at least a feature extraction layer. Based on the feature extraction layer included in the initial risk assessment model, feature extraction is performed on the first risk data example to obtain the risk attribute feature information output by the feature extraction layer; The model parameter values of the feature extraction layer are optimized based on the risk attribute feature information to obtain an optimized feature extraction layer. The initial risk assessment model, which includes an optimized feature extraction layer, is identified as the configured second risk assessment model.
[0015] The technical solutions provided by the embodiments of this disclosure may include the following beneficial effects.
[0016] A landslide risk assessment method and system are disclosed. The method involves acquiring a first risk data example collected in a landslide assessment instruction, a first risk assessment model to be configured in a model configuration instruction, and a second risk assessment model and a risk result generation model configured based on the first risk data example. The second risk assessment model is used for feature extraction, and the risk result generation model is used to generate new risk assessment results, which satisfy the global distribution of the first risk data example. Features are extracted from the new risk assessment results generated based on the risk result generation model according to the configured second risk assessment model and the first risk assessment model to be configured, respectively, to obtain first risk attribute features and second risk attribute features. Based on the first risk attribute features and the second risk attribute features, the first risk assessment model to be configured is configured to obtain a configured first risk assessment model, which improves the accuracy of risk assessment.
[0017] It should be understood that the above general description and the following detailed description are merely exemplary and do not limit this disclosure. Attached Figure Description
[0018] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the specification, serve to explain the principles of this application.
[0019] Figure 1 A flowchart illustrating a landslide risk assessment method provided in this application embodiment; Detailed Implementation
[0020] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.
[0021] Based on the above, please refer to the following: Figure 1 This is a flowchart illustrating a landslide risk assessment method provided in an embodiment of this application. This landslide risk assessment method can be applied to... Figure 1 Furthermore, a landslide risk assessment method may specifically include the content described in the following steps S101-S103.
[0022] S101: Obtain the first risk data example collected in the landslide assessment instruction, the first risk assessment model to be configured in the model configuration instruction, and the second risk assessment model and risk result generation model configured based on the first risk data example; the second risk assessment model is used for feature extraction, and the risk result generation model is used to generate new risk assessment results, and the new risk assessment results satisfy the global distribution of the first risk data example; S102: Based on the configured second risk assessment model and the first risk assessment model to be configured, feature extraction is performed on the new risk assessment result generated by the risk result generation model to obtain the first risk attribute feature and the second risk attribute feature. S103: Configure the first risk assessment model to be configured based on the first risk attribute characteristics and the second risk attribute characteristics to obtain the configured first risk assessment model.
[0023] In this embodiment of the disclosure, the pre-configured model can be a second risk assessment model configured using the first risk data example collected in the landslide assessment instruction. Additionally, the risk result generation model can be a relevant model that generates new risk assessment results that satisfy the global distribution of the first risk data example.
[0024] Firstly, the risk outcome generation model can be configured according to the following steps in the embodiments disclosed herein: Step 1: Obtain the first risk attribute based on the output of the risk outcome verification model; the risk outcome verification model is used to generate elements that decompose the first risk data example into multiple primitives; Step 2: Load the first risk attribute into the risk outcome generation model to be configured, and obtain the second risk attribute output by the risk outcome generation model; Step 3: Based on the risk attribute confidence between the second risk attribute and the first risk attribute, determine the loss function value of the risk outcome generation model to be configured; Step 4: Configure the risk outcome generation model based on the loss function value to obtain the configured risk outcome generation model.
[0025] Here, the first risk attribute output by the risk outcome verification model can be used as the input risk attribute of the risk outcome generation model to be configured. Then, the loss function value of the risk outcome generation model is determined based on the risk attribute confidence between the second risk attribute output by the risk outcome generation model and the first risk attribute.
[0026] To better configure the risk outcome generation model, before loading the first risk attribute into the risk outcome generation model to be configured, the local risk attribute regions in the first risk attribute can be hidden to obtain the hidden first risk attribute. When the hidden first risk attribute is input into the risk outcome generation model to be configured, the unhidden local risk attribute regions can guide the generation of hidden local risk attribute regions. In this way, the model configuration can be achieved based on the closeness between the generated risk attribute and the initial first risk attribute.
[0027] In this embodiment of the disclosure, the aforementioned risk outcome verification model can also be configured based on the first risk data example. The main purpose of this risk outcome verification model is to configure an element that can encode visual features in the upstream data. Then, risk attributes can be restored using multiple primitives covered by the element generated in the risk outcome verification model, thereby obtaining the first risk attribute output by the risk outcome verification model.
[0028] The following section will detail the configuration and application process of the risk outcome verification model.
[0029] In this embodiment of the disclosure, an adversarial model composed of paired compression units and derivative units can be used to configure a risk outcome verification model. Here, a first risk data example can be input into the compression unit to be configured to obtain the elements output by the compression unit; the elements output by the compression unit can be input into the derivative unit to be configured to obtain the risk attributes output by the derivative unit; then, it is verified whether the confidence level between the risk attributes output by the derivative unit and the first risk data example input into the compression unit is greater than a preset threshold. If it is not greater than the preset threshold, the process of inputting the first risk data example into the compression unit to be configured is repeated until the confidence levels of the two risk attributes are greater than the preset threshold.
[0030] It should be noted that, in practical applications, the process of determining the first risk attribute can be determined by the configuration process of the joint risk result verification model. That is, the steps of inputting the first risk data example into the compression unit to be configured, obtaining the elements output by the compression unit, and inputting the elements output by the compression unit into the derivative unit to be configured, obtaining the risk attribute output by the derivative unit, can be repeatedly executed until the confidence level between the risk attribute output by the derivative unit and the first risk data example input into the compression unit is greater than a preset threshold. Then, the risk attribute output by the derivative unit is determined as the first risk attribute.
[0031] Secondly, in the case where the risk outcome generation model includes a first risk outcome generation sub-model for generating elements that decompose the first risk data example into multiple primitives, and a second risk outcome generation sub-model for generating new risk assessment results based on the risk attributes output by the first risk outcome generation sub-model, the risk outcome generation model can be configured according to the following steps in the embodiments of this disclosure: Step 1: Load the first risk data example into the configured first risk result generation sub-model to obtain the first risk attribute output by the first risk result generation sub-model; Step 2: Load the first risk attribute into the second risk result generation sub-model to be configured, and obtain the second risk attribute output by the second risk result generation sub-model; Step 3: Based on the confidence level of the first risk attribute between the first risk attribute and the input first risk data example, and the confidence level of the second risk attribute between the second risk attribute and the first risk attribute, determine the loss function value of the risk outcome generation model to be configured; Step 4: Configure the risk outcome generation model based on the loss function value to obtain the configured risk outcome generation model.
[0032] Here, the loss function value of the risk outcome generation model to be configured can be determined by combining the confidence level of the first risk attribute between the first risk attribute and the input first risk data example, and the confidence level of the second risk attribute between the second risk attribute and the first risk attribute. Both the confidence level of the first risk attribute and the confidence level of the second risk attribute will affect the optimization of the relevant model parameter values. That is, this realizes the synchronous configuration of the first risk outcome generation sub-model and the second risk outcome generation sub-model, which is more efficient.
[0033] With the configuration including the first risk result generation sub-model and the second risk result generation sub-model, a new risk assessment result can be generated for any input first risk data example. This new risk assessment result contains rich data information of landslide assessment instructions, thus better adapting to the configuration requirements of the model.
[0034] In this embodiment of the disclosure, the configuration of the first risk assessment model in the model configuration instruction can be guided by the risk attribute confidence degree between two risk attribute features (i.e., the first risk attribute feature and the second risk attribute feature) extracted from the configured second risk assessment model and the first risk assessment model to be configured. Specifically, this can be achieved through the following steps: Step 1: Based on the risk attribute confidence level between the first risk attribute feature and the second risk attribute feature, determine the loss function value of the first risk assessment model to be configured; Step 2: If the loss function value in the current round is greater than the preset threshold, optimize the model parameter values of the first risk assessment model based on the loss function value, and perform the next round of configuration based on the optimized first risk assessment model until the loss function value is less than or equal to the preset threshold.
[0035] To further extend the generalization performance of the first risk assessment model in the domain of model configuration instructions, the first risk assessment model can be fine-tuned using the second risk data example collected in the model configuration instructions. This can be achieved through the following steps: Step 1: Load the second risk data example into the first risk assessment model to obtain the model's calculation results; Step 2: Based on the comparison between the calculation results and the task labeling results for the second risk data example, determine the loss function value of the first risk assessment model; Step 3: Based on the loss function value, reconfigure the first risk assessment model to obtain the final configured first risk assessment model.
[0036] Here, feature extraction can be performed through the feature extraction layer included in the first risk assessment model. When the feature information output by the feature extraction layer is input into the task layer included in the first risk assessment model, the first risk assessment model can be configured in multiple rounds based on the calculation results and the matching results of the task labeling results for the second risk data example.
[0037] Based on the risk assessment model configuration method provided in the embodiments of this disclosure, the embodiments of this disclosure also provide a target detection method, specifically including the following steps: S201: Obtain the target risk attributes collected in the model configuration instructions; S202: Load the target risk attribute into the first risk assessment model configured using the risk assessment model configuration method, and obtain the detection result of the target object in the target risk attribute.
[0038] It should be understood that this application is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.
Claims
1. A landslide risk assessment method, characterized in that, The methods include: Obtain the first risk data example collected in the landslide assessment instruction, the first risk assessment model to be configured in the model configuration instruction, and the second risk assessment model and risk result generation model configured based on the first risk data example; The second risk assessment model is used for feature extraction, and the risk result generation model is used to generate new risk assessment results, and the new risk assessment results satisfy the global distribution of the first risk data example. Based on the configured second risk assessment model and the first risk assessment model to be configured, feature extraction is performed on the new risk assessment results generated by the risk result generation model to obtain the first risk attribute features and the second risk attribute features. Based on the first risk attribute characteristics and the second risk attribute characteristics, the first risk assessment model to be configured is configured to obtain the configured first risk assessment model; Configure the risk outcome generation model according to the following steps: Obtain the first risk attribute based on the output of the risk outcome validation model; The risk outcome validation model is used to generate elements that break down the first risk data example into multiple primitives; The first risk attribute is loaded into the risk outcome generation model to be configured, and the second risk attribute is obtained from the output of the risk outcome generation model. Based on the risk attribute confidence between the second risk attribute and the first risk attribute, the loss function value of the risk outcome generation model to be configured is determined; The risk outcome generation model to be configured is configured based on the loss function value, resulting in a configured risk outcome generation model.
2. The method according to claim 1, characterized in that, Load the first risk attribute into the risk outcome generation model to be configured, including: The local risk attribute region in the first risk attribute is hidden to obtain the first risk attribute after hiding. Load the hidden first risk attribute into the risk outcome generation model to be configured.
3. The method according to claim 1, characterized in that, The risk outcome verification model includes compression units and derivation units. Configure the risk outcome verification model according to the following steps: Repeat the following steps until the confidence level between the risk attribute output by the derivative unit and the first risk data example input into the compression unit is greater than a preset threshold: Input the first risk data example into the compression unit to be configured to obtain the elements output by the compression unit; input the elements output by the compression unit into the derivative unit to be configured to obtain the risk attributes output by the derivative unit.
4. The method according to claim 3, characterized in that, To obtain the first risk attribute based on the output of the risk outcome validation model, follow these steps: Input the first risk data example into the compression unit included in the risk result verification model to obtain the elements output by the compression unit; The elements output by the compressed unit are input into the derived units included in the risk outcome verification model to obtain the first risk attribute output by the derived unit.
5. The method according to claim 1, characterized in that, The risk outcome generation model includes a first risk outcome generation sub-model for generating elements that decompose the first risk data example into multiple primitives, and a second risk outcome generation sub-model for generating new risk assessment results based on the risk attributes output by the first risk outcome generation sub-model. Configure the risk outcome generation model according to the following steps: Load the first risk data example into the configured first risk result generation sub-model to obtain the first risk attribute output by the first risk result generation sub-model; The first risk attribute is loaded into the second risk result generation sub-model to be configured, and the second risk attribute output by the second risk result generation sub-model is obtained. Based on the confidence level of the first risk attribute between the first risk attribute and the input first risk data example, and the confidence level of the second risk attribute between the second risk attribute and the first risk attribute, the loss function value of the risk outcome generation model to be configured is determined; The risk outcome generation model to be configured is configured based on the loss function value, resulting in a configured risk outcome generation model.
6. The method according to any one of claims 1 to 5, characterized in that, After obtaining the configured first risk assessment model, the method also includes: Example of obtaining the second risk data collected in the model configuration command; Based on the second risk data example, the configured first risk assessment model is reconfigured to obtain the final configured first risk assessment model.
7. The method according to claim 6, characterized in that, Based on the second risk data example, the configured first risk assessment model is reconfigured to obtain the final configured first risk assessment model, including: The second risk data example is loaded into the first risk assessment model to obtain the model's calculation results; Based on the comparison between the calculation results and the task labeling results for the second risk data example, the loss function value of the first risk assessment model is determined; Based on the loss function value, the first risk assessment model is reconfigured to obtain the final configured first risk assessment model.
8. The method according to any one of claims 1 to 5, characterized in that, Configure the second risk assessment model according to the following steps: Obtain an initial risk assessment model; the initial risk assessment model should include at least a feature extraction layer. Based on the feature extraction layer included in the initial risk assessment model, feature extraction is performed on the first risk data example to obtain the risk attribute feature information output by the feature extraction layer; The model parameter values of the feature extraction layer are optimized based on the risk attribute feature information to obtain an optimized feature extraction layer. The initial risk assessment model, which includes an optimized feature extraction layer, is identified as the configured second risk assessment model.
9. A landslide risk assessment system, characterized in that, It includes a processor and a memory that communicate with each other, the processor being used to read a computer program from the memory and execute it to implement the method of any one of claims 1-8.