Rural house anti-seismic evaluation method based on deep learning

Abstract

The invention relates to the technical field of house anti-seismic evaluation, in particular to a rural house anti-seismic evaluation method based on deep learning, and the method comprises the following steps: S1, collecting the house age information of a rural house, and precisely recording the time span from the building of the house; performing image acquisition on the appearance of the house by using image acquisition equipment with multiple lenses; s2, the collected house age, appearance images, cracks and foundation settlement data are cleaned, and repeated, wrong or incomplete data records are removed. According to the method, a house aging degree analysis result, house cracks and foundation settlement data are used as an input model, anti-seismic performance data of houses in multi-region seismic areas under different seismic conditions are integrated, and more factors influencing the anti-seismic performance of the houses are considered by adopting a multi-source data fusion mode; the anti-seismic capability of the rural house can be evaluated more comprehensively and accurately, different data acquisition periods are formulated according to house conditions, and a reliable data basis is provided for subsequent evaluation.

Description

Technical Field

[0001] The present invention relates to the technical field of house seismic resistance assessment, and in particular to a rural house seismic resistance assessment method based on deep learning. Background Art

[0002] Rural houses refer to buildings built in rural areas for farmers to live and use for production and life. Their structural forms are diverse, the most common ones are brick-wood structures, brick-concrete structures, and some are civil structures or steel structures. In order to ensure the safety of people, the seismic resistance of rural houses needs to be evaluated. The purpose is to understand the seismic resistance of rural houses and provide a basis for seismic reinforcement or reconstruction of houses.

[0003] Existing technologies include the invention with publication number CN114792304A, which discloses a method for quickly calculating residents' seismic performance assessment parameters based on computer vision. The method uses a mobile device to capture house building data, and a series of computer vision methods are used to calculate the wall contours and holes on the walls of the captured data, and further obtain relevant seismic performance assessment parameters such as the number of floors of the house, the width and height of the wall, and the proportion of holes, and analyze and calculate data such as the seismic performance indicators of rural house buildings.

[0004] The current existing scheme uses the calculation of wall holes and conducts data analysis to obtain the earthquake resistance index of the house. However, the structure, age and degree of aging of rural houses vary. Therefore, it is difficult to accurately obtain earthquake resistance data based solely on wall holes, so improvements are needed. Summary of the Invention

[0005] The purpose of this invention is to solve the problem in the prior art that the existing scheme adopts the method of calculating wall holes and performing data analysis to obtain the seismic resistance index of the house. However, the structure, age and aging degree of rural houses are different. Therefore, it is difficult to accurately obtain seismic resistance data by relying solely on wall holes. A rural house seismic resistance assessment method based on deep learning is proposed.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a rural house earthquake resistance assessment method based on deep learning, comprising the following steps:

[0007] S1. Collect information on the age of rural houses and accurately record the time span since the houses were built; use image acquisition equipment equipped with multiple lenses to capture images of the house exteriors;

[0008] S2. Clean the collected data on house age, exterior images, cracks, and foundation settlement to remove duplicate, erroneous, or incomplete data records;

[0009] S3. Use a convolutional neural network based on the ResNet-50 architecture to identify weathering, peeling, and fading on the wall surface, damage and deformation on the roof, and analyze the degree of aging of the house.

[0010] S4. Input the collected house age information into a deep learning model based on a long short-term memory network (LSTM);

[0011] S5. Construct a neural network model based on the fusion of multi-layer perceptron (MLP) and convolutional neural network (CNN);

[0012] S6. The results of the building aging analysis are combined with data on building cracks and foundation settlement as input into the model, and the seismic performance data of buildings in multiple earthquake zones under different earthquake conditions are also incorporated.

[0013] S7. Use an independent test dataset to verify the trained earthquake resistance assessment model and calculate the performance indicators of the model on the test dataset;

[0014] S8. Apply the assessment results to the seismic retrofitting planning of rural houses and provide a basis for decision-making for rural house owners.

[0015] Preferably, in step S1, the house age information covers key parts such as the overall outline of the house, walls, roof, doors and windows, and high-resolution images of the house exterior are collected at least four different angles. For cracks on the outside of the house, a crack width gauge is used to accurately measure the width of obvious cracks, and a tape measure is used to assist in measuring the length of the cracks. For minor cracks, a magnifying glass or microscope with a magnification of not less than 50 times is used to assist in observation and recording. The crack positions are accurately marked using digital annotation software that matches the house CAD plan. At the same time, a level is used to measure the elevation change of the house foundation observation point to obtain the settlement amount. The overall inclination of the house is measured using equipment such as a total station. Data is collected on a monthly basis for houses that are found to be settled or tilted, and on a semi-annual basis for houses that are not found to be settled or tilted. The house aging analysis results are used together with the house crack and foundation settlement data as input to the model. At the same time, the seismic performance data of houses in multiple earthquake zones under different earthquake conditions are integrated. The multi-source data fusion method is used to consider more factors affecting the seismic performance of houses, which can more comprehensively and accurately evaluate the seismic capacity of rural houses.

[0016] Preferably, in step S1, damage status data of rural houses in multiple earthquake zones under different earthquake intensities are collected simultaneously, including information such as house structure type, damaged parts, and degree of damage, to construct a house status database for multiple earthquake zones. By accurately recording house age information and collecting house exterior images with multiple lenses and multiple angles at high resolution, covering key parts such as the overall outline of the house, walls, roofs, doors and windows, the exterior condition of the house can be fully reflected. For cracks, professional tools are used for precise measurement and recording, and the crack positions are located based on CAD plan drawings using digital annotation software to ensure the accuracy of crack data. In foundation settlement measurement, levels and total stations are used to obtain settlement and inclination, and different data collection cycles are formulated according to the house conditions, ensuring that focus is placed on problematic houses while taking into account the general situation, providing a rich and reliable data basis for subsequent evaluations.

[0017] Preferably, in step S2, pre-processing operations such as grayscale, noise reduction, and normalization are performed on the house appearance image to improve image quality and consistency, facilitate subsequent deep learning model processing, and standardize numerical data such as crack width, length, foundation settlement, etc. to ensure that the data are analyzed at the same scale. At the same time, the data in the earthquake area house condition database collected in step S1 are sorted and classified to make it compatible with the local collection data format to facilitate fusion analysis. Grayscale, noise reduction, normalization and other pre-processing operations are performed on the house appearance image to improve image quality and consistency, which is conducive to better feature extraction of the deep learning model. The numerical data is standardized and scaled to a unified interval to ensure that different types of data are analyzed at the same scale, thereby improving data processing efficiency and model training effect.

[0018] Preferably, in step S3, during the recognition process, the dilated convolution technique is used to expand the receptive field to better capture the subtle features of the house appearance, and the recognition results are optimized by combining the correlation data of house appearance aging and earthquake damage in multiple earthquake zones to highlight the key features related to house aging and seismic performance.

[0019] Preferably, in step S4, during the training process, the model is trained using more than 10,000 groups of rural house sample data of different house ages with detailed appearance feature annotations and corresponding actual aging degree assessments. At the same time, aging feature data related to house age in the multi-region earthquake zone house status database is introduced as an auxiliary. Through the back propagation algorithm, the mean square error (MSE) is used as the loss function, and the model parameters are continuously adjusted to optimize the model's ability to learn the mapping relationship between house age and appearance aging features. The model can preliminarily judge the basic degree of house aging based on the house age, and the house aging degree is set to A.

[0020] Preferably, in step S5, when determining the model structure, the number of neuron layers is determined to be 5 through at least 50 comparative tests, and the number of neurons in each layer is 256, 128, 64, 32, and 16 respectively. The initial value of the connection weight adopts the Xavier initialization method, and the hyperparameters of the model such as learning rate, number of iterations, regularization parameters, etc. are initialized and set to prepare for subsequent model training. At the same time, the seismic performance data of houses in the earthquake area under different earthquake conditions are feature engineered and converted into a format suitable for model input so that they can be integrated with local house data during the model training and evaluation process.

[0021] Preferably, in step S6, the crack-related feature vector is Where C1 is the average width of the crack, C2 is the total length of the crack, etc. The foundation settlement related characteristic vector is For example, S1 is the maximum differential settlement, S2 is the average settlement rate, etc. The characteristic vector of the seismic performance data of houses in multi-region earthquake areas under different earthquake conditions is For example, E1 is the average degree of damage to houses under a specific earthquake intensity in a certain area;

[0022] Combine input data into feature vectors

[0023] Let the actual seismic assessment level be y, and use One-Hot Encoding to represent it. If there are L seismic assessment levels, then y is an L-dimensional vector, with only the position corresponding to the actual level being 1 and the rest being 0.

[0024] Multilayer Perceptron (MLP) is used as the basic model, and the number of neurons in the input layer is d (d is equal to The hidden layer has h layers, and the number of neurons in the i-th hidden layer is n i , the number of neurons in the output layer is L (i.e. the number of earthquake resistance assessment levels). The input of the i-th hidden layer is With output The relationship is:

[0025]

[0026] Where W i is the weight matrix of the i-th layer, is the bias vector, σ is the activation function, such as the ReLU function σ(x) = max(0, x);

[0027] Output layer output

[0028]

[0029] The softmax function is defined as Convert the output into a probability distribution for each seismic assessment level.

[0030] According to the imbalance of housing data in multiple earthquake zones, oversampling or undersampling techniques are used to adjust the proportion of samples with different seismic assessment levels to improve the model's ability to identify minority samples.

[0031] A comprehensive assessment is conducted based on three factors: degree of aging, average crack width, and maximum differential settlement. The value range of each factor is normalized to between 0 and 1, and then a comprehensive seismic assessment index value is calculated based on these factors, as shown in the following table:

[0032]

[0033]

[0034] Preferably, in step S7, the accuracy of the seismic assessment model is:

[0035]

[0036] Where N is the number of samples in the validation set, Is an indicator function, when the condition in the brackets is true The value is 1, otherwise it is 0.

[0037] Preferably, in step S8, the test results of the model are displayed in the form of a web page, so that users can understand the earthquake resistance of the house in a timely manner.

[0038] Preferably, in step S8, suggestions for house reinforcement are pushed based on the earthquake resistance assessment data of the house, and a reinforced model is generated to facilitate the user to perform house reinforcement operations.

[0039] Compared with the prior art, the advantages and positive effects of the present invention are:

[0040] 1. In the present invention, the results of the house aging analysis are used together with the data on house cracks and foundation settlement as the input model. At the same time, the seismic performance data of houses in multiple earthquake zones under different earthquake conditions are integrated. The multi-source data fusion method is used to consider more factors affecting the seismic performance of houses, which can more comprehensively and accurately evaluate the seismic resistance of rural houses.

[0041] 2. In the present invention, by accurately recording the age information of the house and collecting high-resolution images of the house's appearance from multiple lenses and multiple angles, covering the overall outline of the house, walls, roof, doors and windows and other key parts, the appearance of the house can be fully reflected. For cracks, professional tools are used for precise measurement and recording, and the crack positions are located based on the CAD plan using digital annotation software to ensure the accuracy of the crack data. In the foundation settlement measurement, levels and total stations are used to obtain the settlement and inclination, and different data collection cycles are formulated according to the condition of the house, ensuring that focus is placed on problematic houses while taking into account the general situation, providing a rich and reliable data basis for subsequent evaluation.

[0042] 3. In the present invention, the house exterior image is pre-processed by grayscale conversion, noise reduction, normalization, etc. to improve image quality and consistency, which is conducive to better feature extraction by the deep learning model. The numerical data is standardized and scaled to a unified interval to ensure that different types of data are analyzed at the same scale, thereby improving data processing efficiency and model training effect.

[0043] 4. In the present invention, damage status data of rural houses in multiple earthquake zones under different earthquake intensities are collected to construct a multi-region earthquake zone house status database. This database provides rich reference data for aging feature analysis, model training and comprehensive evaluation, enhances the universality and adaptability of the evaluation method, and can better meet the seismic assessment needs of rural houses in different regions. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 This is a flow chart of the rural house seismic assessment method based on deep learning proposed in the present invention. DETAILED DESCRIPTION

[0045] See also Figure 1 The present invention provides a technical solution: a rural house earthquake resistance assessment method based on deep learning, comprising the following steps:

[0046] S1. Collect information on the age of rural houses and accurately record the time span since the houses were built; use image acquisition equipment equipped with multiple lenses to capture images of the house exteriors;

[0047] S2. Clean the collected data on house age, exterior images, cracks, and foundation settlement to remove duplicate, erroneous, or incomplete data records;

[0048] S3. Use a convolutional neural network based on the ResNet-50 architecture to identify weathering, peeling, and fading on the wall surface, damage and deformation on the roof, and analyze the degree of aging of the house.

[0049] S4. Input the collected house age information into a deep learning model based on a long short-term memory network (LSTM);

[0050] S5. Construct a neural network model based on the fusion of multi-layer perceptron (MLP) and convolutional neural network (CNN);

[0051] S6. The results of the building aging analysis are combined with data on building cracks and foundation settlement as input into the model, and the seismic performance data of buildings in multiple earthquake zones under different earthquake conditions are also incorporated.

[0052] S7. Use an independent test dataset to verify the trained earthquake resistance assessment model and calculate the performance indicators of the model on the test dataset;

[0053] S8. Apply the assessment results to the seismic retrofitting planning of rural houses and provide a basis for decision-making for rural house owners.

[0054] Specifically, in step S1, the house age information covers key parts such as the overall outline of the house, walls, roof, doors and windows, and high-resolution images of the house exterior are collected from at least four different angles. For cracks on the outside of the house, a crack width gauge is used to accurately measure the width of obvious cracks, and a tape measure is used to assist in measuring the crack length. For minor cracks, a magnifying glass or microscope with a magnification of not less than 50 times is used to assist in observation and recording. The crack positions are accurately marked using digital annotation software that matches the house CAD plan. At the same time, a level is used to measure the elevation change of the house foundation observation point to obtain the settlement amount. The overall inclination of the house is measured using equipment such as a total station. Data is collected on a monthly basis for houses that are found to be settled or tilted, and on a semi-annual basis for houses that are not found to be settled or tilted. The house aging analysis results are used together with the house crack and foundation settlement data as input to the model. At the same time, the seismic performance data of houses in multiple earthquake zones under different earthquake conditions are incorporated. The multi-source data fusion method is used to consider more factors affecting the seismic performance of houses, which can more comprehensively and accurately evaluate the seismic capacity of rural houses.

[0055] Specifically, in step S1, damage status data of rural houses in multiple earthquake zones under different earthquake intensities are collected simultaneously, including information such as house structure type, damaged parts, and degree of damage, to build a house status database for multiple earthquake zones. By accurately recording house age information and collecting high-resolution images of house exteriors from multiple lenses and angles, covering key parts such as the overall outline of the house, walls, roofs, doors and windows, the appearance of the house can be fully reflected. For cracks, professional tools are used for precise measurement and recording, and digital annotation software is used to locate the crack position based on the CAD plan to ensure the accuracy of the crack data. In foundation settlement measurement, levels and total stations are used to obtain settlement and inclination, and different data collection cycles are formulated according to the house conditions, ensuring that both focus is placed on problematic houses and the general situation is taken into account, providing a rich and reliable data basis for subsequent evaluations.

[0056] In this embodiment: In step S2, the house exterior image is subjected to pre-processing operations such as grayscale, noise reduction, and normalization to improve image quality and consistency, which is convenient for subsequent deep learning model processing. Numerical data such as crack width, length, foundation settlement, etc. are standardized to ensure that the data are analyzed at the same scale. At the same time, the data in the earthquake area house condition database collected in step S1 are sorted and classified to make it compatible with the local collection data format to facilitate fusion analysis. The house exterior image is subjected to pre-processing operations such as grayscale, noise reduction, and normalization to improve image quality and consistency, which is conducive to better feature extraction by the deep learning model. The numerical data is standardized and scaled to a unified interval to ensure that different types of data are analyzed at the same scale, thereby improving data processing efficiency and model training effect.

[0057] Specifically, in step S3, during the recognition process, the dilated convolution technique is used to expand the receptive field to better capture the subtle features of the house's appearance. Combined with the correlation data of house appearance aging and earthquake damage in multiple earthquake zones, the recognition results are optimized to highlight the key features related to house aging and seismic resistance.

[0058] Specifically, in step S4, during the training process, the model uses more than 10,000 groups of rural house sample data of different house ages with detailed appearance feature annotations and corresponding actual aging degree assessments for training. At the same time, aging feature data related to house age in the house status database of multiple earthquake areas is introduced as an auxiliary. Through the back propagation algorithm, with the mean square error (MSE) as the loss function, the model parameters are continuously adjusted to optimize the model's ability to learn the mapping relationship between house age and appearance aging features. It can preliminarily judge the basic degree of house aging based on the house age, and set the house aging degree as A.

[0059] In this embodiment: In step S5, when determining the model structure, the number of neuron layers is determined to be 5 through at least 50 comparative tests, and the number of neurons in each layer is 256, 128, 64, 32, and 16 respectively. The initial value of the connection weight adopts the Xavier initialization method, and the hyperparameters of the model such as the learning rate, number of iterations, and regularization parameters are initialized and set to prepare for subsequent model training. At the same time, the seismic performance data of houses in the earthquake area under different earthquake conditions are feature engineered and converted into a format suitable for model input so that it can be integrated with local house data during the model training and evaluation process.

[0060] Preferably, in step S6, the crack-related feature vector is Where C1 is the average width of the crack, C2 is the total length of the crack, etc. The foundation settlement related characteristic vector is For example, S1 is the maximum differential settlement, S2 is the average settlement rate, etc. The characteristic vector of the seismic performance data of houses in multi-region earthquake areas under different earthquake conditions is For example, E1 is the average degree of damage to houses under a specific earthquake intensity in a certain area;

[0061] Combine input data into feature vectors

[0062] Let the actual seismic assessment level be y, and use One-Hot Encoding to represent it. If there are L seismic assessment levels, then y is an L-dimensional vector, with only the position corresponding to the actual level being 1 and the rest being 0.

[0063] Multilayer Perceptron (MLP) is used as the basic model, and the number of neurons in the input layer is d (d is equal to The hidden layer has h layers, and the number of neurons in the i-th hidden layer is n i , the number of neurons in the output layer is L (i.e. the number of earthquake resistance assessment levels). The input of the i-th hidden layer is With output The relationship is:

[0064]

[0065] Where W i is the weight matrix of the i-th layer, is the bias vector, σ is the activation function, such as the ReLU function σ(x) = max(0, x);

[0066] Output layer output

[0067]

[0068] The softmax function is defined as Convert the output into a probability distribution for each seismic assessment level.

[0069] According to the imbalance of housing data in multiple earthquake zones, oversampling or undersampling techniques are used to adjust the proportion of samples with different seismic assessment levels to improve the model's ability to identify minority samples.

[0070] A comprehensive assessment is conducted based on three factors: degree of aging, average crack width, and maximum differential settlement. The value range of each factor is normalized to between 0 and 1, and then a comprehensive seismic assessment index value is calculated based on these factors, as shown in the following table:

[0071]

[0072] Preferably, in step S7, the accuracy of the seismic assessment model is:

[0073]

[0074] Where N is the number of samples in the validation set, Is an indicator function, when the condition in the brackets is true The value is 1, otherwise it is 0.

[0075] Specifically, in step S8, the test results of the model are displayed in the form of a web page, so that users can understand the earthquake resistance status of the house in a timely manner.

[0076] Specifically, in step S8, suggestions for house reinforcement are pushed based on the earthquake resistance assessment data of the house, and a reinforced model is generated to facilitate the user to perform house reinforcement operations.

Claims

1. A rural house seismic assessment method based on deep learning, characterized by: The following steps are involved: S1. Collect the age information of rural houses and capture images of the house appearance; S2. Clean the collected data on house age, exterior images, cracks, and foundation settlement to remove duplicate, erroneous, or incomplete data records; S3: Use a convolutional neural network based on the ResNet-50 architecture and dilated convolution technology to identify the appearance of the house and analyze the degree of aging; S4. Input the house age information into the long short-term memory network (LSTM) model; S5. Construct a neural network model based on the fusion of multi-layer perceptron (MLP) and convolutional neural network (CNN); S6. Input the building aging analysis results, crack and foundation settlement data, along with the seismic performance data of buildings in multiple regions, and train the model; S7. Validate the model using an independent test set and calculate performance metrics. S8. Apply the assessment results to the seismic retrofitting planning of rural houses and provide a basis for decision-making for rural house owners.

2. The rural house seismic assessment method based on deep learning according to claim 1 is characterized by: In step S1, the house age information covers key parts such as the overall outline of the house, walls, roof, doors and windows, and high-resolution images of the house exterior are collected from at least four different angles. For cracks on the outside of the house, a crack width gauge is used to accurately measure the width of obvious cracks, and a tape measure is used to assist in measuring the length of the cracks.

3. The rural house seismic assessment method based on deep learning according to claim 1 is characterized by: In step S1, damage status data of rural houses in multiple earthquake zones under different earthquake intensities are collected simultaneously, including information such as house structure type, damaged parts, and damage degree, to construct a multi-region earthquake zone house status database.

4. The rural house seismic assessment method based on deep learning according to claim 1, characterized in that: In step S2, preprocessing operations such as grayscale conversion, noise reduction, and normalization are performed on the house exterior image to improve image quality and consistency, facilitate subsequent processing by the deep learning model, and standardize numerical data such as crack width, length, and foundation settlement.

5. The rural house seismic assessment method based on deep learning according to claim 1, characterized in that: In step S3, during the recognition process, the dilated convolution technique is used to expand the receptive field to better capture the subtle features of the house appearance.

6. The rural house seismic assessment method based on deep learning according to claim 1, characterized in that: In step S4, during the training process of the model, aging feature data related to house age in the multi-region earthquake zone house condition database is introduced as an aid.

7. The rural house seismic assessment method based on deep learning according to claim 1, characterized in that: In step S5, the model's hyperparameters, such as the learning rate, number of iterations, and regularization parameters, are initialized and set to prepare for subsequent model training. At the same time, the seismic performance data of houses in the earthquake zone under different earthquake conditions are feature engineered and converted into a format suitable for model input so that they can be integrated with local house data during model training and evaluation.

8. The rural house seismic assessment method based on deep learning according to claim 1, characterized in that: In step S8, the test results of the model are displayed in the form of a web page, so that users can understand the earthquake resistance status of the house in a timely manner.

9. The rural house seismic assessment method based on deep learning according to claim 1, characterized in that: In step s8, a suggestion for house reinforcement is pushed based on the earthquake resistance assessment data of the house, and a reinforced model is generated to facilitate the user to perform house reinforcement operations.

Images