Floor rubber pad quality detection method based on image processing

Abstract

The application discloses a floor rubber pad quality detection method based on image processing and concretely relates to the technical field of quality detection, which comprises three-dimensional form reconstruction, dense point cloud acquisition of a rubber pad surface through binocular vision and a grid structure light, local reference plane fitting and fusion into a global reference curved surface by using a block weighted least square method, vertical deviation calculation to generate a flatness heat map, and identification of abnormal areas such as warping, hollowing and depression; texture abnormality identification, high-resolution texture image acquisition in a flatness qualified area, multi-scale gray level co-occurrence matrix feature extraction, vibration reduction coefficient prediction by using a support vector machine regression model, and determination of texture abnormality areas caused by stretching, extrusion or aging; and elastic recovery detection, low-frequency vibration excitation to texture abnormality and flatness critical area and dynamic structured light image sequence collection. The application realizes global, non-contact and quantitative detection of floor rubber pad laying quality.

Description

Technical Field

[0001] This invention relates to the field of quality inspection technology, and more specifically, to a method for quality inspection of floor slab rubber pads based on image processing. Background Technology

[0002] Vibration-damping and noise-reducing rubber pads laid in floor slab structures are key materials for improving building sound insulation performance. In actual construction, rubber pads are usually laid on the floor surface after the floor slab is poured, and then covered with other surface layers as needed. The quality of rubber pad installation directly affects the overall vibration reduction effect and service durability. However, in current engineering practice, the quality inspection of laid rubber pads mainly relies on manual visual inspection or simple tool sampling, making it difficult to achieve a comprehensive and quantitative assessment. Especially for large-area floor slabs, manual inspection is inefficient, highly subjective, and cannot accurately identify flatness defects such as local warping, hollow areas, and depressions, nor can it assess the surface texture abnormalities caused by installation tension, local compression, or material aging and their potential impact on vibration reduction performance.

[0003] While existing technologies offer testing methods for the mechanical properties of rubber materials, these methods are mostly applicable to raw material samples that have not yet been laid and cannot be directly applied to rubber mats already installed on floor slabs. Some studies have attempted to use image processing techniques for visual inspection of the rubber mat surface, but these typically only involve simple classification of surface texture and lack a joint evaluation of multi-dimensional quality indicators such as three-dimensional morphology and elastic recovery capability. Furthermore, existing testing methods cannot identify internal structural damage to the rubber mat (such as adhesive layer detachment and material delamination) and its dynamic response characteristics under non-contact conditions, making it difficult to meet the demands of modern buildings for refined acceptance of floor slab sound insulation quality.

[0004] Therefore, there is an urgent need for a non-destructive testing method that can comprehensively evaluate the flatness, texture abnormalities, and elastic recovery ability of laid rubber mats. By combining image processing and dynamic excitation, a large-area, automated, and quantitative quality inspection method can be achieved, providing a reliable basis for the construction acceptance and maintenance of floor slab rubber mats. Summary of the Invention

[0005] To overcome the above-mentioned deficiencies of the prior art, embodiments of the present invention provide a method for quality detection of floor slab rubber pads based on image processing.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] The image processing-based method for quality inspection of floor slab rubber pads includes the following steps:

[0008] Step 1: 3D morphology reconstruction. Dense point cloud of rubber pad surface is obtained through binocular vision and mesh structured light. Local reference plane is fitted by block weighted least squares method and fused into global reference surface. Vertical deviation is calculated to generate flatness heat map and identify flatness abnormal areas such as warping, hollowing, and depression.

[0009] Step 2: Texture anomaly identification. High-resolution texture images are acquired within the flatness-compliant area. Multi-scale gray-level co-occurrence matrix features are extracted. The vibration reduction coefficient is predicted using a support vector machine regression model. After comparison with a threshold, the texture anomaly area caused by stretching, compression, or aging is determined.

[0010] Step 3: Elastic recovery detection. Low-frequency vibration excitation is applied to areas with abnormal texture and critical flatness areas. Dynamic structured light image sequences are acquired, and the recovery curve is fitted using a double exponential decay model. The recovery coefficient and slow recovery time constant are calculated to identify internal structural damage or adhesive layer failure.

[0011] Step 4: Visual integration. The flatness deviation, texture anomaly degree, and dynamic response results are weighted and integrated to calculate the comprehensive quality score. The defect type, location, and severity are marked on the 3D model to generate an evaluation report.

[0012] Specifically, step one includes:

[0013] A binocular vision acquisition system is set up, which includes two high-resolution industrial cameras, a structured light projector, and a set of movable slide rails. The slide rails are controlled to move the acquisition device along a set path, while the projector projects high-density grid structured light onto the surface of the rubber pad. The binocular cameras simultaneously acquire grid deformation images from the left and right perspectives. The three-dimensional spatial coordinates of each grid node are calculated using the principle of triangulation to generate a dense point cloud model.

[0014] The point cloud data is divided into several continuous strips based on the sliding track trajectory. Each strip is further divided into multiple local sub-blocks with an overlap rate of 10% according to a preset length. For each local sub-block, a weighted least squares method is used to fit a local reference plane, where the weight coefficients are set according to the distance from the point cloud to the center of the sub-block, and a Gaussian kernel function is used. The parameters of the local reference plane of each sub-block are obtained by solving the weighted least squares problem. The overlapping areas of adjacent sub-blocks are fused using the inverse distance weighted interpolation method to obtain a continuous theoretical reference surface for the entire region.

[0015] Calculate the vertical deviation of each point cloud relative to the theoretical reference surface, generate a flatness distribution heatmap, and mark the area as a flatness abnormal area when the absolute value of the vertical deviation exceeds the flatness qualification threshold and the area of ​​the continuous region is greater than the preset area threshold.

[0016] Specifically, step two includes:

[0017] High-resolution texture images were acquired within areas with acceptable flatness. A multi-scale gray-level co-occurrence matrix analysis strategy was adopted to extract contrast, correlation, energy, and homogeneity feature parameters at multiple different pixel spacings, and these parameters were then spliced ​​together to form a multi-scale feature vector.

[0018] A sample library of vibration reduction performance is established in advance: rubber pad samples from the same batch are taken, and tensile, extrusion, and aging defect states are simulated. The vibration reduction coefficient of each sample is measured. At the same time, texture images are collected to extract gray-level co-occurrence matrix features. A support vector machine regression model is used to establish the mapping relationship between feature vectors and vibration reduction coefficients.

[0019] During on-site testing, multi-scale feature vectors are extracted from the area to be tested and input into a regression model to obtain the predicted vibration reduction coefficient. When the predicted vibration reduction coefficient is lower than the design requirement threshold, the area is identified as an area with abnormal texture and affecting vibration reduction performance.

[0020] Specifically, in step two:

[0021] After constructing the multi-scale eigenvectors, principal component analysis is used for dimensionality reduction: the multi-scale eigenma matrix is ​​centered, the covariance matrix is ​​calculated, the eigenvalues ​​and eigenvectors are solved, and the top principal components with a cumulative contribution rate of over 95% are selected as the final eigenvectors. The number of principal components is 4 to 6.

[0022] Specifically, step three includes:

[0023] For areas with abnormal texture and critical flatness, dynamic structured light excitation detection is initiated; a projector projects time-series encoded structured light, while a micro-excitation device applies standardized low-frequency vibration excitation; a high-speed camera continuously acquires surface structured light image sequences during and after the excitation process.

[0024] By calculating the phase change of each frame relative to the initial unexcited state, the surface deformation change curve over time is obtained; for the recovery stage after the excitation stops, the recovery curve is extracted and fitted using a double exponential decay model to obtain the fast recovery time constant and the slow recovery time constant, and the double exponential recovery coefficient is defined.

[0025] When the slow recovery time constant exceeds the preset threshold or the double exponential recovery coefficient is lower than the preset threshold, it is determined that there is internal structural damage or adhesive layer failure in the area.

[0026] Specifically, in step three:

[0027] The Levenberg-Marquardt algorithm was used to optimize the parameters of the double exponential decay model. The initial values ​​were set as follows: fast recovery time constant 0.2 seconds, slow recovery time constant 1.0 seconds, initial amplitude of fast recovery component 0.7 times the maximum deformation, initial amplitude of slow recovery component 0.3 times the maximum deformation, and iterative convergence condition was that the change of residual sum of squares was less than 10 to the power of -6.

[0028] After the fitting is completed, the proportion of the fast recovery component is calculated. When the proportion of the fast recovery component is less than 0.5, it is determined that there is internal material aging or delamination defect.

[0029] Simultaneously, the sum of squared residuals between the recovery curve and the double exponential model is calculated. When the sum of squared residuals exceeds a preset threshold, it is determined that the recovery process deviates from the ideal viscoelastic model and is accompanied by nonlinear damage behavior.

[0030] Specifically, step four includes:

[0031] A weighted scoring method was used to calculate the flatness deviation score, texture anomaly score, and dynamic response score for each detection unit, and each score was normalized to the range of 0 to 1. The comprehensive quality score was obtained by multiplying the flatness deviation score, texture anomaly score, and dynamic response score by their respective weighting coefficients and then adding them together. The weight of the flatness deviation score was 0.4, the weight of the texture anomaly score was 0.3, and the weight of the dynamic response score was 0.3.

[0032] When the overall quality score is below 0.6, the detection unit is marked as a defect unit, and the defect type is determined according to the distribution of each individual score: the lowest flatness deviation score is classified as a flatness defect, the lowest texture anomaly score is classified as a texture defect, and the lowest dynamic response score is classified as a structural damage defect. Finally, the location, range, and severity of each type of defect are marked with different colors in the 3D reconstruction model, and an overall quality assessment report containing a defect statistics table, a location distribution map, and processing suggestions is generated.

[0033] The technical effects and advantages of this invention are as follows:

[0034] This invention combines binocular vision with structured light to achieve full-area quantitative detection of the flatness of floor slabs after rubber pad installation. It employs a block-weighted least squares method to fit local reference planes and fuse them into a global reference surface, effectively overcoming the plane fitting distortion problem caused by large-area overall floor slab warping. This allows for accurate identification of local abnormal areas such as warping, hollow areas, and depressions, along with their area and degree of deviation. Simultaneously, this invention introduces a multi-scale gray-level co-occurrence matrix to extract texture features and combines it with a support vector machine regression model to establish a mapping relationship between texture and vibration damping performance. This enables automatic identification of texture anomalies caused by different factors such as tension, compression, and aging, and quantitative assessment of vibration damping performance loss, compensating for the limitations of manual visual inspection in perceiving functional performance.

[0035] This invention employs dynamic structured light excitation and a double-exponential decay model to fit the recovery curve, enabling the differentiation between the elastic recovery of the rubber pad surface and the viscoelastic recovery of the internal structure. By using the slow recovery time constant and the double-exponential recovery coefficient, it accurately identifies hidden defects such as internal structural damage, adhesive layer failure, and material aging delamination. Furthermore, it analyzes the sum of squared residuals to determine whether the recovery process deviates from the ideal viscoelastic model. Finally, a weighted scoring method is used to integrate the results of flatness, texture, and dynamic response, generating a visualized comprehensive quality assessment report. This provides a quantitative and reliable basis for the construction acceptance, defect location, and repair decisions of floor slab rubber pads. Attached Figure Description

[0036] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0037] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0038] like Figure 1 As shown, the steps of the image processing-based method for inspecting the quality of floor slab rubber pads are as follows:

[0039] Step 1: 3D morphology reconstruction. Dense point clouds of the rubber pad surface are acquired using binocular vision and structured light mesh. A block-weighted least squares method is used to fit the local reference plane and fuse it into a global reference surface. Vertical deviation is calculated to generate a flatness heatmap, identifying abnormal areas such as warping, hollow areas, and depressions. The specific process is as follows:

[0040] After the rubber pads on the floor are laid, a binocular vision acquisition system is first installed above the area to be inspected, including two high-resolution industrial cameras, a structured light projector and a set of movable slide rails.

[0041] The control slide rail drives the acquisition device to move along the set path. At the same time, a high-density grid structured light is projected onto the surface of the rubber pad through the projector. The binocular camera simultaneously acquires grid deformation images from the left and right perspectives. The three-dimensional spatial coordinates of each grid node are calculated using the principle of triangulation to generate a dense point cloud model of the rubber pad surface, and then the theoretical reference plane is fitted.

[0042] After generating a dense point cloud model, the point cloud data is divided into several continuous strips according to the motion trajectory of the movable slide rail. Each strip corresponds to a complete image sequence acquired during one slide rail movement. Furthermore, each strip is divided into N local sub-blocks with an overlap rate of 10% according to a preset length L. The side length L of each sub-block is 500mm-1000mm. The overlap design ensures a smooth transition between adjacent sub-blocks.

[0043] For each local sub-block, a weighted least squares method is used to fit the local reference plane to overcome the fitting bias caused by overall warping when fitting a large area. Let the point cloud coordinates in the (m,n)th sub-block be... ,in For striped indexes, For sub-block index, Given the point cloud index within the sub-block, the fitted plane equation is: ;in, This represents the height of the local reference plane at coordinates (x, y). Let be the slope of the plane fitted to the (m,n)th sub-block in the x-direction. Let be the slope of the y-direction of the fitted plane of the (m,n)th sub-block. Let x and y be the intercept of the plane fitting the (m,n)th sub-block, and let x and y be the two-dimensional coordinates of the point cloud on the horizontal plane.

[0044] Weighting coefficient Based on the distance from the point cloud to the center of the sub-block, a Gaussian kernel function is used. , Let be the weight coefficients for the i-th strip, the j-th sub-block, and the k-th point cloud. For point ( Euclidean distance to the center of the sub-block The Gaussian kernel bandwidth parameter is set to 1 / 4 of the sub-block side length L to enhance the contribution of the central region to the plane fitting and suppress edge noise.

[0045] Solve the weighted least squares problem to minimize the objective function: ;in, To minimize the objective function, For the plane parameters to be optimized, This represents the total number of point clouds within the sub-block. These are the actual height coordinates of the point cloud. , The horizontal coordinates of the point cloud;

[0046] After obtaining the local reference plane parameters of each sub-block, the overlapping areas of adjacent sub-blocks are fused using the inverse distance weighted interpolation method to obtain a continuous theoretical reference surface for the entire region. , This is the fused global theoretical reference surface function. For any point cloud coordinates ( Its vertical deviation value is corrected as follows: ;in, Let be the vertical deviation value of the i-th point cloud. Let i be the actual height of the i-th point cloud. For in ( Process the height value of the theoretical reference surface; when the point is located in the overlapping area of ​​sub-blocks. The weighted average of the fitted values ​​at that point on the reference plane of adjacent sub-blocks is taken, and the weight is inversely proportional to the distance from the point to the center of each sub-block.

[0047] Calculate the vertical deviation of each point cloud relative to the reference plane, generate a flatness distribution heatmap, identify local abnormal areas such as warping, hollowing, and depression, and calculate their area and maximum deviation value.

[0048] Specifically, before use, the binocular vision acquisition system requires binocular calibration. The Zhang Zhengyou calibration method is used to obtain the intrinsic parameter matrices, distortion coefficients, and relative pose relationships of the left and right cameras to ensure the accuracy of 3D reconstruction. The grid projected by the structured light projector is a binary checkerboard pattern with unique coding features. Each grid node contains unique spatial coding information, facilitating accurate matching of corresponding points in the left and right images. After generating a dense point cloud model, the least squares method is used to fit the theoretical reference plane. That is, after iteratively removing outliers, the Z-coordinate values ​​of all points in the point cloud are used for plane fitting to obtain the equation of the reference plane. ;in The height of the overall reference plane, , Let be the slope of the plane in the x and y directions. This is the plane intercept. Calculate the vertical deviation of each point cloud relative to the reference plane. Set the flatness qualification threshold as ,when And the area of ​​the continuous region is greater than the preset area threshold. When this occurs, the area is marked as an area of ​​abnormal flatness, where The value range is 2mm-5mm. The value ranges from 100cm² to 400cm², and the specific value is determined according to the project acceptance standards.

[0049] Step 2: Texture Anomaly Identification. High-resolution texture images are acquired within areas of acceptable flatness. Multi-scale gray-level co-occurrence matrix features are extracted, and a support vector machine regression model is used to predict the vibration reduction coefficient. This coefficient is then compared with a threshold to determine areas of texture anomalies caused by stretching, compression, or aging. The specific process is as follows:

[0050] Within the area where the flatness meets the requirements, high-resolution texture images of the rubber pad surface are further acquired;

[0051] Texture feature parameters, including contrast, correlation, energy, and homogeneity, are extracted using gray-level co-occurrence matrix (GLCM). A multi-scale GLCM analysis strategy is employed to extract texture feature parameters at multiple different pixel spacings. Let the set of pixel spacings be denoted as . , Let M be the set of pixel spacing, and M be the total number of scales. (s=1,…,M) represents the pixel spacing (in pixels) at the s-th scale, with values ​​of: , To capture multi-layered texture structures from the microscopic to the mesoscopic. For each scale Calculate the gray-level co-occurrence matrix in each of the four directions. The average value of the eigenvalues ​​in each direction is taken as the feature parameter at that scale, thus obtaining the scale. Contrast at lower levels Correlation ,energy Homogeneity The feature parameters at each scale are concatenated to form a multi-scale feature vector: ; It is a multi-scale texture feature vector with a dimension of 12, which can simultaneously reflect fine-grained (r=1) micro-texture changes and coarse-grained (r=3) structural texture anomalies, effectively improving the ability to distinguish different defect types.

[0052] The vector is compared with a pre-established vibration reduction performance sample library, which is established by testing the vibration reduction coefficient of rubber pad samples with different texture states.

[0053] By identifying texture abnormalities caused by laying stretching, local compression, or material aging through feature matching, the potential impact on vibration reduction performance can be quantified.

[0054] Specifically, the method for constructing the vibration reduction performance sample library is as follows: Rubber pad samples of the same batch and specifications as the rubber pads laid on-site are taken, and typical defect states such as tension, compression, and aging are simulated. A universal testing machine is used in conjunction with a vibration reduction coefficient test bench to determine the vibration reduction coefficient of each sample under fixed load and frequency. , defined as the ratio of transmission loss to input vibrational energy. Simultaneously, high-resolution texture images of each sample surface are acquired, and four feature parameters of the gray-level co-occurrence matrix are extracted: contrast... Correlation ,energy Homogeneity Construct feature vectors ,in This represents a single-scale texture feature vector. A support vector machine regression model is used to establish the mapping relationship between the feature vector and the vibration reduction coefficient. ,in The vibration reduction coefficient is... This is a support vector machine regression model function. During on-site detection, feature vectors are extracted from the region to be tested. The predicted vibration reduction coefficient is obtained by inputting it into the regression model. ,when Vibration reduction coefficient threshold below design requirements At that time, the area was identified as having abnormal texture and affecting vibration reduction performance. Determined by architectural design requirements, and not less than 0.85.

[0055] Preferably, to reduce feature dimensionality and redundant information, principal component analysis is further used for dimensionality reduction after constructing the multi-scale feature vectors. This involves processing the multi-scale feature matrices of the collected N samples. Perform centralization and calculate the covariance matrix. ,in Let covariance matrix be the variance matrix. For the sample size, It is a multi-scale feature matrix. This is the matrix transpose.

[0056] Solve for the eigenvalues ​​and eigenvectors, and select the top p principal components with a cumulative contribution rate exceeding 95% as the final eigenvectors. ,in The feature vectors after dimensionality reduction. The number of principal components should be between 4 and 6, which improves the computational efficiency of subsequent regression models while maintaining information integrity.

[0057] Step 3: Elastic recovery detection. Low-frequency vibration excitation is applied to areas with texture anomalies and critical smoothness, and dynamic structured light image sequences are acquired. The recovery curve is fitted using a double exponential decay model, and the recovery coefficient and slow recovery time constant are calculated to identify internal structural damage or adhesive layer failure. The specific process is as follows:

[0058] Dynamic structured light stimulation detection is initiated for areas with abnormal textures and areas with critical flatness.

[0059] Time-series encoded structured light is projected onto the surface of the rubber pad using a projector, and standardized low-frequency vibration excitation is applied in conjunction with a micro-excitation device.

[0060] A high-speed camera was used to continuously acquire surface structured light image sequences of a rubber pad during and after excitation.

[0061] By analyzing the phase change and recovery time of structured light fringes, the local elastic recovery coefficient and residual deformation of the rubber pad are calculated, and abnormal response areas caused by internal structural damage, adhesive layer detachment, or material failure are identified.

[0062] Specifically, in the dynamic structured light excitation detection, the time-series encoded structured light projected by the projector adopts a three-step phase-shifting method, that is, the phase difference is sequentially projected. The surface phase distribution is obtained by solving the three sinusoidal fringe patterns. The micro-excitation device uses an electromagnetic exciter, applying a standardized low-frequency vibration excitation frequency of 10Hz, an amplitude of 0.5mm, and a duration of 2s, with the excitation direction perpendicular to the rubber pad surface. A high-speed camera continuously acquires structured light image sequences during the excitation process and within 5s after the excitation stops at a frame rate of 200fps. The phase change of each frame relative to the initial unexcited state is calculated. The surface deformation over time curve is obtained. For the recovery phase after excitation stops, the recovery curve of each pixel is extracted, and an exponential decay model is used. Perform fitting, where The surface deformation varies with time t. This is the initial shape variable (maximum shape variable). Define the elastic recovery coefficient as the recovery time constant. , For the largest deformation, The residual deformation is 3 seconds after the excitation stops.

[0063] Preferably, to distinguish the contributions of the elastic recovery of the rubber pad surface from the viscoelastic recovery of the internal structure, a double exponential decay model is used to fit the recovery curve. The model expression is as follows: ;in, and The initial amplitudes of the fast recovery component and the slow recovery component are respectively, satisfying... ; The fast recovery time constant reflects the instantaneous elastic recovery characteristics of the rubber pad surface material; The slow recovery time constant reflects the viscoelastic recovery characteristics of the internal structure of the rubber pad.

[0064] The parameters were obtained by fitting using the nonlinear least squares method. Define the double exponential recovery coefficient: ;in It is the double exponential recovery coefficient. To assess the time point, This is the evaluation time point after the stimulus stops. When the slow recovery time constant... Exceeding the preset threshold or double exponential recovery coefficient Below the preset threshold At that time, it was determined that there was internal structural damage or adhesive layer failure in the area, among which The value is 2.0s. The value is 0.85.

[0065] Furthermore, during the fitting of the double exponential model, the Levenberg-Marquardt algorithm was used for parameter optimization, with the initial values ​​set as follows: The iterative convergence condition is that the change in the sum of squared residuals is less than 1%. After fitting is complete, the proportion of the fast recovery component is calculated. ,when A value <0.5 indicates that the rubber pad primarily relies on slow viscoelastic recovery, potentially indicating internal material aging or delamination defects. Simultaneously, the sum of squared residuals between the recovery curve and the double exponential model is calculated. ,in This represents the number of sampling points in the time series. For the actual measured deformation, The deformation variables fitted to the double exponential model; when When the threshold is exceeded, it indicates that the recovery process deviates from the ideal viscoelastic model and may be accompanied by nonlinear damage behavior.

[0066] Step 4: Visual Integration. The flatness deviation, texture anomaly degree, and dynamic response results are weighted and integrated to calculate a comprehensive quality score. Defect types, locations, and severity are then marked with different colors on the 3D model, generating an evaluation report containing remedial recommendations. The specific process is as follows:

[0067] The results of three-dimensional morphological deviation, texture anomaly degree and dynamic response anomaly are integrated to generate a comprehensive quality assessment report that includes defect type, location, area, severity and vibration reduction performance impact; and the report is visualized and annotated in the three-dimensional model to provide a quantitative basis for subsequent repair or replacement.

[0068] Specifically, the fusion of the test results adopts a weighted scoring method, calculating a flatness deviation score for each test unit (the smallest evaluation unit after grid division, with a side length of 10cm × 10cm). Texture anomaly score Dynamic response score Each score is normalized to the [0,1] interval, with lower scores indicating lower quality. Overall quality score. The weighting coefficient Use values ​​of 0.4, 0.3, and 0.3 respectively, adjusting according to the project's priorities. When this occurs, the unit is marked as a defective unit, and the defect type is determined based on the score distribution: if If the lowest, it is classified as a flatness defect; if If the lowest, it is classified as a texture defect; if The lowest severity is classified as structural damage. Finally, the location, extent, and severity of each type of defect are marked with different colors in the 3D reconstruction model, and a comprehensive quality assessment report is generated, including a defect statistics table, a location distribution map, and treatment recommendations.

[0069] The above formulas are all dimensionless calculations. Dimensionless calculations can be performed using various methods such as standardization, which will not be elaborated here. The formulas are derived from software simulations based on a large amount of collected data, and the preset parameters in the formulas can be set by those skilled in the art according to the actual situation.

[0070] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, ATA hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. The semiconductor medium can be a solid-state ATA hard disk.

[0071] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0072] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0073] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0074] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0075] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0076] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable ATA hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0077] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for quality inspection of floor slab rubber pads based on image processing, characterized in that, Includes the following steps: Step 1: 3D morphology reconstruction. Dense point cloud of rubber pad surface is obtained through binocular vision and mesh structured light. Local reference plane is fitted by block weighted least squares method and fused into global reference surface. Vertical deviation is calculated to generate flatness heat map and identify flatness abnormal areas such as warping, hollowing, and depression. Step 2: Texture anomaly identification. High-resolution texture images are acquired within the flatness-compliant area. Multi-scale gray-level co-occurrence matrix features are extracted. The vibration reduction coefficient is predicted using a support vector machine regression model. After comparison with a threshold, the texture anomaly area caused by stretching, compression, or aging is determined. Step 3: Elastic recovery detection. Low-frequency vibration excitation is applied to areas with abnormal texture and critical flatness areas. Dynamic structured light image sequences are acquired, and the recovery curve is fitted using a double exponential decay model. The recovery coefficient and slow recovery time constant are calculated to identify internal structural damage or adhesive layer failure. Step 4: Visual integration. The flatness deviation, texture anomaly degree, and dynamic response results are weighted and integrated to calculate the comprehensive quality score. The defect type, location, and severity are marked on the 3D model to generate an evaluation report.

2. The method for quality inspection of floor slab rubber pads based on image processing according to claim 1, characterized in that, Step one specifically includes: A binocular vision acquisition system is set up, which includes two high-resolution industrial cameras, a structured light projector, and a set of movable slide rails. The slide rails are controlled to move the acquisition device along a set path, while the projector projects high-density grid structured light onto the surface of the rubber pad. The binocular cameras simultaneously acquire grid deformation images from the left and right perspectives. The three-dimensional spatial coordinates of each grid node are calculated using the principle of triangulation to generate a dense point cloud model. The point cloud data is divided into several continuous strips based on the sliding track trajectory. Each strip is further divided into multiple local sub-blocks with an overlap rate of 10% according to a preset length. For each local sub-block, a weighted least squares method is used to fit a local reference plane, where the weight coefficients are set according to the distance from the point cloud to the center of the sub-block, and a Gaussian kernel function is used. The parameters of the local reference plane of each sub-block are obtained by solving the weighted least squares problem. The overlapping areas of adjacent sub-blocks are fused using the inverse distance weighted interpolation method to obtain a continuous theoretical reference surface for the entire region. Calculate the vertical deviation of each point cloud relative to the theoretical reference surface, generate a flatness distribution heatmap, and mark the area as a flatness abnormal area when the absolute value of the vertical deviation exceeds the flatness qualified threshold and the area of ​​the continuous region is greater than the preset area threshold.

3. The method for quality inspection of floor slab rubber pads based on image processing according to claim 1, characterized in that, Step two specifically includes: High-resolution texture images were acquired within areas with acceptable flatness. A multi-scale gray-level co-occurrence matrix analysis strategy was adopted to extract contrast, correlation, energy, and homogeneity feature parameters at multiple different pixel spacings, and these parameters were then spliced ​​together to form a multi-scale feature vector. A sample library of vibration reduction performance is established in advance: rubber pad samples from the same batch are taken, and tensile, extrusion, and aging defect states are simulated. The vibration reduction coefficient of each sample is measured. At the same time, texture images are collected to extract gray-level co-occurrence matrix features. A support vector machine regression model is used to establish the mapping relationship between feature vectors and vibration reduction coefficients. During on-site testing, multi-scale feature vectors are extracted from the area to be tested and input into a regression model to obtain the predicted vibration reduction coefficient. When the predicted vibration reduction coefficient is lower than the design requirement threshold, the area is identified as an area with abnormal texture and affecting vibration reduction performance.

4. The method for quality inspection of floor slab rubber pads based on image processing according to claim 3, characterized in that, In step two: After constructing the multi-scale eigenvectors, principal component analysis is used for dimensionality reduction: the multi-scale eigenma matrix is ​​centered, the covariance matrix is ​​calculated, the eigenvalues ​​and eigenvectors are solved, and the top principal components with a cumulative contribution rate of over 95% are selected as the final eigenvectors. The number of principal components is 4 to 6.

5. The method for quality inspection of floor slab rubber pads based on image processing according to claim 1, characterized in that, Step three specifically includes: For areas with abnormal texture and critical flatness, dynamic structured light excitation detection is initiated; a projector projects time-series encoded structured light, while a micro-excitation device applies standardized low-frequency vibration excitation; a high-speed camera continuously acquires surface structured light image sequences during and after the excitation process. By calculating the phase change of each frame relative to the initial unexcited state, the surface deformation change curve over time is obtained; for the recovery stage after the excitation stops, the recovery curve is extracted and fitted using a double exponential decay model to obtain the fast recovery time constant and the slow recovery time constant, and the double exponential recovery coefficient is defined. When the slow recovery time constant exceeds the preset threshold or the double exponential recovery coefficient is lower than the preset threshold, it is determined that there is internal structural damage or adhesive layer failure in the area.

6. The method for quality inspection of floor slab rubber pads based on image processing according to claim 5, characterized in that, In step three: The Levenberg-Marquardt algorithm was used to optimize the parameters of the double exponential decay model. The initial values ​​were set as follows: fast recovery time constant 0.2 seconds, slow recovery time constant 1.0 seconds, initial amplitude of fast recovery component 0.7 times the maximum deformation, initial amplitude of slow recovery component 0.3 times the maximum deformation, and iterative convergence condition was that the change of residual sum of squares was less than 10 to the power of -6. After the fitting is completed, the proportion of the fast recovery component is calculated. When the proportion of the fast recovery component is less than 0.5, it is determined that there is internal material aging or delamination defect. Simultaneously, the sum of squared residuals between the recovery curve and the double exponential model is calculated. When the sum of squared residuals exceeds a preset threshold, it is determined that the recovery process deviates from the ideal viscoelastic model and is accompanied by nonlinear damage behavior.

7. The method for quality inspection of floor slab rubber pads based on image processing according to claim 1, characterized in that, Step four specifically includes: A weighted scoring method was used to calculate the flatness deviation score, texture anomaly score, and dynamic response score for each detection unit, and each score was normalized to the range of 0 to 1. The comprehensive quality score was obtained by multiplying the flatness deviation score, texture anomaly score, and dynamic response score by their respective weighting coefficients and then adding them together. The weight of the flatness deviation score was 0.4, the weight of the texture anomaly score was 0.3, and the weight of the dynamic response score was 0.

3. When the overall quality score is below 0.6, the detection unit is marked as a defect unit, and the defect type is determined according to the distribution of each individual score: the lowest flatness deviation score is classified as a flatness defect, the lowest texture anomaly score is classified as a texture defect, and the lowest dynamic response score is classified as a structural damage defect. Finally, the location, range, and severity of each type of defect are marked with different colors in the 3D reconstruction model, and an overall quality assessment report containing a defect statistics table, a location distribution map, and processing suggestions is generated.

Images