Acquisition method and extraction method of house crack information extraction model

Abstract

The invention provides an acquisition method and an extraction method of a house crack information extraction model, and the method comprises the steps: constructing a house data set, and carrying out the model training of each learning model in a pre-constructed model set through the house data set; performing performance comparison on the trained learning models, and selecting the learning model with the optimal performance as a target model of the working area; according to the method, the optimal extraction model matched with the house crack generation characteristics in the working area can be rapidly, efficiently and accurately obtained, the extraction efficiency and accuracy of the house crack information are effectively improved, and the method provided by the invention also has relatively high adaptability and expansibility.

Description

Technical Field

[0001] This invention relates to the field of remote sensing monitoring technology, and in particular to a method for obtaining a model for extracting building crack information, a method for extracting building crack information, electronic equipment, and computer program products. Background Technology

[0002] Cracks in buildings are one of the important factors affecting the safety of building structures. To ensure the safety of building structures, it is usually necessary to extract and continuously monitor information such as whether cracks have occurred, the length of cracks, or the density of cracks.

[0003] In existing technologies, the extraction of information about building cracks is often based on deformation measurement methods, such as IoT-based building monitoring and civil engineering-based building monitoring. These methods typically require the installation of numerous sensors around the building to obtain real-time monitoring data on settlement, tilt, and other factors. However, these methods not only rely on a large amount of specialized equipment and instruments, but they are also usually only applicable to crack detection in detached buildings, resulting in low efficiency and an inability to quickly extract and monitor crack information across a large area of ​​buildings in a short period of time.

[0004] Although there are currently methods to extract building crack information by extracting surface subsidence in the area and further obtaining building tilt based on this, these methods usually only consider ground subsidence information and do not fully consider other factors such as the building's own structure and the weather conditions in which the building is located, resulting in low accuracy in extracting crack information.

[0005] Therefore, how to quickly and accurately extract building crack information over a large area has become a technical problem that needs to be solved in this field. Summary of the Invention

[0006] In view of the shortcomings of the prior art described above, the purpose of this invention is to provide a method for obtaining a model for extracting building crack information, a method for extracting building crack information, an electronic device and a computer program product, to solve the problems that existing building crack information extraction methods have low extraction efficiency and cannot be applied to large-scale / long-term rapid and accurate extraction.

[0007] To achieve the above and other related objectives, the present invention first provides a method for obtaining a model for extracting information about building cracks, comprising: constructing a building dataset; the building dataset containing building feature information, geological environment feature information, meteorological environment feature information, and crack information corresponding to each building element; based on the building dataset, performing model training on each learning model in a pre-constructed model set; comparing the performance of each trained learning model, and selecting the learning model with the best performance as the target model for the working area.

[0008] In an optional embodiment, the construction of the housing dataset includes: acquiring housing survey data within the work area, and geological environment data, meteorological environment data, and housing crack data that match the housing survey data in terms of collection time and space; extracting crack information corresponding to each housing element based on the housing crack data; acquiring corresponding housing feature information, geological environment feature information, and meteorological environment feature information from the housing survey data, the geological environment data, and the meteorological environment data according to selected housing feature factors, geological environment feature factors, and meteorological environment feature factors; summarizing the housing feature information, geological environment feature information, and meteorological environment feature information corresponding to the same housing element into a feature information vector; and constructing the housing dataset based on the feature information vector corresponding to each housing element and the housing crack information; wherein the housing feature factors, the geological environment feature factors, and the meteorological environment feature factors are feature factors related to the generation of housing cracks.

[0009] In one optional embodiment, the construction of the housing dataset further includes: Using correlation analysis, each of the aforementioned building characteristic factors, geological environment factors, and meteorological environment factors is screened based on the correlation of crack information to obtain building characteristic factors, geological environment factors, and meteorological environment factors with a correlation greater than a threshold; wherein, the correlation analysis method includes Pearson correlation coefficient and random forest importance analysis.

[0010] In one optional embodiment, the construction of the housing dataset further includes: Data preprocessing is performed on the building feature information, the geological environment feature information, and the meteorological environment feature information, respectively, including: using a label encoding method to convert the text information in the building feature information, the geological environment feature information, and the meteorological environment feature information into corresponding numerical information; and using the natural breakpoint method to convert the continuously distributed numerical values ​​in the building feature information, the geological environment feature information, and the meteorological environment feature information into discrete interval distributed numerical values.

[0011] In one optional embodiment, the performance comparison of the trained learning models includes: Based on the preset model evaluation metrics, the performance values ​​corresponding to each of the trained learning models are extracted; the performance values ​​of each learning model are comprehensively compared to obtain the performance comparison results corresponding to each learning model; wherein, the model evaluation metrics include F1-score, precision, recall and overall accuracy.

[0012] In an optional embodiment, the method for obtaining the building crack information extraction model further includes: The working area is divided into several sub-regions according to region type, and a housing dataset corresponding to each sub-region is constructed. The housing dataset is partitioned, and the learning model is trained based on each partitioned data subset. The step of training each learning model in the pre-constructed model set based on the housing dataset includes: training each learning model in the pre-constructed model set based on the housing dataset corresponding to each sub-region to obtain the trained learning model for each sub-region. The step of comparing the performance of each trained learning model includes: comparing the performance of each sub-region with the trained learning model, and selecting the learning model with the best performance as the target model corresponding to that sub-region.

[0013] In one optional embodiment, the method for constructing the housing dataset corresponding to each sub-region includes: Select housing characteristic factors, geological environment characteristic factors, and meteorological environment characteristic factors that are associated with the occurrence of housing cracks in the sub-region; based on the housing survey data, the geological environment data, and the meteorological environment data, according to the housing characteristic factors, geological environment characteristic factors, and meteorological environment characteristic factors corresponding to the sub-region, obtain the corresponding housing characteristic information, geological environment characteristic information, and meteorological environment characteristic information for each housing element in the sub-region; based on the housing characteristic information, geological environment characteristic information, meteorological environment characteristic information, and housing crack information for each housing element in the sub-region, construct the housing data corresponding to each sub-region.

[0014] The second aspect of this application provides a method for extracting information about cracks in a building, including: The area to be extracted for building crack information is determined; the building crack information is extracted using a building crack information extraction model to obtain the building crack information corresponding to each building element; wherein, the crack information extraction model is a model obtained by using any of the building crack information extraction models described above.

[0015] A third aspect of this application provides an electronic device, including at least one processor and a memory connected to the processor, wherein: the memory is used to store a computer program; the processor is used to execute the computer program to enable the electronic device to implement the method for obtaining a house crack information extraction model as described above, and / or to implement the steps of the method for extracting house crack information as described above.

[0016] A fourth aspect of this application provides a computer program product including computer-readable instructions that, when executed on an electronic device, cause the electronic device to implement the method for obtaining a house crack information extraction model as described above, and / or to implement the steps of the method for extracting house crack information as described above.

[0017] As described above, the method for obtaining the building crack information extraction model provided in this application first constructs a building dataset for crack information extraction based on building feature factors, geological environment factors, and meteorological environment factors. Then, it uses this building dataset to train a pre-constructed learning model and comprehensively compares the performance of each trained learning model. This allows for the rapid, efficient, and accurate acquisition of the optimal extraction model that best matches the crack generation characteristics of buildings within the working area. Furthermore, using the model obtained through the model acquisition method provided in this application for building crack information extraction not only improves the accuracy of crack information extraction but also makes the method applicable to different working areas, exhibiting high adaptability and scalability. This provides an effective monitoring method for large-area / long-term building crack monitoring. Attached Figure Description

[0018] Figure 1 The diagram shows a flowchart of an embodiment of the method for obtaining the building crack information extraction model provided in this application; Figure 2 The diagram shown is a flowchart of another embodiment of the method for obtaining the building crack information extraction model provided in this application; Figure 3 The diagram shown is a flowchart illustrating the extraction of building crack information provided in this application in one embodiment. Figure 4 The diagram shown is a structural schematic of the electronic device provided by the present invention in one embodiment. Detailed Implementation The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that, unless otherwise specified, the following embodiments and features described therein can be combined with each other.

[0019] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention, and are therefore not intended to be considered as drawings.

[0020] To address the technical problems existing in the prior art, this invention provides a method for obtaining a model for extracting building crack information, a method for extracting building crack information, an electronic device, and a computer program product. By extracting various feature factors related to the generation of building cracks and extracting crack information from buildings, a building dataset for model training is constructed. Using this building dataset, multiple preset learning models are trained separately, and the performance of each trained model is evaluated to quickly and efficiently obtain a target model best suited to the working area. Then, based on the selected target model, building crack information corresponding to each building element is obtained, thereby effectively improving the extraction efficiency and accuracy of building crack information.

[0021] Please see Figure 1 The diagram shows a flowchart of an embodiment of the method for obtaining the building crack information extraction model provided in this application.

[0022] S100, acquire housing survey data within the work area, and acquire geological environment data, meteorological environment data, and housing crack data that match the housing survey data in terms of collection time and space; The housing census data refers to data obtained after investigating factors related to the structural safety of buildings (such as the building structure and materials). The geological environment data includes data on geological elements related to the structural safety of buildings; The meteorological environmental data includes meteorological elements related to the structural safety of buildings; The data on building cracks is obtained after investigating crack information in building structures.

[0023] In this application, the crack information refers to data characterizing whether cracks have occurred in the building and the extent of crack occurrence.

[0024] In one embodiment of this application, the house crack information is whether cracks have occurred or not; those skilled in the art will know that in other embodiments, the crack information may also be the number of cracks or the crack density, etc., and the specific implementation of the house crack information does not affect the execution effect of the method described in this application.

[0025] Specifically, collect housing census data within the work area and obtain the collection time corresponding to the housing census data; based on the collection time, use the time interval with the collection time not greater than the interval threshold as the matching time period for the housing census data; Wherein, the interval threshold is a preset threshold; optionally, the time precision of the threshold is the same as the time precision of the collection time. For example, if the collection time is September 2020 and the interval threshold is 2 months, then July 2020 to November 2020 will be used as the matching time period.

[0026] Based on the matching time period of the housing census data, environmental data within the matching time period is extracted from the environmental data corresponding to the work area, and used as environmental data that matches the housing census data in terms of collection time; similarly, housing crack information within the matching time period is extracted, and used as housing crack information that matches the housing census data in terms of collection time.

[0027] In this embodiment, the geological environment data includes InSAR imagery; the meteorological environment data includes meteorological station observation data; those skilled in the art should know that in other embodiments, the geological environment data may also include other types of geological environment data such as DEM; similarly, the meteorological environment data may also include other types of meteorological environment data such as remote sensing meteorological data.

[0028] S200, Based on the house crack data, extract the crack information corresponding to each house element; Specifically, the identification information corresponding to each building element is obtained; based on the identification information of each building element, the crack information corresponding to each building element is extracted from the building crack data.

[0029] It should be noted that, in some optional embodiments, the house crack data is the house disaster-bearing body survey data, that is, the house crack information is directly contained in the house disaster-bearing body survey data. In this case, the crack information corresponding to each house element can be obtained directly by extracting the house crack information in the house disaster-bearing body survey data.

[0030] S300, based on the building characteristic factors, geological environment characteristic factors, and meteorological environment characteristic factors, obtain the corresponding building characteristic information, geological environment characteristic information, and meteorological environment characteristic information for each building element from the building survey data, the geological environment data, and the meteorological environment data; Among them, the building characteristic factor, the geological environment characteristic factor, and the meteorological environment characteristic factor are characteristic factors related to the generation of building cracks.

[0031] Specifically, among the characteristic factors included in the housing survey data, several characteristic factors associated with the occurrence of housing cracks are pre-selected as housing characteristic factors; similarly, among the characteristic factors included in the geological environment data, several characteristic factors associated with the occurrence of housing cracks are pre-selected as geological environment characteristic factors; and among the characteristic factors included in the meteorological environment data, several characteristic factors associated with the occurrence of housing cracks are pre-selected as meteorological environment characteristic factors.

[0032] For example, the building characteristic factors include building design factors, construction factors, building structure factors, building material factors, renovation and reinforcement factors, and building aging factors. The geological environment characteristic factors include ground subsidence rate and slope; wherein, the ground subsidence rate is used to characterize the change of subsidence corresponding to the ground sampling point over time; optionally, the ground subsidence rate is obtained based on InSAR images within the working area.

[0033] The meteorological environmental characteristics include temperature change and acid rain amount; wherein, the temperature change is the difference between the highest and lowest temperatures in the region during the year; and the acid rain amount is the amount of acid rain that falls in the region during the year.

[0034] In other examples, the geological environment characteristic factors also include other characteristic factors such as elevation and distribution of geological fault zones; the meteorological environment characteristic factors also include other characteristic factors such as maximum wind speed and maximum rainfall, which will not be elaborated here.

[0035] In the embodiments of this application, the building characteristic factors, the geological environment characteristic factors, and the meteorological environment characteristic factors are characteristic factors that are predetermined based on expert experience and determined through manual screening.

[0036] After selecting the aforementioned building characteristic factors, geological environment characteristic factors, and meteorological environment characteristic factors, based on the identification information of each building element, the values ​​of the building characteristic factors corresponding to each identification information are extracted from the building census data as the building characteristic information corresponding to each building element; and based on the location information of each building element, the values ​​of the geological environment characteristic factors located at the corresponding locations are extracted from the geological environment data as the geological environment characteristic information corresponding to each building element; and based on the location information of each building element, the values ​​of the meteorological environment characteristic factors located at the corresponding locations are extracted from the meteorological environment data as the meteorological environment characteristic information corresponding to each building element.

[0037] For example, if the identification information is a house ID, then based on the house ID corresponding to each house element, the value of the building structure factor (house feature factor) corresponding to the house ID is extracted from the house survey data to obtain the building structure information corresponding to each house element.

[0038] For example, the location information is the spatial range of the building element; then, based on the spatial range corresponding to each building element, the value of the ground subsidence rate (geological environment characteristic factor) corresponding to each pixel within the range of the building element is extracted from the ground subsidence rate distribution data; the ground subsidence rate corresponding to each pixel is integrated to obtain the ground subsidence rate corresponding to each building element. Similarly, based on the spatial range corresponding to each building element, the acid rain amount (meteorological environmental characteristic factor) corresponding to each pixel within the range of the building element is extracted from the acid rain distribution data; the acid rain amount values ​​corresponding to each pixel are combined to obtain the acid rain amount corresponding to each building element.

[0039] In a more specific embodiment, the housing survey data is housing disaster-bearing body census data, which is data obtained after conducting a basic survey of various housing elements within the region; the housing characteristic factors selected from this data are shown in Table 1 below, including: Table 1 Housing Characteristic Factors Since the accuracy of feature factor selection directly affects the accuracy of subsequent model training, and the number of selected feature factors also affects the efficiency of model training—that is, the more accurate and fewer the selected feature factors, the higher the accuracy and efficiency of model training—in some preferred embodiments, to more accurately and objectively select feature factors highly correlated with building cracks and to minimize the number of selected feature factors, the method for obtaining the building crack information extraction model further includes, during step S300: Using correlation analysis, each of the building characteristic factors, the geological environment factors, and the meteorological environment factors is screened based on the correlation with crack information to obtain the building characteristic factors, the geological environment characteristic factors, and the meteorological environment characteristic factors whose correlation with the crack information is greater than a threshold.

[0040] Specifically, the building characteristic factors, geological environment characteristic factors, and meteorological environment characteristic factors, which are pre-determined through manual screening, will be used as initial characteristic factors; Correlation analysis was used to extract the correlation between each initial feature factor and the crack information. Initial feature factors with a correlation below the threshold were removed, while initial feature factors with high correlation were retained.

[0041] The above method was used to screen each initial building characteristic factor, geological environment characteristic factor, and meteorological environment characteristic factor to obtain building characteristic factors, geological environment characteristic factors, and meteorological environment characteristic factors that are highly correlated with crack information after screening.

[0042] In the embodiments of this application, two factor correlation analysis methods, Pearson correlation coefficient and random forest importance analysis, are used to screen each initial housing characteristic factor, geological environment characteristic factor, or meteorological environment characteristic factor (hereinafter referred to as "initial characteristic factors") to eliminate initial characteristic factors with low correlation.

[0043] The Pearson correlation coefficient is a statistic used to measure the strength and direction of the linear relationship between two characteristic factors. Its calculation formula is as follows: In the formula, For the first i Initial feature factors of each sample data X Factor values, For the first i Crack information values ​​for each sample data point. Initial feature factors for each sample data X The factor mean; The mean value of crack information for each sample data; r Initial characteristic factor X The Pearson correlation coefficient between crack information and crack information; the larger the value, the higher the correlation.

[0044] The random forest feature importance analysis involves randomly building multiple decision trees and recording the contribution of each feature factor to the model accuracy at each split node. The results from all trees are then aggregated to determine the importance of different feature factors to the dependent variable (split information), thus evaluating the importance of the feature factors. The calculation formula is as follows: in, T The total number of decision trees in the random forest; It is the first t The weight of a tree is usually determined by the proportion of the training samples it uses in the total training samples. It is a characteristic factor j In the t The decrease in Gini index caused by node splitting in a tree; The larger the value, the more important the feature factor.

[0045] In one optional embodiment, random forest importance analysis is used to perform the screening of each initial feature factor; initial feature factors with correlation below a threshold are used as first low-correlation feature factors; and Pearson correlation analysis is used to perform the screening of each initial feature factor, and initial feature factors with correlation below a threshold are used as second low-correlation feature factors. Based on this, the first low-correlation feature factors selected by random forest importance analysis and the second low-correlation feature factors selected by Pearson correlation analysis are combined. Feature factors with low correlation in both are removed, and the remaining feature factors are taken as high-correlation feature factors.

[0046] For example, a total of 18 characteristic factors were selected as the final highly correlated characteristic factors: construction time, construction method, building purpose, building type, structural type, building area, building height, highest temperature, average temperature, maximum settlement rate, minimum settlement rate, average settlement rate, settlement rate difference, maximum relative tilt rate, average relative tilt rate, atmospheric SO2 concentration, atmospheric NO2 concentration, and precipitation.

[0047] S400, Based on the house feature information, the geological environment feature information, the meteorological environment feature information, and the house crack information, construct a house dataset for model training; Specifically, after obtaining the house feature information, the geological environment feature information, the meteorological environment feature information, and the crack information, for each house element, the house feature information, the geological environment feature information, and the meteorological environment feature information corresponding to the same house element are summarized into a vector to obtain the feature information vector corresponding to the house element.

[0048] Based on the feature information vectors corresponding to each building element and the building crack information, a dataset for model training is constructed; that is, the building feature information, geological environment feature information, and meteorological environment feature information are used as the input features corresponding to each building element, and the crack information is used as the label information corresponding to the building element to construct sample data corresponding to the building element, thereby obtaining a building dataset for subsequent model training.

[0049] In practical applications, since the data types corresponding to the collected feature information are not the same—for example, "ground settlement rate" is a numerical value, and its corresponding data type is floating-point; "building structure type" is text, and its corresponding data type is string; to facilitate subsequent model training and ensure the stability and accuracy of the trained model, in some optional embodiments, step S400 further includes the following when executed: Data preprocessing is performed on the building feature information, the geological environment feature information, and the meteorological environment feature information respectively, so as to construct a dataset based on the preprocessed feature information; Specifically, a tag encoding method is used to convert the text data in the house feature information, the geological environment feature information, and the meteorological environment feature information into corresponding numerical data, so as to transform text-type feature information into numerical-type feature information.

[0050] Furthermore, the natural breakpoint method is used to transform the continuously distributed numerical data in the house feature information, the geological environment feature information, and the meteorological environment feature information into discrete interval distributed numerical data, so as to transform the floating-point feature information into integer feature information; for example, the house area is divided into 10 levels using the natural breakpoint method; while the house height, due to its relatively small numerical range, is divided into 5 levels using the natural breakpoint method.

[0051] In this embodiment of the application, data preprocessing converts different types of feature factors into feature factors of the same data type, thereby effectively improving the efficiency and accuracy of model training in subsequent steps, making it easier for the model to understand the impact of different feature factors on house cracks.

[0052] S500, Based on the housing dataset, perform model training on each learning model in the pre-built model set to obtain the trained learning model; The model set includes several different types of learning models, and the learning model is used to extract information about cracks in buildings.

[0053] In this application, the learning model includes several machine learning models and several deep learning models; the machine learning models are models trained based on manually selected feature metrics; the deep learning models are models trained based on input raw data by automatically extracting features.

[0054] In this embodiment, the machine learning model includes SVM, RF, and LightGBM; the deep learning model includes MLP, CRNN, and TabNet.

[0055] Specifically, using the housing datasets constructed in step S300, the pre-constructed SVM model, RF model, and LightGBM model are trained respectively to obtain the trained SVM model, RF model, and LightGBM model accordingly.

[0056] Similarly, using the housing datasets constructed in step S300, model training is performed on the pre-constructed MLP model, CRNN model, and TabNet model respectively to obtain the trained MLP model, CRNN model, and TabNet model accordingly.

[0057] For a single learning model, during model training, the feature information vectors corresponding to each building element, namely the building feature information, the geological environment feature information, and the meteorological environment feature information, are used as the independent variables of the model, and the crack information corresponding to each building element is used as the label information of the model.

[0058] More specifically, after obtaining the housing dataset, the dataset is divided according to a preset ratio to obtain a training set, a test set, and a validation set respectively. For example, 20% of the housing sample data in the dataset is divided as the validation set, and then the remaining housing sample data is divided into the training set and the test set at a ratio of 4 / 1, that is, the ratio between the training set, the test set, and the validation set is 64%:16%:20%.

[0059] S600 compares the performance of each trained learning model and selects the learning model with the best performance as the target model for the working region.

[0060] Specifically, based on the preset model evaluation metrics, the performance values ​​corresponding to each learning model after training are extracted; the performance values ​​of each learning model are comprehensively compared to obtain the performance comparison results corresponding to each learning model.

[0061] Among all the learning models, the one with the best performance comparison results is selected as the target model for extracting building crack information.

[0062] The model evaluation metrics include, but are not limited to, one or more of the following metrics: F1-score, precision, recall, and overall accuracy.

[0063] In one specific implementation, the target model focuses more on accurately identifying and recognizing buildings with complete cracks (i.e., pinpointing both the exact location and the completeness of the cracks). Therefore, while ensuring high model accuracy, the recall rate of the model for building crack information is relatively more important. Based on this, when evaluating the overall performance of the model, the average accuracy and recall rate of building crack information are mainly considered. If the average accuracy is similar, the model with the higher recall rate for building crack information is selected as the target model.

[0064] The method for obtaining the building crack information extraction model provided in this application constructs a building crack information extraction dataset by acquiring building feature factors, geological environment factors, and meteorological environment factors corresponding to each building element in the work area. Furthermore, by using the constructed building dataset, multiple pre-built learning models are trained simultaneously, and the performance of each trained learning model is comprehensively compared. This allows for the rapid, efficient, and accurate acquisition of the target model that best matches the work area—that is, the optimal extraction model that best fits the features associated with building cracks within the work area. This not only improves the accuracy of building crack information extraction but also extends the method to different work areas, exhibiting high adaptability and scalability. It provides an effective monitoring method for large-area / long-term building crack monitoring.

[0065] In the model training process, the representativeness (feature saliency) of the sample dataset has a significant impact on the accuracy and efficiency of model training. That is, the stronger the representativeness (the stronger the feature saliency) of the sample dataset, the higher the accuracy and efficiency of model training.

[0066] The housing features contained in the housing dataset are often associated with the regional type of the area where the housing is located. That is, the features of houses located in the same region are often the same or similar. For example, houses located in central towns are mainly high-rise shear structure buildings, while houses located in rural areas are mainly low-rise frame structure buildings.

[0067] Based on this, in order to quickly construct a housing dataset with significant housing features for efficient training of various learning models; and to further determine the matching effect between housing sample data corresponding to different regional types and different learning models, so as to select the learning model most suitable for the regional type from among the different learning models, in an optional embodiment, the method for obtaining the housing crack information extraction model may further include the following when performing step S300: S300' divides the work area into several sub-regions according to region type; for each sub-region, a corresponding housing dataset is constructed. Specifically, the work area is divided into different types of areas, and the boundaries of each type of area are obtained. The work area is then divided based on these boundaries to obtain sub-areas corresponding to different types of areas. The type of area is a type used to divide administrative areas based on characteristics such as building structure and distribution.

[0068] In one specific implementation, the region type includes central towns, urban-rural fringe areas, and rural areas; accordingly, the working area is divided into central town sub-regions, urban-rural fringe sub-regions, and rural sub-regions according to the region type. After obtaining each sub-region, for a single sub-region, building characteristic factors, geological environment characteristic factors, and meteorological environment characteristic factors associated with the generation of building cracks in that sub-region are selected as characteristic factors of that sub-region; Within each sub-region, based on the housing survey data, the geological environment data, and the meteorological environment data, and according to the housing characteristic factors, geological environment characteristic factors, and meteorological environment characteristic factors corresponding to that sub-region, the housing characteristic information, geological environment characteristic information, and meteorological environment characteristic information corresponding to each housing element in that sub-region are obtained accordingly.

[0069] This step is performed on each sub-region to obtain the building characteristic information, geological environment characteristic information, and meteorological environment characteristic information corresponding to the building elements in each sub-region.

[0070] In some preferred embodiments, for a single sub-region, a factor correlation analysis method is used to select building characteristic factors, geological environment characteristic factors, and meteorological environment characteristic factors that are associated with the generation of building cracks in that sub-region; wherein, the implementation method of the factor correlation analysis method is the same as that in the above embodiments, and will not be described again here.

[0071] For example, in the central town sub-region, after factor correlation analysis, it is found that the characteristic factors of nitrogen dioxide concentration and sulfur dioxide concentration have a low correlation with the information on building cracks. Therefore, nitrogen dioxide concentration and sulfur dioxide concentration are removed from the characteristic factors of the central town sub-region. However, in the urban-rural fringe sub-region, after factor correlation analysis, it is found that the characteristic factors of nitrogen dioxide concentration and sulfur dioxide concentration have a high correlation with the information on building cracks. Therefore, nitrogen dioxide concentration and sulfur dioxide concentration are retained from the characteristic factors of the urban-rural fringe sub-region.

[0072] Furthermore, for the central town sub-region and the urban-rural fringe sub-region, the buildings within these regions often exhibit high consistency in terms of building materials, construction methods, and years of construction. Therefore, the safety status of an individual building is often strongly correlated with that of its neighboring buildings. Based on this, the building characteristic factors corresponding to the central town sub-region have been supplemented with factors related to neighboring buildings and the proportion of cracked buildings within the neighboring area. The adjacent building correlation factor is used to quantify the degree of consistency between the target building (such as a building with cracks) and its adjacent buildings in terms of building characteristics such as materials, structural type and construction time; the higher the value of the factor, the stronger the consistency of building characteristics between the target building and its adjacent buildings.

[0073] Specifically, taking the target building as the center, a certain spatial radius is selected as the neighborhood range; buildings other than the target building within this neighborhood range are extracted as the neighboring buildings of the target building; the similarity of the target building and each neighboring building in terms of building materials, structural type, and year of construction is compared. When the similarity is greater than the corresponding similarity threshold, the current building feature factor is determined to be the same and the similarity determination result of the current building feature factor is recorded as 1; otherwise, the similarity determination result of the current building feature factor is recorded as 0; the similarity determination results of different building feature factors are accumulated and normalized to the interval [0,1], and the normalized value is used as the relevant factor value of the neighboring buildings of the target building.

[0074] The cracked building proportion factor within the adjacent area is the proportion of buildings with cracks within a preset area to the total number of buildings in that area. It is used to reflect the overall risk level of cracks in buildings within a community or block, thereby providing contextual information for the crack risk assessment of individual buildings.

[0075] Based on this, in this embodiment, the method for obtaining the building crack information extraction model, after completing step S300', performs the following accordingly: S400', Based on the house feature information, geological environment feature information, meteorological environment feature information, and house crack information of each house element in the sub-region, construct the house dataset corresponding to each sub-region; Specifically, for a single sub-region, after obtaining the building feature information, geological environment feature information, meteorological environment feature information, and crack information corresponding to each building element in the sub-region, the building feature information, geological environment feature information, and meteorological environment feature information corresponding to the same building element are summarized into a vector to obtain the feature information vector corresponding to the building element.

[0076] Based on the feature information vectors corresponding to each house element in the sub-region and the house crack information, a house dataset corresponding to the sub-region is constructed.

[0077] This step is performed on each sub-region within the work area to obtain the corresponding housing dataset for each sub-region.

[0078] S500', based on the housing dataset corresponding to each sub-region, performs model training on each learning model in the pre-built model set to obtain the trained learning model for each sub-region.

[0079] Specifically, for a single sub-region, the model training is performed on each learning model in the pre-built model set using the housing dataset corresponding to that sub-region; the specific implementation method of this process is the same as that in the above embodiments, and will not be repeated here.

[0080] For example, for a sub-region of a central town, the corresponding housing data subset is used to train SVM, RF, LightGBM, MLP, CRNN and TabNet models respectively, so as to obtain the trained SVM, RF and LightGBM, MLP, CRNN and TabNet models corresponding to the sub-region of the central town.

[0081] S600' compares the performance of the trained learning model for each sub-region and selects the learning model with the best performance as the target model for that sub-region.

[0082] Specifically, for each sub-region, a test dataset matching that sub-region is obtained; each trained learning model is used to perform tests on the test dataset to obtain performance metrics corresponding to different learning models; the performance metrics corresponding to each learning model are combined to obtain a comprehensive metric for each learning model; by comparing the comprehensive metrics of each learning model, the learning model with the highest comprehensive performance metric is selected as the target model for that sub-region.

[0083] When constructing housing datasets corresponding to different sub-regions, the amount of minority class samples (houses with cracks) and majority class samples (houses without cracks) in the housing datasets corresponding to different sub-regions often differs. For example, the proportion of minority class samples in housing datasets corresponding to rural sub-regions and central town sub-regions is often relatively small compared to the majority class samples, while the proportion of minority class samples in housing datasets corresponding to urban-rural fringe sub-regions is often relatively large compared to the majority class samples. Therefore, to better balance the ratio between minority and majority class samples during model training and avoid the model biasing towards the majority class (houses without cracks), thereby significantly reducing the ability to identify the minority class (houses with cracks) and leading to the risk of missed identification of house cracks, in some preferred embodiments, the method for obtaining the house crack information extraction model further includes, during step S400': Based on the distribution of minority and majority class samples in different sub-regions, corresponding data balancing methods are used to process the housing dataset. Specifically, for a single sub-region, the ratio of minority class samples to majority class samples within that sub-region is detected. When this ratio is greater than a ratio threshold, it indicates that the proportion of minority class samples is relatively large, and undersampling is used to process the housing dataset. Conversely, when the ratio is less than or equal to the ratio threshold, it indicates that the proportion of minority class samples is relatively small, and SMOTE oversampling is used to process the housing dataset by interpolating and synthesizing new sample data between minority class samples.

[0084] Based on the same inventive concept, this application also provides a method for extracting information about building cracks; please refer to [link to relevant documentation]. Figure 3 The diagram shows a flowchart of an embodiment of the method for extracting information on building cracks provided in this application.

[0085] like Figure 3 As shown, the method, when executed, includes the following steps: S10, Determine the area to be extracted for building crack information; The area to be extracted refers to the area where the information on house cracks has not yet been extracted or updated.

[0086] S20, Obtain the feature factor vector associated with the generation of house cracks in the area to be extracted; Specifically, the area to be extracted is obtained; based on the area, housing survey data within the area is collected; and based on the collection time of the housing survey data, geological environment data, meteorological environment data, and housing crack data that match the housing survey data in terms of collection time and space are obtained. Furthermore, from the housing survey data, the geological environment data, and the meteorological environment data, housing feature factors, geological environment factors, and meteorological environment factors corresponding to each housing element are selected respectively to construct a feature factor vector associated with the generation of housing cracks in the area to be extracted.

[0087] After obtaining the building feature factors, geological environment factors, and meteorological environment factors corresponding to each building element, the feature factors are summarized to construct the feature vector corresponding to each building element in the area to be extracted.

[0088] In this embodiment, the extraction methods for the building characteristic factors, the geological environment factors, and the meteorological environment factors are the same as those in the above embodiments, and will not be repeated here.

[0089] S30, based on the feature factor vectors corresponding to each building element, the crack information of the building is extracted using the crack information extraction model to obtain the crack information of the building corresponding to each building element.

[0090] The crack information extraction model is a model obtained by using the method for obtaining the house crack information extraction model as described in the above embodiments.

[0091] In one specific embodiment, the region type of the region to be extracted is determined, and a crack information extraction model matching the region type is determined; feature information vectors corresponding to each building element in the region to be extracted are extracted; and the feature information vectors corresponding to each building element are respectively input into the crack information extraction model to obtain the building crack information corresponding to each building element.

[0092] The crack information refers to data characterizing whether cracks have occurred in the building and the extent of crack formation. In this embodiment, the crack information indicates whether there are cracks or not; those skilled in the art will understand that in other embodiments, the crack information may also be the number of cracks or the crack density, etc.

[0093] To address the technical problems existing in the prior art, embodiments of the present invention also provide an electronic device, please refer to... Figure 4 The diagram shows a schematic of the structure of the electronic device 5 of the present invention; the electronic device 5 includes: at least one processor, at least one communication interface, at least one memory and at least one communication bus.

[0094] In this embodiment, the number of processor, communication interface, memory, and communication bus is at least one, and the processor, communication interface, and memory communicate with each other through the communication bus. The memory 51 is used to store computer programs, and the processor 52 is used to execute the computer programs stored in the memory.

[0095] Optionally, the number of memories can be one or more, and the number of processors can be one or more.

[0096] Optionally, the processor in the electronic device loads one or more instructions corresponding to application processes into the memory according to the steps in the method for obtaining the house crack information extraction model as described above, and the processor runs the application stored in the memory to realize the functions in the method for obtaining the house crack information extraction model as described above, which will not be elaborated here.

[0097] It should be noted that memory includes, but is not limited to, random access memory (RAM), and may also include non-volatile memory, such as at least one disk storage device. Similarly, processors can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0098] Furthermore, this application embodiment also provides a computer storage medium carrying one or more computer programs. When the one or more computer programs are executed by an electronic device, the electronic device can implement the steps of the method for obtaining the house crack information extraction model provided in the above embodiments, and / or implement each step in the method for extracting house crack information provided in the above embodiments.

[0099] This invention also provides a computer-readable storage medium storing a computer program thereon, which, when called by a processor, implements the method for obtaining the house crack information extraction model as described in the above embodiments.

[0100] A computer-readable storage medium can be a tangible device capable of holding and storing instructions used by an instruction execution device. Computer-readable storage media can be, for example, (but not limited to) electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, and mechanical encoding devices.

[0101] The computer-readable program described herein can be downloaded from a computer-readable storage medium to various computing / processing devices, or downloaded via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network, to an external computer or external storage device. A network adapter card or network interface in each computing / processing device receives computer-readable program instructions from the network and forwards these instructions to the computer-readable storage medium in the respective computing / processing device.

[0102] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the invention. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in the present invention should still be covered by the claims of the present invention.

Claims

1. A method for obtaining a model for extracting information about building cracks, characterized in that, include: Build a housing dataset; The housing dataset contains housing feature information, geological environment feature information, meteorological environment feature information, and crack information corresponding to each housing element; Based on the housing dataset, model training is performed on each learning model in the pre-built model set; the performance of each trained learning model is compared, and the learning model with the best performance is selected as the target model for the working area.

2. The method for obtaining the building crack information extraction model according to claim 1, characterized in that, The method for constructing the housing dataset includes: Acquire housing survey data within the work area, and geological environmental data, meteorological environmental data, and housing crack data that match the housing survey data in terms of collection time and space; Based on the house crack data, crack information corresponding to each house element is extracted; according to the selected house characteristic factors, geological environment characteristic factors and meteorological environment characteristic factors, the house characteristic information, geological environment characteristic information and meteorological environment characteristic information corresponding to each house element are obtained from the house survey data, the geological environment data and the meteorological environment data respectively. The building feature information, geological environment feature information, and meteorological environment feature information corresponding to the same building element are summarized into a feature information vector; based on the feature information vectors corresponding to each building element and the building crack information, the building dataset is constructed; wherein... The building characteristic factor, the geological environment characteristic factor, and the meteorological environment characteristic factor are characteristic factors related to the generation of building cracks.

3. The method for obtaining the building crack information extraction model according to claim 2, characterized in that, The method for constructing the housing dataset also includes: Using correlation analysis, each of the aforementioned building characteristic factors, geological environment factors, and meteorological environment factors is screened based on the correlation of crack information to obtain building characteristic factors, geological environment factors, and meteorological environment factors with a correlation greater than a threshold; wherein, the correlation analysis method includes Pearson correlation coefficient and random forest importance analysis.

4. The method for obtaining the building crack information extraction model according to claim 2, characterized in that, The method for constructing the housing dataset also includes: Data preprocessing is performed on the building characteristic information, the geological environment characteristic information, and the meteorological environment characteristic information, respectively, including: A tag encoding method is used to convert the text information in the house feature information, the geological environment feature information, and the meteorological environment feature information into corresponding numerical information; and, The natural breakpoint method is used to transform the continuously distributed values ​​in the building feature information, the geological environment feature information, and the meteorological environment feature information into discrete interval distributed values.

5. The method for obtaining the building crack information extraction model according to claim 1, characterized in that, The performance comparison of each trained learning model includes: Based on the preset model evaluation metrics, extract the performance values ​​corresponding to each trained learning model; The performance values ​​of each learning model are comprehensively compared to obtain the performance comparison results for each learning model; among them, The model evaluation metrics include F1-score, precision, recall, and overall accuracy.

6. The method for obtaining the building crack information extraction model according to claim 1, characterized in that, Also includes: The work area is divided into several sub-regions according to region type, and a housing dataset corresponding to each sub-region is constructed. The housing dataset is partitioned to train the learning model based on the partitioned subsets of data. The step of training each learning model in the pre-built model set based on the housing dataset includes: training each learning model in the pre-built model set based on the housing dataset corresponding to each sub-region to obtain the trained learning model for each sub-region. Furthermore, the performance comparison of each trained learning model includes: comparing the performance of each sub-region with the trained learning model, and selecting the learning model with the best performance as the target model corresponding to that sub-region.

7. The method for obtaining the building crack information extraction model according to claim 6, characterized in that, The construction method of the housing dataset corresponding to each sub-region includes: Select housing characteristic factors, geological environment characteristic factors, and meteorological environment characteristic factors that are associated with the occurrence of housing cracks in the sub-region; Based on the housing survey data, the geological environment data, and the meteorological environment data, according to the housing characteristic factors, geological environment characteristic factors, and meteorological environment characteristic factors corresponding to the sub-region, the housing characteristic information, geological environment characteristic information, and meteorological environment characteristic information corresponding to each housing element in the sub-region are obtained accordingly. Based on the building feature information, geological environment feature information, meteorological environment feature information, and building crack information of each building element in the sub-region, the building data corresponding to each sub-region is constructed.

8. A method for extracting information about cracks in a building, characterized in that, include: Determine the area from which information on building cracks will be extracted; The building crack information extraction model is used to extract building crack information to obtain the building crack information corresponding to each building element. The building crack information extraction model is a model obtained by using the building crack information extraction model acquisition method as described in any one of claims 1 to 7.

9. An electronic device, characterized in that, It includes at least one processor and a memory connected to the processor, wherein: The memory is used to store computer programs; The processor is used to execute the computer program so that the electronic device can implement the method for obtaining the house crack information extraction model as described in any one of claims 1 to 7, and / or implement the steps of the method for extracting house crack information as described in claim 8.

10. A computer program product, characterized in that, The device includes computer-readable instructions that, when executed on an electronic device, cause the electronic device to implement the method for obtaining a housing crack information extraction model as described in any one of claims 1 to 7, and / or to implement the steps of the method for extracting housing crack information as described in claim 8.

Images