A building settlement detection method and related products suitable for construction engineering quality

Through multi-source data fusion and image processing technology, a support vector regression model is built, which realizes high-precision real-time monitoring of building settlement, solves the problems of low accuracy and low efficiency in the existing technology, and improves the reliability and timeliness of detection.

CN118758253BActive Publication Date: 2024-11-15SICHUAN HELI CONSTR ENG INSPECTION & APPRAISAL CONSULTING CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411241716.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-05
Publication Date
2024-11-15
Estimated Expiration
2044-09-05

AI Technical Summary

Technical Problem

The existing building settlement detection methods have low accuracy and low efficiency, making them difficult to monitor in real time, and cannot meet the high requirements for settlement monitoring in modern high-rise buildings.

Method used

Using multi-source data fusion and image processing technology, data is collected through lidar, inertial detection unit, BDS acquisition unit and multi-spectral camera, image processing and feature extraction are carried out, support vector regression model is constructed, and high-precision real-time monitoring of building settlement is achieved.

Benefits of technology

It improves the accuracy and efficiency of detection, enhances the reliability of data, can accurately reflect the settlement trend of buildings, reduces manual intervention, improves the continuity and timeliness of detection, and early detection and prevents safety hazards caused by building settlement.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118758253B_ABST
    Figure CN118758253B_ABST
Patent Text Reader

Abstract

The present invention relates to the field of intelligent detection of construction projects, and specifically to a building settlement detection method and related products suitable for the quality of construction projects. The method comprises acquiring training data, processing the acquired data through image processing technology, and obtaining a denoised difference image after processing, then constructing an SVR model by extracting the feature vector of the denoised difference image, and iteratively training through training samples to finally obtain a trained support vector regression model. The present invention realizes high-precision real-time monitoring of building settlement through multi-source data fusion and image processing technology, and uses the SVR model to predict settlement, which can accurately reflect the settlement trend of the building.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of intelligent detection of construction projects, and in particular to a building settlement detection method and related products suitable for construction project quality. Background Art

[0002] With the acceleration of urbanization, the number of high-rise buildings has increased rapidly. However, due to uneven foundation settlement, mistakes in the construction process or material quality problems, high-rise buildings may experience different degrees of settlement during use. Excessive settlement will affect the structural safety and service life of the building. Therefore, real-time monitoring of building settlement is particularly important.

[0003] Traditional settlement monitoring methods mainly rely on manual measurement, which is not only time-consuming and labor-intensive, but also has low measurement accuracy and frequency, making it difficult to meet the high requirements of settlement monitoring for modern high-rise buildings. With the development of sensor technology, computer technology and digital image processing technology, high-precision and automated settlement monitoring methods based on multi-source data fusion have gradually become a research hotspot.

[0004] Therefore, it is necessary to study a detection method that integrates multiple measurement methods and combines advanced data processing and analysis methods to monitor building settlement in real time and accurately. Summary of the invention

[0005] The technical problem to be solved by the present invention is that the existing building settlement detection methods have low accuracy, low efficiency, and difficulty in real-time monitoring. The purpose is to provide a building settlement detection method and related products suitable for the quality of construction projects, which realizes rapid settlement detection of buildings, improves detection efficiency and accuracy, and ensures the safety and service life of building structures.

[0006] The present invention is achieved through the following technical solutions:

[0007] A building settlement detection method applicable to construction engineering quality, comprising:

[0008] Obtain training data, which includes collected data and actual data;

[0009] The collected data is processed by image processing technology to obtain the difference images of building settlement at different times, and the accurate final difference images are obtained through texture change, shape change and similarity;

[0010] Decomposing the final difference image into low-frequency sub-bands and high-frequency sub-bands of multiple scales and directions, and performing adaptive threshold processing on the high-frequency sub-bands to obtain a denoised difference image;

[0011] Extract the feature vector of the denoised difference image and form a training sample with the actual data;

[0012] Construct the SVR model and obtain the final model parameters through iterative training of training samples;

[0013] Obtain the trained support vector regression model through the final model parameters;

[0014] The real-time collected data is obtained, and after processing the real-time collected data and the collected data, they are input into the support vector regression model to obtain the prediction results.

[0015] Specifically, the data collection equipment used includes: laser radar, inertial detection unit, BDS collection unit and multispectral camera; the types of data collected include: three-dimensional point cloud information, posture information, position information and multispectral image information; the actual data is the actual measured building settlement value;

[0016] Methods for obtaining collected data include:

[0017] Collect corresponding data through acquisition equipment, and fuse multi-source data to obtain fused data;

[0018] Set quality thresholds and calculate the comprehensive quality score of the fused data;

[0019] If the comprehensive quality score is lower than the quality threshold, the acquisition parameters are adjusted; if the comprehensive quality score is not lower than the quality threshold, the fluctuation intensity of the fused data is calculated;

[0020] Set the fluctuation threshold. If the fluctuation intensity is greater than the fluctuation threshold, re-acquire the parameters. If the fluctuation intensity is not greater than the fluctuation threshold, output the fused data as the collected data.

[0021] Specifically, PCA is performed on the three-dimensional point cloud data to obtain a three-dimensional point cloud feature vector, the posture information is used as the posture feature vector, the position information is used as the position feature vector, and the multispectral image information is extracted through CNN to obtain a multispectral feature vector;

[0022] The three-dimensional point cloud feature vector, posture feature vector, position feature vector and multi-spectral feature vector are standardized to obtain the corresponding standardized vector , , and ;

[0023] Get fused data ,in, , , and is the weight of the corresponding information;

[0024] Comprehensive quality score of fused data ,in, is the noise level, For data integrity, For equipment stability, is the weight corresponding to the noise level, is the weight corresponding to data integrity, is the weight corresponding to the equipment stability;

[0025] Acquisition parameters include sampling rate and resolution. , then adjust the sampling rate and resolution, ,in, is the quality threshold, is the sampling rate adjustment factor, is the resolution adjustment factor, is the adjustment amplitude of the sampling rate, is the adjustment range of resolution;

[0026] Fluctuation intensity ,in, is the data collected by the laser radar at adjacent collection times, is the data collected by the inertial measurement unit at adjacent collection times, is the collected data of the BDS collection unit at adjacent collection times, It is the data collected by the multispectral camera at adjacent collection moments.

[0027] Specifically, the method for obtaining the difference image includes:

[0028] The fused data is converted into an image form by using image processing technology to generate an initial image;

[0029] Choose any two initial images at different times;

[0030] Extracting texture features of the initial image through the gray level co-occurrence matrix and determining the texture changes of two different initial images;

[0031] Extracting shape features of the initial image through an active contour model and determining shape changes of two different initial images;

[0032] Determine the similarity between two images by using the coherence coefficient method;

[0033] The texture change, shape change and similarity are weighted and summed to obtain the difference image.

[0034] Optionally, texture features Contrast ,energy ,entropy and uniformity ,in, Indicates the gray value is and The co-occurrence frequency of pixel pairs at specific distances and angles;

[0035] Pixels Texture changes , For time point No. Texture features at pixel points The value of For time point No. Texture features at pixel points The value of For the The weight of the texture features, They represent contrast, energy, entropy, and uniformity, respectively;

[0036] The methods for obtaining shape features include:

[0037] Initialize closed curve ,in represents a point on the curve;

[0038] Determine the total energy function ,in, is the internal energy used to constrain the smoothness of the curve, is the external energy used to attract the curve close to the edge of the image, is the weight parameter used to control the elasticity of the curve, is the weight parameter used to control the rigidity of the curve;

[0039] By iterative optimization, the total energy function is minimized so that the curve gradually approaches the edge of the image. The optimization function is: ,in, is the learning rate, For the The position of the curve at the iteration, For the The position of the curve at the iteration;

[0040] After multiple iterations, the curve Converge to the edge of the image and extract the obtained shape features ;

[0041] Pixels Shape changes , For time point The shape features are at the pixel point The value of For time point The shape features are at the pixel point The value of

[0042] Coherence coefficient , and is the area range of the pixel point, and For time point and No. The pixel value of the pixel;

[0043] Get the final difference image ,in, and is the weight coefficient for fusing texture and shape changes with the coherence coefficient.

[0044] Specifically, the method for obtaining the denoised difference image includes:

[0045] Decomposing the final difference image into low-frequency sub-bands and high-frequency sub-bands of multiple scales and directions;

[0046] Divide the high frequency subband into multiple local image blocks, and calculate the variance of each local image block;

[0047] A first threshold and a second threshold are set. If the variance corresponding to the local image block is less than the first threshold, the pixel points of the local image block are subjected to soft threshold processing; if the local image block is not less than the second threshold, the pixel points of the local image block are subjected to Bayesian shrinkage processing; if it is the other case, the pixel points of the local image block are subjected to hard threshold processing;

[0048] The threshold processing results of different local image blocks are combined to obtain the final threshold processing result;

[0049] Complete the processing of all pixel points of all high-frequency sub-bands to obtain multiple high-frequency sub-bands after noise reduction;

[0050] The low-frequency subband and the denoised high-frequency subband are reconstructed to obtain a denoised difference image.

[0051] Optionally, the final difference image is decomposed, ,in, is the low frequency subband, For the High frequency sub-bands;

[0052] Set the first threshold and the second threshold , , and make a judgment on each local image block;

[0053] like , then use soft threshold processing to obtain the soft threshold processing result ,in, is the soft threshold parameter, For the In the local image block pixels, is the variance corresponding to the local image block;

[0054] like , then use hard threshold processing to obtain the hard threshold processing result , is the hard threshold parameter;

[0055] like , then use Bayesian Shrinkage processing to obtain the Bayesian processing result , is the noise variance;

[0056] Reconstruct the denoised difference image , for High frequency subband after denoising.

[0057] Specifically, the method of training the SVR model includes:

[0058] Calculate the mean and standard deviation of the feature vector, and standardize the feature vector based on the mean and standard deviation to obtain a standardized feature vector;

[0059] Determine the kernel function of the SVR model, and determine the objective function, constraints and loss function of the SVR model optimization;

[0060] Randomly initialize model parameters;

[0061] Determine the weight vector gradient and bias gradient for optimization;

[0062] Iteratively update the model parameters until the convergence condition is reached, output the final parameter model, and obtain the trained support vector regression model based on the parameter model;

[0063] The objective function is , is the weight vector, is the bias, and is the slack variable, is the weight coefficient of the error term, is the kernel function mapping, is the weight coefficient in multi-task learning, is the shared weight vector in multi-task learning, For the The weight vector of each task, is the number of tasks, is the loss function, is the L1 norm, is the L2 norm, is the sample size, is the weight coefficient of L2 norm regularization, is the weight coefficient of L1 norm regularization, For the The true value of the samples, For the The input value of samples;

[0064] The constraints are ,in, To control the complexity of the model, For the slack variables, For all sample indexes All apply;

[0065] The loss function is ,in, is the true value, is the model prediction value, is the smoothing parameter of the loss function;

[0066] Randomly initialize model parameters , , , and shared weight vectors in multi-task learning ;

[0067] The weight vector gradient is ;

[0068] The bias gradient is ,in, is the gradient of the objective function with respect to the weight vector, is the gradient of the target vector with respect to the bias, is the gradient of the loss function with respect to the weight vector, is the gradient of the loss function with respect to the bias;

[0069] Update the model parameter expression: .

[0070] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the building settlement detection method applicable to construction engineering quality as described above.

[0071] A computer program product includes a computer program / instruction, which, when executed by a processor, implements the building settlement detection method applicable to construction engineering quality as described above.

[0072] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0073] The method provided by the present invention comprises acquiring training data, and processing the acquired data by image processing technology, and performing denoising on the difference image after processing, and then constructing an SVR model by extracting the feature vector of the denoised difference image, and iteratively training through training samples, and finally obtaining a trained support vector regression model;

[0074] The multi-source data fusion and image processing technology of the present invention realizes high-precision real-time monitoring of building settlement, which not only improves the accuracy and efficiency of detection, but also enhances the reliability of data through adaptive threshold processing and denoising technology. At the same time, the SVR model is used for settlement prediction, which can accurately reflect the settlement trend of the building, reduce human intervention, improve the continuity and timeliness of detection, and contribute to the early discovery and prevention of safety hazards caused by building settlement. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] The accompanying drawings illustrate exemplary embodiments of the present invention and, together with the description thereof, are used to explain the principles of the present invention. These drawings are included to provide a further understanding of the present invention, and the accompanying drawings are included in and constitute a part of this specification and do not constitute a limitation of the embodiments of the present invention.

[0076] Figure 1 It is a flow chart of a building settlement detection method applicable to construction engineering quality according to the present invention. DETAILED DESCRIPTION

[0077] To make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and implementation methods. It is understood that the specific implementation methods described herein are only used to explain the relevant content, rather than to limit the present invention.

[0078] It should also be noted that, for the convenience of description, only the parts related to the present invention are shown in the drawings.

[0079] In the absence of conflict, the embodiments and features of the embodiments of the present invention may be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0080] Embodiment 1, as Figure 1 As shown, this embodiment provides a building settlement detection method applicable to construction engineering quality, including:

[0081] The first step is to obtain training data, which includes collected data and actual data.

[0082] The collected data refers to the building settlement information obtained through various sensors and equipment (such as lidar, inertial measurement unit, BDS acquisition unit and multispectral camera); the actual data refers to the actual building settlement value obtained through traditional measurement methods.

[0083] In the second step, the collected data is processed through image processing technology to obtain difference images of building settlement at different times, and the accurate final difference image is obtained through texture change, shape change and similarity.

[0084] The image processing technology in this application includes a series of image analysis methods, such as texture analysis, shape feature extraction and similarity calculation, which are used to extract useful information from images and generate difference images.

[0085] In the third step, the final difference image is decomposed into low-frequency sub-bands and high-frequency sub-bands of multiple scales and directions, and the high-frequency sub-bands are subjected to adaptive threshold processing to obtain the denoised difference image.

[0086] Subband decomposition refers to dividing an image into different sub-images according to its frequency components. The low-frequency sub-band contains the overall structural information of the image, while the high-frequency sub-band contains the detail information. Adaptive threshold processing is performed on the high-frequency sub-band to automatically adjust the threshold according to the local features of the image, filter out noise, retain useful information, and finally obtain a denoised difference image.

[0087] The fourth step is to extract the feature vector of the denoised difference image and form a training sample with the actual data; the feature vector refers to a set of values ​​that can describe certain important properties of the image and is used to represent the characteristics of the image.

[0088] The fifth step is to build the SVR model and obtain the final model parameters through iterative training of training samples; support vector regression is used for prediction and regression analysis. The SVR model is iteratively trained using training samples, and the model parameters are continuously adjusted until the optimal model parameters are obtained.

[0089] The sixth step is to obtain the trained support vector regression model through the final model parameters; this model can be used to predict the settlement of the building.

[0090] In the application step, real-time collected data is obtained, and after processing the real-time collected data and the collected data, they are input into the support vector regression model to obtain the prediction results.

[0091] In practical applications, real-time data is acquired, and historical data recorded at any time in the previous period is selected, and the second and fourth steps are performed on the real-time data and historical data. After obtaining the feature vector of the difference image corresponding to the real-time data and the historical data, it is input into the trained support vector regression model, and the settlement prediction result of the building is obtained after processing. In this way, the settlement of the building can be monitored in real time, providing reliable data support for building safety.

[0092] Embodiment 2: This embodiment provides a method for obtaining collected data.

[0093] The data collection equipment used includes: LiDAR, inertial detection unit, BDS collection unit and multispectral camera; the types of data collected include: three-dimensional point cloud information, posture information, position information and multispectral image information; the actual data is the actual measured building settlement value.

[0094] The laser radar collects the three axial position information (xyz coordinates) of the target, the inertial measurement unit collects the elevation, side angle and heading information of the target, the BDS acquisition unit collects the longitude, latitude and elevation information of the building, and the multispectral camera collects multispectral image information (near infrared spectrum, red spectrum, green spectrum, blue spectrum) of the building surface.

[0095] Before data collection, initialize the device first. According to the selected data collection device, initialize the working parameters of each device to ensure the normal operation of the device. Then perform data synchronization and time reference. In order to ensure the synchronization of data collection, all devices use a unified time reference. Set the time parameters of each device to synchronize it with the reference clock.

[0096] Perform PCA processing on the three-dimensional point cloud data to obtain the three-dimensional point cloud feature vector, use the posture information as the posture feature vector, use the position information as the position feature vector, and extract the multispectral image information through CNN to obtain the multispectral feature vector;

[0097] Standardize the 3D point cloud feature vector, posture feature vector, position feature vector and multispectral feature vector to ensure that the data is on the same scale and obtain the corresponding standardized vector , , and ;

[0098] Collect the corresponding data through the acquisition equipment, and fuse the multi-source data to obtain fused data ,in, , , and is the weight of the corresponding information. The weight value is obtained by iteratively using optimization algorithms (such as genetic algorithm, particle swarm optimization algorithm, etc.).

[0099] Setting quality thresholds , and calculate the comprehensive quality score of the fused data ,in, is the noise level (calculate the ratio of noise in the data to assess the noise level, signal-to-noise ratio), Data integrity (by calculating the integrity of the collected data to assess whether the data is lost or incomplete, packet loss rate), For device stability (by evaluating the working status of the acquisition device to determine whether the device is stable, device jitter and drift), is the weight corresponding to the noise level, is the weight corresponding to data integrity, is the weight corresponding to the equipment stability. Use multi-objective optimization methods, such as AHP (Analytic Hierarchy Process) or weighted linear regression, to comprehensively consider multiple quality indicators and determine the optimal weight value.

[0100] If the comprehensive quality score is lower than the quality threshold, the acquisition parameters are adjusted; if the comprehensive quality score is not lower than the quality threshold, the fluctuation intensity of the fused data is calculated.

[0101] Acquisition parameters include sampling rate and resolution. , then adjust the sampling rate and resolution, ,in, is the quality threshold, is the sampling rate adjustment factor, is the resolution adjustment factor, is the adjustment amplitude of the sampling rate, The adjustment range of resolution.

[0102] Set the fluctuation threshold. If the fluctuation intensity is greater than the fluctuation threshold, re-acquire the parameters. If the fluctuation intensity is not greater than the fluctuation threshold, output the fused data as the collected data. That is, if the fluctuation intensity is large, it means that the data is unstable and needs to be collected again.

[0103] Fluctuation intensity ,in, is the data collected by the laser radar at adjacent collection times, is the data collected by the inertial measurement unit at adjacent collection times, is the collected data of the BDS collection unit at adjacent collection times, It is the data collected by the multispectral camera at adjacent collection moments.

[0104] Embodiment 3: This embodiment provides a method for obtaining a difference image.

[0105] The fused data is converted into an image form through image processing technology to generate an initial image; in this embodiment, multi-source data (such as three-dimensional point cloud, posture information, position information and multi-spectral image information) is converted into a two-dimensional image representation through image processing technology

[0106] Select any two initial images at different times; by selecting two initial images with a time interval, the settlement changes of the building during this period can be observed and analyzed, thereby evaluating the trend and degree of building settlement.

[0107] The texture features of the initial image are extracted through the gray level co-occurrence matrix, and the texture changes of two different initial images are determined; the gray level co-occurrence matrix is ​​a matrix used to describe the spatial relationship between gray levels in an image. By calculating the gray level co-occurrence matrix of the initial image, the texture features of the image can be extracted. In this embodiment, by extracting and comparing the texture features of the images at different times, the texture changes of the building between these times are determined, reflecting the changes in details during the settlement process.

[0108] The shape features of the initial image are extracted through the active contour model, and the shape changes of two different initial images are determined; the active contour model is used for image edge detection technology, and an initial contour gradually converges to the edge of the image through iteration. The shape features of the initial image can be extracted through the active contour model. That is, the edge information in the image is extracted, and the shape feature changes of the images at different times are compared to reflect the shape changes during the building settlement process.

[0109] The similarity between two images is determined by the coherence coefficient method; by calculating the coherence coefficient between two initial images at different times, their similarity can be determined. The higher the coherence coefficient, the more similar the two images are; the lower the coherence coefficient, the greater the difference between the two images.

[0110] The texture change, shape change and similarity are weighted and summed to obtain the difference image. Through this weighted summation method, the changes of the initial image at different times can be comprehensively reflected to obtain an accurate difference image.

[0111] A specific implementation of the above method is provided below:

[0112] Texture features Contrast ,energy ,entropy and uniformity ,in, Indicates the gray value is and The co-occurrence frequency of pixel pairs at specific distances and angles; by comparing images at two different times, the texture changes of the building between these times can be determined, reflecting the detailed changes in the settlement process.

[0113] Pixels Texture changes , For time point No. Texture features at pixel points The value of For time point No. Texture features at pixel points The value of For the The weight of the texture features, They represent contrast, energy, entropy, and uniformity, respectively;

[0114] The methods for obtaining shape features include:

[0115] Initialize closed curve ,in represents a point on the curve;

[0116] Determine the total energy function ,in, is the internal energy used to constrain the smoothness of the curve, is the external energy used to attract the curve close to the edge of the image, is the weight parameter used to control the elasticity of the curve, is the weight parameter used to control the rigidity of the curve;

[0117] By iterative optimization, the total energy function is minimized so that the curve gradually approaches the edge of the image. The optimization function is: ,in, is the learning rate, For the The position of the curve at the iteration, For the The position of the curve at the iteration;

[0118] After multiple iterations, the curve Converge to the edge of the image and extract the obtained shape features ;

[0119] Pixels Shape changes , For time point The shape features are at the pixel point The value of For time point The shape features are at the pixel point The value of

[0120] Coherence coefficient , and is the area range of the pixel point, and For time point and No. The pixel value of the pixel;

[0121] Get the final difference image ,in, and is the weight coefficient for fusing texture and shape changes with the coherence coefficient.

[0122] Embodiment 4: This embodiment provides a method for obtaining a denoised difference image.

[0123] Decomposing the final difference image into low-frequency sub-bands and high-frequency sub-bands of multiple scales and directions; ,in, is the low frequency subband, For the The difference image is decomposed into sub-bands of multiple scales and directions using wavelet transform or other multi-scale analysis techniques. The low-frequency sub-band contains the overall structural information of the image, while the high-frequency sub-band contains the detailed information of the image.

[0124] The high-frequency subband is divided into multiple local image blocks, and the variance of each local image block is calculated; variance is an indicator to measure the degree of change of pixel values ​​within an image block. The larger the variance, the richer the detail information of the image block. By calculating the variance of the local image block, the appropriate denoising method can be selected according to the detail level of the image block.

[0125] Set the first threshold and the second threshold , , and make a judgment on each local image block;

[0126] If the variance corresponding to the local image block is less than the first threshold, the pixels of the local image block are subjected to soft threshold processing; if the local image block is not less than the second threshold, the pixels of the local image block are subjected to Bayesian shrinkage processing; if it is the other case, the pixels of the local image block are subjected to hard threshold processing;

[0127] like , then use soft threshold processing to obtain the soft threshold processing result ,in, is the soft threshold parameter, For the In the local image block pixels, is the variance corresponding to the local image block; soft threshold processing reduces the impact of noise by reducing the coefficients below the threshold.

[0128] like , then use hard threshold processing to obtain the hard threshold processing result , is the hard threshold parameter; hard threshold processing eliminates noise by directly setting coefficients less than the threshold to zero.

[0129] like , then use Bayesian Shrinkage processing to obtain the Bayesian processing result , is the noise variance; Bayesian Shrinkage processing denoises the high variance area according to the statistical characteristics of the noise.

[0130] The threshold processing results of different local image blocks are combined to obtain the final threshold processing result;

[0131] The processing of all pixel points of all high-frequency sub-bands is completed to obtain multiple high-frequency sub-bands after noise reduction; by performing noise reduction processing on all high-frequency sub-bands, the overall quality of the image is improved.

[0132] The denoised high-frequency subband is recombined with the unprocessed low-frequency subband, and the structure and details of the original image are restored through inverse transformation. That is, the low-frequency subband and the denoised high-frequency subband are reconstructed to obtain the denoised difference image. , for High frequency subband after denoising.

[0133] In Example 5, before training the SVR model, it is necessary to extract the feature vector of the difference image. In this example, a convolutional autoencoder (CAE) can be used for feature extraction. The fused difference image is input into the convolutional autoencoder to extract the feature vector of the image. The convolutional autoencoder consists of an encoder and a decoder. Generally, the encoder is used for feature extraction.

[0134] The encoder compresses the input image into a low-dimensional latent space representation. The encoder consists of multiple convolutional layers and pooling layers.

[0135] The convolution layer extracts local features of the image through convolution operations. , For the The output of the convolutional layer, For the The convolution kernel weights of the layer, For the The bias of the layer, is the activation function, is the convolution operation;

[0136] The output of the convolutional layer is pooled to obtain the Output of the pooling layer , and obtain the feature vector of the difference image , is the number of encoder layers.

[0137] Methods for training SVR models include:

[0138] Calculate the mean and standard deviation of the feature vector, and standardize the feature vector based on the mean and standard deviation to obtain a standardized feature vector; feature vector standardization is to convert the data to zero mean and unit variance to eliminate the dimensional differences between different features, making the model training process more stable and effective. The calculation formula is: standardized feature vector = (feature vector-mean) / standard deviation.

[0139] Determine the kernel function of the SVR model, and determine the objective function, constraints, and loss function for SVR model optimization.

[0140] The kernel function is used to map the input data to a high-dimensional feature space. Commonly used kernel functions include linear kernel, radial basis function (RBF) kernel, polynomial kernel, etc.

[0141] The objective function is used to minimize the prediction error and model complexity. The objective function is , is the weight vector, is the bias, and is the slack variable, is the weight coefficient of the error term, is the kernel function mapping, is the weight coefficient in multi-task learning, is the shared weight vector in multi-task learning, For the The weight vector of each task, is the number of tasks, is the loss function, is the L1 norm, is the L2 norm, is the sample size, is the weight coefficient of L2 norm regularization, is the weight coefficient of L1 norm regularization, For the The true value of the samples, For the The input value of the samples.

[0142] Constraints are used to constrain iterations with objective functions. The constraints are ,in, To control the complexity of the model, For the slack variables, For all sample indexes All apply.

[0143] The purpose of the loss function is to find the optimal model parameters by optimizing the objective function and satisfying the constraints to minimize the prediction error and control the model complexity. The loss function is ,in, is the true value, is the model prediction value, is the smoothing parameter of the loss function.

[0144] Randomly initialize model parameters , , , and shared weight vectors in multi-task learning ; The initial value is usually generated through a random distribution to avoid the model falling into a local optimal solution at the beginning of training.

[0145] Determine the weight vector gradient and bias gradient to be optimized; determine the update direction and amplitude of the model parameters through gradient calculation to gradually minimize the objective function.

[0146] The weight vector gradient is ;

[0147] The bias gradient is ,in, is the gradient of the objective function with respect to the weight vector, is the gradient of the target vector with respect to the bias, is the gradient of the loss function with respect to the weight vector, is the gradient of the loss function with respect to the bias;

[0148] Update the model parameter expression: Through iterative optimization, the optimal model parameters that minimize the objective function are found, and finally a trained support vector regression model is obtained. The slack variables are used to deal with the errors in the model prediction and ensure that the model has good generalization ability.

[0149] The model parameters are iteratively updated until the convergence condition is reached (when the objective function value and the parameter update amplitude are less than the preset threshold), the final parameter model is output, and the trained support vector regression model is obtained based on the parameter model.

[0150] Embodiment 6: A building settlement detection terminal applicable to construction project quality, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned building settlement detection applicable to construction project quality when executing the computer program.

[0151] The memory can be used to store software programs and modules. The processor executes various functional applications and data processing of the terminal by running the software programs and modules stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, an execution program required for at least one function, etc.

[0152] The data storage area can store data created according to the use of the terminal, etc. In addition, the memory may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0153] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the building settlement detection method applicable to construction engineering quality as described above.

[0154] Without loss of generality, computer readable media may include computer storage media and communication media. Computer storage media include volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information such as computer readable instruction data structures, program modules or other data. Computer storage media include RAM, ROM, EPROM, EEPROM, flash memory or other solid-state storage technology, CD-ROM, DVD or other optical storage, cassettes, magnetic tapes, disk storage or other magnetic storage devices. Of course, those skilled in the art will appreciate that computer storage media are not limited to the above. The above-mentioned system memory and mass storage devices can be collectively referred to as memory.

[0155] A computer program product includes a computer program / instruction, which, when executed by a processor, implements the building settlement detection method applicable to construction engineering quality as described above.

[0156] A computer program product includes a computer program or set of instructions for performing specific tasks or implementing specific functions. These programs or instructions are designed to be executed by a processor to implement a series of predefined steps or operations. The program product may be stored in various forms of computer storage media, such as memory, hard disk, solid-state drive, optical disk or other forms of digital storage devices. It may exist in the form of compiled binary code or in the form of scripts or bytecodes that can be executed by an interpreter. The program product uses carefully designed algorithms and logical instructions to enable the processor to process data in a specific order and manner to complete various functions such as data analysis, user interaction, device control, etc.

[0157] In the description of this specification, the description with reference to the terms "one embodiment / method", "some embodiments / methods", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment / method or example are included in at least one embodiment / method or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment / method or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments / methods or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments / methods or examples described in this specification and the features of the different embodiments / methods or examples, unless they are contradictory.

[0158] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of this application, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0159] It should be understood by those skilled in the art that the above embodiments are only for the purpose of clearly illustrating the present invention, and are not intended to limit the scope of the present invention. For those skilled in the art, other changes or modifications may be made based on the above invention, and these changes or modifications are still within the scope of the present invention.

Claims

1. A building settlement detection method suitable for construction engineering quality, characterized in that: include: Obtain training data, which includes collected data and actual data; The collected data is processed by image processing technology to obtain the difference images of building settlement at different times, and the accurate final difference images are obtained through texture change, shape change and similarity; Decomposing the final difference image into low-frequency sub-bands and high-frequency sub-bands of multiple scales and directions, and performing adaptive threshold processing on the high-frequency sub-bands to obtain a denoised difference image; Extract the feature vector of the denoised difference image and form a training sample with the actual data; Construct the SVR model and obtain the final model parameters through iterative training of training samples; Obtain the trained support vector regression model through the final model parameters; Obtaining real-time collected data, processing the real-time collected data and the collected data, and inputting them into a support vector regression model to obtain a prediction result; The data collection equipment used include: laser radar, inertial detection unit, BDS acquisition unit and multispectral camera; the types of data collected include: three-dimensional point cloud information, attitude information, position information and multispectral image information; the actual data is the actual measured building settlement value; Methods for obtaining collected data include: Collect corresponding data through acquisition equipment, and fuse multi-source data to obtain fused data; Set quality thresholds and calculate the comprehensive quality score of the fused data; If the comprehensive quality score is lower than the quality threshold, the acquisition parameters are adjusted; if the comprehensive quality score is not lower than the quality threshold, the fluctuation intensity of the fused data is calculated; Set the fluctuation threshold. If the fluctuation intensity is greater than the fluctuation threshold, re-acquire the parameters. If the fluctuation intensity is not greater than the fluctuation threshold, output the fused data as the collected data. Among them, PCA processing is performed on the three-dimensional point cloud data to obtain the three-dimensional point cloud feature vector, the posture information is used as the posture feature vector, the position information is used as the position feature vector, and the multispectral image information is extracted through CNN to obtain the multispectral feature vector; The three-dimensional point cloud feature vector, posture feature vector, position feature vector and multi-spectral feature vector are standardized to obtain the corresponding standardized vector , , and ; Get fused data ,in, , , and is the weight of the corresponding information; Comprehensive quality score of fused data ,in, is the noise level, For data integrity, For equipment stability, is the weight corresponding to the noise level, is the weight corresponding to data integrity, is the weight corresponding to the equipment stability; Acquisition parameters include sampling rate and resolution. , then adjust the sampling rate and resolution, ,in, is the quality threshold, is the sampling rate adjustment factor, is the resolution adjustment factor, is the adjustment amplitude of the sampling rate, is the adjustment range of resolution; Fluctuation intensity ,in, is the data collected by the laser radar at adjacent collection times, is the data collected by the inertial measurement unit at adjacent collection times, is the collected data of the BDS collection unit at adjacent collection times, The data collected by the multispectral camera at adjacent collection moments; The method for obtaining the difference image includes: The fused data is converted into an image form by using image processing technology to generate an initial image; Choose any two initial images at different times; Extracting texture features of the initial image through the gray level co-occurrence matrix and determining the texture changes of two different initial images; Extracting shape features of the initial image through an active contour model and determining shape changes of two different initial images; Determine the similarity between two images by using the coherence coefficient method; The texture change, shape change and similarity are weighted and summed to obtain a difference image; Among them, texture features Contrast ,energy ,entropy and uniformity ,in, Indicates the gray value is and The co-occurrence frequency of pixel pairs at specific distances and angles; Pixels Texture changes , For time point No. Texture features at pixel points The value of For time point No. Texture features at pixel points The value of For the The weight of the texture features, They represent contrast, energy, entropy, and uniformity, respectively; The methods for obtaining shape features include: Initialize closed curve ,in represents a point on the curve; Determine the total energy function ,in, is the internal energy used to constrain the smoothness of the curve, is the external energy used to attract the curve close to the edge of the image, is the weight parameter used to control the elasticity of the curve, is the weight parameter used to control the rigidity of the curve; By iterative optimization, the total energy function is minimized so that the curve gradually approaches the edge of the image. The optimization function is: ,in, is the learning rate, For the The position of the curve at the iteration, For the The position of the curve at the iteration; After multiple iterations, the curve Converge to the edge of the image and extract the obtained shape features ; Pixels Shape changes , For time point The shape features are at the pixel point The value of For time point The shape features are at the pixel point The value of Coherence coefficient , and is the area range of the pixel point, and For time point and No. The pixel value of the pixel; Get the final difference image ,in, and is the weight coefficient for fusing texture and shape changes with the coherence coefficient.

2. A building settlement detection method suitable for construction engineering quality according to claim 1, characterized in that: The method for obtaining the denoised difference image includes: Decomposing the final difference image into low-frequency sub-bands and high-frequency sub-bands of multiple scales and directions; Divide the high frequency subband into multiple local image blocks, and calculate the variance of each local image block; A first threshold and a second threshold are set. If the variance corresponding to the local image block is less than the first threshold, the pixel points of the local image block are subjected to soft threshold processing; if the local image block is not less than the second threshold, the pixel points of the local image block are subjected to Bayesian shrinkage processing; if it is the other case, the pixel points of the local image block are subjected to hard threshold processing; The threshold processing results of different local image blocks are combined to obtain the final threshold processing result; Complete the processing of all pixel points of all high-frequency sub-bands to obtain multiple high-frequency sub-bands after noise reduction; The low-frequency subband and the denoised high-frequency subband are reconstructed to obtain a denoised difference image.

3. A building settlement detection method suitable for construction engineering quality according to claim 2, characterized in that: Decompose the final difference image, ,in, is the low frequency subband, For the High frequency sub-bands; Set the first threshold and the second threshold , , and make a judgment on each local image block; like , then use soft threshold processing to obtain the soft threshold processing result ,in, is the soft threshold parameter, For the In the local image block pixels, is the variance corresponding to the local image block; like , then use hard threshold processing to obtain the hard threshold processing result , is the hard threshold parameter; like , then use Bayesian Shrinkage processing to obtain the Bayesian processing result , is the noise variance; Reconstruct the denoised difference image , for High frequency subband after denoising.

4. A building settlement detection method applicable to construction engineering quality according to claim 1, characterized in that: Methods for training SVR models include: Calculate the mean and standard deviation of the feature vector, and standardize the feature vector based on the mean and standard deviation to obtain a standardized feature vector; Determine the kernel function of the SVR model, and determine the objective function, constraints and loss function of the SVR model optimization; Randomly initialize model parameters; Determine the weight vector gradient and bias gradient for optimization; Iteratively update the model parameters until the convergence condition is reached, output the final parameter model, and obtain the trained support vector regression model based on the parameter model; The objective function is , is the weight vector, is the bias, and is the slack variable, is the weight coefficient of the error term, is the kernel function mapping, is the weight coefficient in multi-task learning, is the shared weight vector in multi-task learning, For the The weight vector of each task, is the number of tasks, is the loss function, is the L1 norm, is the L2 norm, is the sample size, is the weight coefficient of L2 norm regularization, is the weight coefficient of L1 norm regularization, For the The true value of the samples, For the The input value of samples; The constraints are ,in, To control the complexity of the model, For the slack variables, For all sample indexes All apply; The loss function is ,in, is the true value, is the model prediction value, is the smoothing parameter of the loss function; Randomly initialize model parameters , , , and shared weight vectors in multi-task learning ; The weight vector gradient is ; The bias gradient is ,in, is the gradient of the objective function with respect to the weight vector, is the gradient of the target vector with respect to the bias, is the gradient of the loss function with respect to the weight vector, is the gradient of the loss function with respect to the bias; Update the model parameter expression: .

5. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the building settlement detection method applicable to construction project quality as described in any one of claims 1 to 4 is implemented.

6. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instruction is executed by the processor, the building settlement detection method applicable to the quality of construction projects as described in any one of claims 1 to 4 is implemented.