Geological disaster monitoring method and system based on social cost loss optimization

Abstract

The application provides a geological disaster monitoring method and system based on social cost loss optimization, and belongs to the technical field of disaster monitoring. The monitoring method is applied to a monitoring system including a sample analysis module, a model construction module and a disaster monitoring module. Specifically, a disaster and non-disaster area ratio of a target region is calculated based on historical remote sensing data, a unit social cost loss ratio of disaster identification deviation is set, a social cost loss evaluation index is constructed, an optimal sample ratio is solved with the index minimized as a target, a corresponding learning sample set is constructed, a preset geological disaster evaluation model is trained and learned, and a geological disaster probability is obtained according to the trained model. The social cost loss evaluation index is set through the disaster and non-disaster area ratio and the social cost loss, the optimal sample ratio is determined with the value being the lowest as an optimization target, the model training and output result can be closer to the actual disaster prevention demand, and the disaster prevention and control effect is improved.

Description

Technical Field

[0001] This invention relates to the field of geological disaster monitoring technology, and in particular to a geological disaster monitoring method and system optimized based on social cost loss. Background Technology

[0002] In the process of geological hazard susceptibility assessment, sample selection is the core link in model training. Its rationality directly determines the accuracy and practicality of the assessment model, and thus affects subsequent emergency rescue and disaster prevention resource allocation. Currently, the sample ratios used in geological hazard susceptibility assessment methods mainly rely on three selection methods: First, balanced sampling of hazard and non-hazard samples at the same ratio. To improve sample quality, optimization is achieved through multiple random samplings and selecting non-hazard samples outside the hazard buffer zone to reduce interference from potential hazard factors in non-hazard samples. Second, sampling is based on the actual area ratio of hazard to non-hazard within the target area. By restoring the true spatial proportion of hazard distribution within the area, the representativeness of the sample is improved. Third, comparative experiments are conducted with multiple groups of different sample ratios. Statistical indicators such as the overall model accuracy and F1 score are used as evaluation criteria, and the ratio with the best performance is selected as the final sample ratio. The common goal of these three methods is to improve the statistical fitting accuracy of the model, essentially relying on the fitting ability of machine learning or deep learning models to sample features for assessment.

[0003] Traditional sample selection methods, such as balanced sampling or sampling based on actual area, are prone to over-identifying or under-identifying disasters. In the actual implementation of geological disaster emergency rescue and disaster prevention and mitigation, under-identification of disasters may lead to the failure of emergency rescue plans and losses, while misidentification of disasters may result in the over-deployment of disaster prevention materials and waste of manpower. The severity of the consequences of these two types of identification biases is drastically different. However, traditional sample selection methods mostly treat the impact of these two types of disaster identification biases equally, failing to reflect this difference in sample ratio optimization. This can easily lead to the misallocation of subsequent disaster prevention resources, affecting disaster prevention efficiency. Furthermore, using model statistical indicators as the optimization target for sample ratios can easily lead to a disconnect between subsequent model training results and actual needs. This results in evaluation results that meet theoretical accuracy but lack practicality, failing to meet actual disaster prevention requirements. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of neglecting the differences in social costs of missed and incorrect disaster identification when selecting samples for disaster assessment, the optimization objectives being out of touch with actual needs, and the poor disaster prevention and control effect. This invention provides a geological disaster monitoring method and system based on social cost loss optimization. By quantifying the social cost loss weights of the ratio of disaster to non-disaster area and disaster identification bias, and using the lowest social cost loss assessment index as the optimization objective, the optimal sample ratio is determined. This makes the geological disaster probability assessment results more aligned with the actual needs of emergency rescue and disaster prevention and mitigation, reduces the overall social cost loss in the region, and improves the disaster prevention and control effect.

[0005] The objective of this invention is achieved through the following technical solution: Geological hazard monitoring methods optimized based on social cost loss include: The ratio of disaster-affected to non-disaster-affected areas in the target region is calculated based on historical remote sensing data, and the ratio of unit social cost loss for disaster identification bias is set. Set up a social cost loss assessment index for disaster identification bias, and solve for the optimal sample ratio with the goal of minimizing the social cost loss assessment index; A learning sample set is constructed based on the optimal sample ratio. The preset geological hazard assessment model is trained using the learning sample set, and the probability of geological hazards is obtained based on the trained geological hazard assessment model.

[0006] Furthermore, the calculation of the disaster-to-non-disaster area ratio of the target region based on historical remote sensing data includes: Based on historical remote sensing data, the boundaries between disaster-prone and non-disaster-prone areas within the target region are identified, and the boundary identification results are corrected by combining the corresponding survey data; The target area is divided into several pixel units of equal area, and each pixel unit is divided into a disaster grid or a non-disaster grid based on the corrected boundary. Count the number of disaster grids and non-disaster grids, and calculate the total area of ​​the disaster area and the total area of ​​the non-disaster area by combining the area of ​​the corresponding pixel unit; By comparing the total area of ​​the disaster area with the total area of ​​the non-disaster area, the ratio of the disaster area to the non-disaster area of ​​the target area can be obtained.

[0007] Furthermore, the disaster identification bias includes both missed disaster identification and incorrect disaster identification.

[0008] Furthermore, the ratio of unit social cost loss for setting disaster identification bias includes: Based on the basic distribution information of the target area, the severity of the consequences of missed disaster reporting and incorrect disaster reporting are calculated respectively; The relationship between the unit social cost loss multiples for disaster underreporting and disaster misreporting is calculated based on the severity of the corresponding consequences; The unit social cost loss ratio is set based on the relationship between the unit social cost loss multiple of disaster underreporting and disaster misreporting.

[0009] Furthermore, the social cost loss assessment index for setting disaster identification bias includes: The probability of missing disasters is defined by the model recall metric, and the probability of false disasters is defined by the model specificity metric. By combining the ratio of unit social cost loss and the ratio of disaster area to non-disaster area, a coupling relationship between the probability of disaster omission and the probability of disaster misidentification is established, and a social cost loss assessment index for disaster identification bias is set.

[0010] Furthermore, after determining the optimal sample proportion with the goal of minimizing the social cost loss assessment index, the following steps are also performed: Several sets of sample ratios are constructed based on the optimal sample ratio and its adjacent sample ratios, and corresponding training sets and test sets are constructed by combining historical disaster data of the target area. The pre-set geological disaster assessment model is trained using training and testing sets for each sample proportion to obtain the social cost loss assessment index value for each sample proportion. The optimal sample proportion is updated based on the social cost loss assessment index value of each sample proportion.

[0011] Furthermore, updating the optimal sample proportion based on the social cost loss assessment index value of each sample proportion includes: When the social cost loss assessment index value of the adjacent sample ratio is less than the current optimal sample ratio, the optimal sample ratio is updated according to the adjacent sample ratio. Reduce the refinement gradient, select neighboring samples of the updated optimal sample ratio based on the reduced refinement gradient, recalculate the corresponding social cost loss assessment index value, update the optimal sample ratio, until the corresponding update termination condition is met.

[0012] Furthermore, the update termination conditions include: The social cost loss assessment index values ​​of adjacent sample proportions are all greater than the current optimal sample proportion; Alternatively, the number of times the decrease in the social cost loss assessment index value is less than the preset threshold reaches the predetermined number threshold. Alternatively, refine the gradient to a preset gradient limit.

[0013] Furthermore, the step of constructing a learning sample set based on the optimal sample ratio, training the preset geological hazard assessment model using the learning sample set, and obtaining the geological hazard probability based on the trained geological hazard assessment model includes: Based on historical disaster data, disaster samples are constructed. Non-disaster samples are randomly selected from non-disaster areas of the target region according to the optimal sample ratio. A learning sample set is constructed based on the disaster samples and non-disaster samples. The training set and test set are divided according to a preset ratio, and the preset geological hazard assessment model is trained and learned. Obtain the current remote sensing data of the target area, input the current remote sensing data into the trained geological hazard assessment model, and obtain the geological hazard probability of each pixel unit in the target area.

[0014] A geological hazard monitoring system based on social cost loss optimization, used to perform any of the above-mentioned geological hazard monitoring methods, includes: The sample analysis module is used to calculate the ratio of disaster-affected to non-disaster-affected areas of the target region based on historical remote sensing data, set the unit social cost loss ratio of disaster identification deviation, and set the social cost loss assessment index of disaster identification deviation. With the goal of minimizing the social cost loss assessment index, the optimal sample ratio is solved. The model building module is used to build a learning sample set according to the optimal sample ratio, and to train the preset geological disaster assessment model using the learning sample set; The disaster monitoring module is used to obtain the probability of geological disasters based on the trained geological disaster assessment model.

[0015] The beneficial effects of this invention are: By combining historical remote sensing data with survey data to correct boundaries and divide the data into pixel units, the ratio of disaster-affected to non-disaster-affected areas is quantified, ensuring the accuracy of the quantification results. Furthermore, based on the actual regional situation, the difference in unit social cost loss between missed and incorrect disaster assessments is quantified, avoiding the irrationality of the traditional cost equivalence assumption. Simultaneously, with the goal of minimizing social cost loss, the sample ratio is dynamically iteratively optimized to ensure a balanced and efficient learning sample set. This enables the trained model to output pixel-level geological disaster probabilities, reducing rescue delays caused by missed assessments and resource waste caused by incorrect assessments, while providing precise spatial guidance for emergency disaster prevention. Ultimately, this comprehensively reduces the overall social cost loss in the region, achieving a high degree of matching between geological disaster monitoring and actual needs. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of a process of the present invention; Figure 2 This is a schematic diagram of a structure according to an embodiment of the present invention.

[0017] The modules are: 1. Sample analysis module, 2. Model building module, and 3. Disaster monitoring module. Detailed Implementation

[0018] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0019] Example: A geological hazard monitoring method optimized based on social cost loss, such as... Figure 1 As shown, it includes: The ratio of disaster-affected to non-disaster-affected areas in the target region is calculated based on historical remote sensing data, and the ratio of unit social cost loss for disaster identification bias is set. Set up a social cost loss assessment index for disaster identification bias, and solve for the optimal sample ratio with the goal of minimizing the social cost loss assessment index; A learning sample set is constructed based on the optimal sample ratio. The preset geological hazard assessment model is trained using the learning sample set, and the probability of geological hazards is obtained based on the trained geological hazard assessment model.

[0020] Geological disaster monitoring relies on model training to assess disaster risk, and the representativeness of the sample set is fundamental to model performance. Without a reference ratio of disaster-to-non-disaster areas, sample selection can easily fall into the trap of subjective equilibrium or random sampling, leading to a disconnect between sample distribution and the actual spatial characteristics of disasters in the region. This can cause the model to become overly biased towards certain sample characteristics, resulting in subsequent identification biases. Only by using the actual area proportion as a benchmark can the constructed sample set objectively reflect the spatial weights of the two types of areas, providing the model with a training basis that fits the actual regional situation and avoiding model performance deviations caused by sample imbalance.

[0021] Historical remote sensing data has the advantages of full coverage, spatiotemporal traceability, and the ability to reflect the macroscopic distribution of the Earth's surface. It can not only fully cover the target area, but also trace the surface state after historical disasters through multi-time series data, accurately identify the boundary between disaster and non-disaster areas, and avoid the loss or deviation of regional distribution information due to the limitations of local field surveys.

[0022] Therefore, historical remote sensing data is used to calculate the ratio of disaster-affected to non-disaster-affected areas in the target region, and survey data is further introduced for correction to ensure the accuracy of the area ratio calculation.

[0023] The calculation of the ratio of disaster-affected to non-disaster-affected areas of the target region based on historical remote sensing data includes: Based on historical remote sensing data, the boundaries between disaster-prone and non-disaster-prone areas within the target region are identified, and the boundary identification results are corrected by combining the corresponding survey data; The target area is divided into several pixel units of equal area, and each pixel unit is divided into a disaster grid or a non-disaster grid based on the corrected boundary. Count the number of disaster grids and non-disaster grids, and calculate the total area of ​​the disaster area and the total area of ​​the non-disaster area by combining the area of ​​the corresponding pixel unit; By comparing the total area of ​​the disaster area with the total area of ​​the non-disaster area, the ratio of the disaster area to the non-disaster area of ​​the target area can be obtained.

[0024] When identifying regional boundaries based on historical remote sensing data, the corresponding historical remote sensing data is first selected according to the spatial range of the target area and preprocessed. Interference factors such as atmospheric scattering and illumination differences are eliminated through radiometric correction to ensure that the surface feature information is accurately reflected. Then, geometric correction is used to align the data with the standard geographic coordinate system to avoid positional deviations affecting boundary positioning.

[0025] Based on preprocessed historical remote sensing data, and considering the differences in spectral and topographic features between disaster-affected and non-disaster-affected areas, edge detection and other algorithms are used to extract the boundaries between disaster-affected and non-disaster-affected areas within the target region. Simultaneously, field survey data is incorporated to compare the identified boundaries with the actual disaster distribution, allowing for adjustments to any discrepancies and ensuring the accuracy of the final disaster-affected and non-disaster-affected area boundaries.

[0026] Then, using the target area boundary as the scope, the target area is transformed into a set of regular pixel units, and a unified coordinate system is established. Next, the disaster attribute of each pixel unit is determined by calculating the coverage ratio of the corrected disaster boundary within each pixel unit. If the coverage ratio of the disaster boundary exceeds the corresponding threshold, it is judged as a disaster raster; otherwise, it is judged as a non-disaster raster. If there are cases where the edge of a pixel unit crosses the disaster boundary, the corresponding auxiliary image verification unit attributes are further considered.

[0027] Finally, by using the number of grid cells and the area of ​​the corresponding pixel unit, the total area of ​​the disaster area and the non-disaster area is calculated. The ratio of the total area of ​​the disaster area to the non-disaster area of ​​the target area is then determined by the ratio of the total area of ​​the disaster area to the non-disaster area.

[0028] The disaster identification biases described in this embodiment include both missed disaster identification and incorrect disaster identification.

[0029] Disaster omission occurs when the model fails to identify actual disaster areas, potentially leading to inadequate emergency response and irreversible losses such as building damage and ecological destruction. Disaster misidentification, on the other hand, occurs when the model misclassifies non-disaster areas as disaster zones, resulting in misallocation of disaster prevention resources. However, the losses in misidentification are mostly adjustable resource consumption, and the severity of the consequences is vastly different. Attributing both to identification bias would fail to accurately reflect the differences in actual losses. Therefore, to address these two types of disaster identification bias, corresponding unit social cost loss ratios are further established to transform the impact of losses into quantifiable cost parameters, ensuring the accuracy of subsequent social cost assessments.

[0030] The unit social cost loss ratio for setting disaster identification bias includes: Based on the basic distribution information of the target area, the severity of the consequences of missed disaster reporting and incorrect disaster reporting are calculated respectively; The relationship between the unit social cost loss multiples for disaster underreporting and disaster misreporting is calculated based on the severity of the corresponding consequences; The unit social cost loss ratio is set based on the relationship between the unit social cost loss multiple of disaster underreporting and disaster misreporting.

[0031] The basic distribution information includes population density distribution, infrastructure layout, and ecological sensitivity of the target area. Combined with preset rules, the severity of consequences for missed and incorrect disaster reporting is calculated. Specifically, the severity of the corresponding consequences can be calculated by assigning values ​​to each factor in the basic distribution information and then performing a weighted summation of these factors.

[0032] Then, based on the severity of the consequences, the cost calculation rules for disaster omissions and misreporting are matched to determine the corresponding verification cost items and quantification precision. Based on the cost calculation rules and combined with basic distribution information, the corresponding unit social cost loss is calculated. Furthermore, the unit social cost loss refers to the social cost loss incurred per unit area when a disaster omission or misreporting event occurs.

[0033] Furthermore, by comparing the unit social cost losses of disaster underreporting and disaster misreporting, the corresponding unit social cost loss ratio is determined, and this ratio is the unit social cost loss ratio.

[0034] Based on this, a social cost loss assessment index is further set up, which integrates the probability of omission, the probability of false identification, the ratio of unit social cost loss, and the proportion of disaster and non-disaster areas. The social cost loss assessment index value reflects the overall social cost under the current identification performance, so that the assessment standard shifts from technical fit to actual cost impact, avoiding the problem of meeting technical indicators but cost out of control.

[0035] The social cost loss assessment index for setting disaster identification bias includes: The probability of missing disasters is defined by the model recall metric, and the probability of false disasters is defined by the model specificity metric. By combining the ratio of unit social cost loss and the ratio of disaster area to non-disaster area, a coupling relationship between the probability of disaster omission and the probability of disaster misidentification is established, and a social cost loss assessment index for disaster identification bias is set.

[0036] The model recall rate is the proportion of disaster samples correctly identified by the model to the total number of actual disaster samples in the target area. It directly reflects the model's ability to cover disaster areas. If the recall rate is low, it means that a large number of actual disaster areas have not been identified by the model, that is, the phenomenon of disaster omission is prominent. Therefore, the model recall rate is used to define the probability of disaster omission.

[0037] The model specificity index is the proportion of non-disaster samples correctly identified by the model to the total number of actual non-disaster samples in the target area. It directly reflects the model's ability to distinguish non-disaster areas. If the specificity is low, it means that a large number of areas that are actually non-disasters are misclassified as disasters by the model, that is, the phenomenon of disaster misidentification is prominent. Therefore, the disaster misidentification probability is defined by the model specificity index.

[0038] Disaster omission events occur only in disaster areas, while disaster misidentification events occur only in non-disaster areas. By combining the differences in corresponding disaster identification biases, cost weights are assigned to corresponding events based on the ratio of unit social cost loss and the ratio of disaster to non-disaster area. Thus, the overall loss is quantified through the linear superposition of the two types of disaster identification bias events.

[0039] The expression for the social cost loss assessment index is as follows: ; ; ; in, As a social cost loss assessment index, The false negative rate is the probability of a disaster being missed. The false positive rate, i.e., the probability of incorrectly identifying a disaster. The model recall rate metric. As a model-specific indicator, The ratio of unit social cost loss. This represents the ratio of disaster-affected to non-disaster-affected areas.

[0040] Then, taking the minimum social cost loss assessment index as the optimization objective, the optimal sample ratio is derived through mathematical derivation.

[0041] Specifically, the expression for the social cost loss assessment index... and Considered as sample proportion The function of this function, the social cost loss assessment index, can then be expressed as: .

[0042] right about By taking the derivative and setting it to zero, and combining this with the statistical patterns of model training, the optimal sample ratio can be determined. .

[0043] To circumvent the limitations of theoretical assumptions in mathematical derivation, after determining the optimal sample ratio, the following is also performed: Several sets of sample ratios are constructed based on the optimal sample ratio and its adjacent sample ratios, and corresponding training sets and test sets are constructed by combining historical disaster data of the target area. The pre-set geological disaster assessment model is trained using training and testing sets for each sample proportion to obtain the social cost loss assessment index value for each sample proportion. The optimal sample proportion is updated based on the social cost loss assessment index value of each sample proportion.

[0044] Previously, the theoretically optimal sample ratio, derived mathematically, relied on assumptions such as the model's response to the sample ratio exhibiting an ideal pattern and historical data perfectly matching the actual disaster distribution. However, in real-world scenarios, historical disaster data for the target area may suffer from labeling bias and uneven distribution. Furthermore, the pre-set geological disaster assessment model may exhibit deviations in its adaptability to the theoretical ratio due to differences in algorithm characteristics. To avoid the potential limitations of a single theoretical ratio, this paper takes the theoretically optimal ratio as the core and extends it to both sides to set adjacent ratios, forming multiple sets of sample ratios to be verified, thus covering potential effective ratios beyond the theoretical assumptions.

[0045] Meanwhile, from the historical disaster data of the target area, disaster and non-disaster samples are extracted according to the proportion of each group of samples to construct corresponding training and test sets. It is ensured that the datasets of all groups are consistent in terms of data source range and basic characteristics of disaster and non-disaster samples, so as to avoid interference from the performance comparison between different sample proportions due to the differences in the data itself.

[0046] The pre-defined geological hazard assessment model is trained on the training set for each sample proportion. After training, each model is used to predict on the corresponding test set. Based on the real and non-hazard labels in the test set, the probability of missed detection and the probability of false detection for each model are calculated. Then, combined with the previously determined weights for missed detection and false detection, these are substituted into the social cost loss assessment index formula to obtain the social cost loss assessment index value corresponding to each sample proportion. The pre-defined geological hazard assessment model can be a deep learning model such as random forest or convolutional neural network.

[0047] Then, the optimal sample proportion is updated based on the social cost loss assessment index values ​​of each sample proportion, including: When the social cost loss assessment index value of the adjacent sample ratio is less than the current optimal sample ratio, the optimal sample ratio is updated according to the adjacent sample ratio. Reduce the refinement gradient, select neighboring samples of the updated optimal sample ratio based on the reduced refinement gradient, recalculate the corresponding social cost loss assessment index value, update the optimal sample ratio, until the corresponding update termination condition is met.

[0048] When the social cost loss assessment index value of the adjacent sample proportion is less than the current optimal sample proportion, it proves that the current optimal sample proportion is not a local optimum, and there are sample proportions in its vicinity that can better reduce social cost loss. In this case, the optimal proportion is first initially updated, and then the search interval is further narrowed by reducing the refinement gradient to improve the accuracy of proportion positioning. Furthermore, by gradually narrowing the possible range of the optimal proportion, and through repeated calculations and iterative updates, the social cost loss assessment index value can be continuously made to approach the minimum value.

[0049] To avoid excessive refinement leading to resource waste, corresponding update termination conditions are further set, including: The social cost loss assessment index values ​​of adjacent sample proportions are all greater than the current optimal sample proportion; Alternatively, the number of times the decrease in the social cost loss assessment index value is less than the preset threshold reaches the predetermined number threshold. Alternatively, refine the gradient to a preset gradient limit.

[0050] After one of the update termination conditions is met, the current optimal sample ratio is used as the final output, and a corresponding learning sample set is constructed to train and obtain the final geological hazard assessment model for disaster monitoring. Furthermore, the geological hazard assessment model selected for training must be consistent with the geological hazard assessment model used when determining the sample ratio.

[0051] The step of constructing a learning sample set according to the optimal sample ratio, training the preset geological hazard assessment model using the learning sample set, and obtaining the geological hazard probability based on the trained geological hazard assessment model includes: Based on historical disaster data, disaster samples are constructed. Non-disaster samples are randomly selected from non-disaster areas of the target region according to the optimal sample ratio. A learning sample set is constructed based on the disaster samples and non-disaster samples. The training set and test set are divided according to a preset ratio, and the preset geological hazard assessment model is trained and learned. Obtain the current remote sensing data of the target area, input the current remote sensing data into the trained geological hazard assessment model, and obtain the geological hazard probability of each pixel unit in the target area.

[0052] Another aspect of this embodiment also provides a geological disaster monitoring system based on social cost loss optimization, such as... Figure 2 As shown, it includes: Sample analysis module 1 is used to calculate the ratio of disaster-affected to non-disaster-affected areas of the target region based on historical remote sensing data, set the unit social cost loss ratio of disaster identification deviation, and set the social cost loss assessment index of disaster identification deviation. With the goal of minimizing the social cost loss assessment index, the optimal sample ratio is solved. Model building module 2 is used to build a learning sample set according to the optimal sample ratio, and to train the preset geological disaster assessment model using the learning sample set; Disaster monitoring module 3 is used to obtain the probability of geological disasters based on the trained geological disaster assessment model.

[0053] The sample analysis module, model building module, and disaster monitoring module are all equipped with computers and other devices that have corresponding data processing capabilities, and all have communication ports that can extract the required remote sensing data, disaster data, and other data from external platforms.

[0054] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any way. Other variations and modifications are possible without departing from the technical solutions described in the claims.

Claims

1. A geological hazard monitoring method based on social cost loss optimization, characterized in that, include: The ratio of disaster-affected to non-disaster-affected areas in the target region is calculated based on historical remote sensing data, and the ratio of unit social cost loss for disaster identification bias is set. Set up a social cost loss assessment index for disaster identification bias, and solve for the optimal sample ratio with the goal of minimizing the social cost loss assessment index; A learning sample set is constructed based on the optimal sample ratio. The preset geological hazard assessment model is trained using the learning sample set, and the probability of geological hazards is obtained based on the trained geological hazard assessment model.

2. The geological hazard monitoring method based on social cost loss optimization according to claim 1, characterized in that, The calculation of the disaster-to-non-disaster area ratio of the target region based on historical remote sensing data includes: Based on historical remote sensing data, the boundaries between disaster-prone and non-disaster-prone areas within the target region are identified, and the boundary identification results are corrected by combining the corresponding survey data; The target area is divided into several pixel units of equal area, and each pixel unit is divided into a disaster grid or a non-disaster grid based on the corrected boundary. Count the number of disaster grids and non-disaster grids, and calculate the total area of ​​the disaster area and the total area of ​​the non-disaster area by combining the area of ​​the corresponding pixel unit; By comparing the total area of ​​the disaster area with the total area of ​​the non-disaster area, the ratio of the disaster area to the non-disaster area of ​​the target area can be obtained.

3. The geological hazard monitoring method based on social cost loss optimization according to claim 1, characterized in that, The disaster identification bias includes both missed disaster identification and incorrect disaster identification.

4. The geological hazard monitoring method based on social cost loss optimization according to claim 3, characterized in that, The unit social cost loss ratio for setting disaster identification bias includes: Based on the basic distribution information of the target area, the severity of the consequences of missed disaster reporting and incorrect disaster reporting are calculated respectively; The relationship between the unit social cost loss multiples for disaster underreporting and disaster misreporting is calculated based on the severity of the corresponding consequences; The unit social cost loss ratio is set based on the relationship between the unit social cost loss multiple of disaster underreporting and disaster misreporting.

5. The geological hazard monitoring method based on social cost loss optimization according to claim 3, characterized in that, The social cost loss assessment index for setting disaster identification bias includes: The probability of missing disasters is defined by the model recall metric, and the probability of false disasters is defined by the model specificity metric. By combining the ratio of unit social cost loss and the ratio of disaster area to non-disaster area, a coupling relationship between the probability of disaster omission and the probability of disaster misidentification is established, and a social cost loss assessment index for disaster identification bias is set.

6. The geological hazard monitoring method based on social cost loss optimization according to claim 1, characterized in that, After finding the optimal sample proportion with the goal of minimizing the social cost loss assessment index, the following steps are also performed: Several sets of sample ratios are constructed based on the optimal sample ratio and its adjacent sample ratios, and corresponding training sets and test sets are constructed by combining historical disaster data of the target area. The pre-set geological disaster assessment model is trained using training and testing sets for each sample proportion to obtain the social cost loss assessment index value for each sample proportion. The optimal sample proportion is updated based on the social cost loss assessment index value of each sample proportion.

7. The geological hazard monitoring method based on social cost loss optimization according to claim 6, characterized in that, The process of updating the optimal sample proportion based on the social cost loss assessment index value of each sample proportion includes: When the social cost loss assessment index value of the adjacent sample ratio is less than the current optimal sample ratio, the optimal sample ratio is updated according to the adjacent sample ratio. Reduce the refinement gradient, select neighboring samples of the updated optimal sample ratio based on the reduced refinement gradient, recalculate the corresponding social cost loss assessment index value, update the optimal sample ratio, until the corresponding update termination condition is met.

8. The geological disaster monitoring method based on social cost loss optimization according to claim 7, characterized in that, The update termination conditions include: The social cost loss assessment index values ​​of adjacent sample proportions are all greater than the current optimal sample proportion; Alternatively, the number of times the decrease in the social cost loss assessment index value is less than the preset threshold reaches the predetermined number threshold. Alternatively, refine the gradient to a preset gradient limit.

9. The geological hazard monitoring method based on social cost loss optimization according to claim 1, characterized in that, The step of constructing a learning sample set based on the optimal sample ratio, training the preset geological hazard assessment model using the learning sample set, and obtaining the geological hazard probability based on the trained geological hazard assessment model includes: Based on historical disaster data, disaster samples are constructed. Non-disaster samples are randomly selected from non-disaster areas of the target region according to the optimal sample ratio. A learning sample set is constructed based on the disaster samples and non-disaster samples. The training set and test set are divided according to a preset ratio, and the preset geological hazard assessment model is trained and learned. Obtain the current remote sensing data of the target area, input the current remote sensing data into the trained geological hazard assessment model, and obtain the geological hazard probability of each pixel unit in the target area.

10. A geological hazard monitoring system based on social cost loss optimization, used to execute the geological hazard monitoring method according to any one of claims 1 to 9, characterized in that, include: The sample analysis module is used to calculate the ratio of disaster-affected to non-disaster-affected areas of the target region based on historical remote sensing data, set the unit social cost loss ratio of disaster identification deviation, and set the social cost loss assessment index of disaster identification deviation. With the goal of minimizing the social cost loss assessment index, the optimal sample ratio is solved. The model building module is used to build a learning sample set according to the optimal sample ratio, and to train the preset geological disaster assessment model using the learning sample set; The disaster monitoring module is used to obtain the probability of geological disasters based on the trained geological disaster assessment model.

Images