A method for measuring weld parameters immune to interference

Abstract

The present disclosure provides a kind of weld parameter anti-interference measurement method, the spectral characteristics of high reflectivity surface are obtained by laser scattering technology and Fourier transform processing and matching correction coefficient, realize the weighted adjustment of scattering signal and geometric parameter extraction, accurately quantize groove size by combining edge enhancement algorithm and three-dimensional profile modeling technology, and effectively overcome the interference such as surface oil stain, rust and stripe fracture by using image filtering, morphological reconstruction and outlier rejection algorithm, finally through multi-source data fusion and adaptive calculation range adjustment, high-precision, anti-interference measurement to weld groove angle, blunt edge thickness and group pair error is realized under high reflectivity, rust and complex working conditions, significantly improve the reliability and automation level of welding quality detection.

Description

Technical Field

[0001] This invention relates to the field of pipeline welding technology, and in particular to a method for measuring weld parameters against interference. Background Technology

[0002] In actual production, when collecting three-dimensional topographic data of the workpiece bevel using a 3D scanner, the process of obtaining the bevel angle and blunt edge thickness faces a complex technical challenge.

[0003] Specifically, when the workpiece is made of highly reflective metal and has fine oil stains on its surface, the light reflection signal of the 3D scanner will become unstable due to the material characteristics and oil stain interference, resulting in a deviation of ±2° in the bevel angle measurement. At the same time, the blunt edge thickness data will have an error amplified to more than 0.5 mm due to the blurred boundary.

[0004] Next, in the pipe assembly process, when using structured light scanning technology to measure the misalignment data and the bevel dimensions after assembly, if there is local corrosion or slight deformation on the pipe surface, the fringe projection of the structured light will break due to the change in surface texture, causing the measured misalignment value to deviate from the actual value by up to 1 mm. This technical contradiction is that the high-precision scanning requirement and the surface condition under complex working conditions are mutually restrictive, making it difficult to maintain the continuity and reliability of the entire data chain from bevel preparation, assembly to post-weld inspection in special scenarios. Summary of the Invention

[0005] A first aspect of this disclosure provides a method for measuring weld parameters against interference, comprising:

[0006] The scattered light distribution data of a highly reflective surface is obtained by laser scattering technology, and the scattered light distribution data is subjected to Fourier transform processing to obtain spectral characteristics;

[0007] Based on the spectral characteristics, correction coefficients are matched in a pre-established model, and the scattered light distribution data are weighted and adjusted using the correction coefficients to generate a corrected signal.

[0008] The initial measured values ​​of the geometric parameters are extracted from the corrected signal, and the initial measured values ​​are adjusted through smoothing to obtain the geometric data;

[0009] The boundary feature signal is separated from the corrected signal, and the boundary feature signal is processed by an edge enhancement algorithm to generate boundary enhancement data;

[0010] A three-dimensional contour model is constructed using the boundary enhancement data and the geometric data. The three-dimensional contour model is then quantized using segmentation techniques to obtain quantized data.

[0011] The original image of the projection data is acquired, and the original image is denoised by filtering to generate a filtered image;

[0012] Continuous features are extracted from the filtered image, and the continuous features are processed to generate repaired data;

[0013] The deviation is calculated based on the repaired data and the principle of triangulation. The deviation is then processed using an outlier removal algorithm to obtain the deviation data.

[0014] Depth map data is acquired, processed by classification and segmentation, and the calculation range is adjusted by combining the deviation data and the quantization data to generate quantization data.

[0015] In conjunction with the first aspect, the method of acquiring scattered light distribution data from a highly reflective surface using laser scattering technology, and performing Fourier transform processing on the scattered light distribution data to obtain spectral characteristics, includes:

[0016] Initial light distribution data of highly reflective surfaces are collected using laser scattering technology, and the distribution pattern of scattered light at different angles is recorded to obtain an initial light distribution dataset.

[0017] Scattered light intensity information is extracted from the initial light distribution dataset, and the light intensity value at each angle is normalized to obtain scattered light intensity sequence data.

[0018] The scattered light intensity sequence data is processed using a fast Fourier transform algorithm to generate a spectral data set;

[0019] Spectral peaks and frequency distribution features are extracted from the spectral data set, and significant feature points are determined based on a preset threshold to obtain a spectral feature set;

[0020] The spectral feature set is reduced in dimensionality using principal component analysis algorithm to obtain a dimensionality-reduced feature vector set.

[0021] The reduced feature vector set is trained using a support vector machine classifier to generate a classification model and obtain surface characteristic determination data.

[0022] In conjunction with the first aspect, the step of matching correction coefficients in a pre-established model based on the spectral characteristics, and using the correction coefficients to weight and adjust the scattered light distribution data to generate a corrected signal, includes:

[0023] The statistical feature values ​​of the distribution data are obtained through the spectral features, and the statistical feature set is obtained by using the mean and variance calculation methods.

[0024] The amplitude information corresponding to significant frequencies is extracted from the statistical feature set, and the amplitude significance is filtered according to a preset threshold to obtain the amplitude feature set;

[0025] The deviation value is calculated based on the amplitude feature set and the standard amplitude data in the pre-established model. The combination of correction coefficients is determined by the mapping relationship between the deviation value and the correction coefficient.

[0026] The scattered light distribution data is weighted and adjusted using the combination of correction coefficients to obtain a corrected signal dataset;

[0027] The peak offset of the frequency distribution is extracted from the corrected signal dataset, and the offset distribution pattern is determined by a support vector machine classifier to obtain the offset classification result;

[0028] The offset classification results are reduced in dimensionality using principal component analysis to obtain a set of feature vectors.

[0029] In conjunction with the first aspect, the step of extracting initial measured values ​​of geometric parameters from the corrected signal, and adjusting the initial measured values ​​through smoothing processing to obtain geometric data, includes:

[0030] The spectral distribution characteristics of the geometric parameters are obtained through the modified signal, and the spectral distribution is decomposed by Fast Fourier Transform to obtain the spectral feature set.

[0031] Significant frequency components related to geometric parameters are extracted from the spectral feature set, and frequency significance is filtered according to a preset threshold to obtain the frequency feature set;

[0032] The frequency deviation value is calculated based on the frequency feature set and the standard frequency data in the preset constraints. The adjustment coefficient is determined by using the mapping relationship between the deviation value and the geometric parameters.

[0033] The initial measurements are weighted and adjusted using the adjustment coefficients to obtain geometric data;

[0034] If the deviation between the geometric data and the preset constraints exceeds a preset threshold, the geometric data is smoothed using a Gaussian filtering method to obtain the new geometric data.

[0035] Statistical features of the distribution are extracted from the geometric data, and a set of statistical features is obtained by using the mean and standard deviation calculation method.

[0036] In conjunction with the first aspect, separating the boundary feature signal from the corrected signal and processing the boundary feature signal using an edge enhancement algorithm to generate boundary enhancement data includes:

[0037] The boundary feature signal is separated by signal decomposition using the corrected signal, and then frequency domain analysis is performed on the boundary feature signal using wavelet transform technology to obtain a frequency domain feature set.

[0038] Significant boundary frequency components are extracted from the frequency domain feature set, and frequency significance is filtered according to a preset threshold to obtain the boundary frequency set;

[0039] The boundary frequency set is convolved using a two-dimensional Gaussian kernel to obtain the boundary enhancement set;

[0040] Based on the boundary enhancement set and the boundary feature signal, principal component analysis is used to reduce the dimensionality of the enhanced data to obtain feature-reduced data.

[0041] If the deviation between the reduced feature data and the preset boundary constraints exceeds a preset threshold, the reduced feature data is smoothed using a Gaussian filtering method to obtain a smoothed feature set.

[0042] Boundary distribution statistical features are extracted from the smoothed feature set, and the mean and variance are calculated to obtain the statistical feature set.

[0043] In conjunction with the first aspect, the construction of a three-dimensional contour model using the boundary enhancement data and the geometric data, and the quantization calculation of the three-dimensional contour model using segmentation techniques to obtain quantized data, includes:

[0044] Using the boundary enhancement data and the geometric data, the target region is spatially discretized using triangulation mesh generation technology to generate a discrete mesh set;

[0045] Curvature values ​​of each grid vertex are extracted from the discrete grid set to form surface curvature features;

[0046] Gaussian filtering is used to smooth the curvature features to obtain a curvature feature set.

[0047] For the curvature feature set, principal component analysis is used to compress the feature dimension to three dimensions, generating a dimensionality-reduced feature set;

[0048] If the standard deviation of the dimensionality-reduced feature set exceeds a preset threshold, the dimensionality-reduced feature set is smoothed by mean filtering to obtain a smoothed feature set.

[0049] Based on the smoothed feature set, the feature points are grouped using the K-means clustering algorithm to generate grouped feature sets.

[0050] In conjunction with the first aspect, the step of extracting continuous features from the filtered image and processing the continuous features using a repair algorithm to generate repaired data includes:

[0051] Continuity features are extracted from the filtered image, and continuous quantization data is obtained by calculating the ratio of continuous length to spacing.

[0052] If the continuity quantization data is lower than a preset threshold, a morphological reconstruction algorithm is used to perform expansion and erosion operations on the fractured area to generate preliminary repair data.

[0053] Using the preliminary repaired data, the boundary point set of features is extracted using the Canny edge detection algorithm to obtain boundary feature data;

[0054] For the boundary feature data, the linear distribution characteristics are detected by Hough linear transform to generate a linear distribution feature set;

[0055] The distribution density of the fracture region is extracted from the linear distribution feature set, and the distribution density is smoothed by median filtering to obtain smoothed density data;

[0056] Based on the smooth density data, the K-nearest neighbor algorithm is used to classify the feature points of the fracture region and generate a set of classified feature points.

[0057] Repair data is obtained by matching the set of classification feature points with a pre-established template database.

[0058] In conjunction with the first aspect, the step of calculating the deviation based on the repaired data and the triangulation principle, and processing the deviation using an outlier removal algorithm to obtain deviation data includes:

[0059] The location and spatial information are extracted from the repaired data, and the initial deviation value is calculated using triangulation to generate the initial deviation value data.

[0060] For the initial deviation data, the isolated forest algorithm is used to detect outliers, outlier points are determined by calculating outlier scores, and deviation data is generated after removing outliers.

[0061] The spatial distribution characteristics of the deviation values ​​are obtained from the deviation data, and the distribution data is smoothed using a mean filter to obtain smoothed deviation distribution data.

[0062] Based on the smoothed deviation distribution data, the DBSCAN algorithm is used to cluster the feature points in the deviation region to generate a clustered feature point set.

[0063] The clustered feature point set is matched with a pre-established distribution template database, and the distance deviation between each feature point and the template is calculated using Euclidean distance to determine the optimized deviation distribution data.

[0064] Local density data is obtained from the optimized deviation distribution data, and the density data is smoothed by a median filter to obtain the final deviation value data.

[0065] In conjunction with the first aspect, after generating the quantized data, the method further includes:

[0066] The quantified data is compared with the preset bevel size standard range. If it exceeds the standard range, an alarm signal is generated and adjustment suggestions are output.

[0067] A second aspect of this disclosure provides an electronic device, comprising:

[0068] One or more processors;

[0069] A storage unit is used to store one or more programs that, when executed by one or more processors, enable the one or more processors to implement the method.

[0070] Beneficial Effects: The present invention provides an anti-interference measurement method for weld parameters. By using laser scattering technology and Fourier transform processing to obtain the spectral characteristics of highly reflective surfaces and matching correction coefficients, the method achieves weighted adjustment of the scattering signal and extraction of geometric parameters. Combined with edge enhancement algorithms and three-dimensional contour modeling technology, the method accurately quantifies the bevel size. Furthermore, by utilizing image filtering, morphological reconstruction, and outlier removal algorithms, the method effectively overcomes interference from surface oil, corrosion, and stripe fractures. Finally, through multi-source data fusion and adaptive calculation range adjustment, the method achieves high-precision and anti-interference measurement of weld bevel angle, blunt edge thickness, and misalignment amount under high reflectivity, corrosion, and complex working conditions, significantly improving the reliability and automation level of welding quality inspection. Attached Figure Description

[0071] Figure 1 This is a flowchart illustrating an anti-interference measurement method for weld parameters according to an embodiment of the present disclosure.

[0072] Figure 2 An electronic device according to an embodiment of this disclosure. Detailed Implementation

[0073] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with those disclosed herein.

[0074] The terminology used in this disclosure is for the purpose of describing particular embodiments only and is not intended to be limiting of the present disclosure. The singular forms “a,” “the,” and “the” as used in this disclosure and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any and all possible combinations of one or more of the associated listed items.

[0075] like Figure 1The diagram shown is a flowchart illustrating a method for measuring weld parameters against interference according to an embodiment of this disclosure, including:

[0076] S101: Obtain the scattered light distribution data of a highly reflective surface through laser scattering technology, and perform Fourier transform processing on the scattered light distribution data to obtain spectral characteristics;

[0077] The process involves acquiring scattered light distribution data from a highly reflective surface using laser scattering technology, performing Fourier transform on the scattered light distribution data to obtain spectral characteristics, including:

[0078] Initial light distribution data of highly reflective surfaces are collected using laser scattering technology, and the distribution pattern of scattered light at different angles is recorded to obtain an initial light distribution dataset.

[0079] Scattered light intensity information is extracted from the initial light distribution dataset, and the light intensity value at each angle is normalized to obtain scattered light intensity sequence data.

[0080] The scattered light intensity sequence data is processed using a fast Fourier transform algorithm to generate a spectral data set;

[0081] Spectral peaks and frequency distribution features are extracted from the spectral data set, and significant feature points are determined based on a preset threshold to obtain a spectral feature set;

[0082] The spectral feature set is reduced in dimensionality using principal component analysis algorithm to obtain a dimensionality-reduced feature vector set.

[0083] The reduced feature vector set is trained using a support vector machine classifier to generate a classification model and obtain surface characteristic determination data.

[0084] Specifically, in step S101, the scattered light distribution data of the highly reflective surface is obtained by laser scattering technology, and the data is processed by Fourier transform to extract the spectral features characterizing the surface properties.

[0085] In practice, the highly reflective metal surface is first scanned using laser scattering technology. When a laser beam irradiates the workpiece surface, a unique scattering distribution is generated due to differences in surface microstructure, material, and adhering substances (such as oil stains). The intensity of scattered light at different angles is collected by a high-precision sensor to form an initial light distribution dataset containing spatial and intensity information.

[0086] Subsequently, the core light intensity information was extracted from the dataset. To eliminate the influence of dimensions caused by differences in ambient lighting and equipment, the raw light intensity values ​​collected from each angle were normalized and converted to a unified scale, resulting in a stable and comparable sequence of scattered light intensity data.

[0087] Next, to reveal the periodic or characteristic patterns contained in the scattered signal, the Fast Fourier Transform (FFT) algorithm is used to transform this set of light intensity sequences in the time or spatial domain to the frequency domain. After FFT processing, a set of spectral data is generated, which contains amplitude and phase information representing the frequency components of different surface features.

[0088] Then, from the complex spectral data set, the algorithm automatically identifies and extracts key spectral peaks and specific frequency distribution features.

[0089] Based on thresholds calibrated through numerous experiments, the system can intelligently determine which are significant feature points characterizing surface material, oil stains, and other conditions, thereby selecting effective components that constitute the spectral feature set.

[0090] Since the initial spectral feature set has high dimensionality and may contain redundancy, Principal Component Analysis (PCA) is used to reduce the dimensionality of the spectral feature set in order to further condense information and highlight the main features. This process transforms the original features into a new set of fewer, independent features that contain the most information, resulting in a dimensionality-reduced feature vector set.

[0091] Finally, machine learning methods are used to enable the system to intelligently determine surface characteristics. A support vector machine (SVM) classifier is used to train the dimensionality-reduced feature vector set through supervised learning, generating a classification model that can accurately classify different surface states (such as clean metal, oily or slightly rusted surfaces).

[0092] By applying this model, new scattered light data can be evaluated, and the final judgment data characterizing the surface properties can be output, providing a key basis for subsequent signal correction.

[0093] S102: Match correction coefficients in a pre-established model based on the spectral characteristics, and use the correction coefficients to weight and adjust the scattered light distribution data to generate a corrected signal;

[0094] The step of matching correction coefficients in a pre-established model based on the spectral characteristics, and then using the correction coefficients to weight and adjust the scattered light distribution data to generate a corrected signal includes:

[0095] The statistical feature values ​​of the distribution data are obtained through the spectral features, and the statistical feature set is obtained by using the mean and variance calculation methods.

[0096] The amplitude information corresponding to significant frequencies is extracted from the statistical feature set, and the amplitude significance is filtered according to a preset threshold to obtain the amplitude feature set;

[0097] The deviation value is calculated based on the amplitude feature set and the standard amplitude data in the pre-established model. The combination of correction coefficients is determined by the mapping relationship between the deviation value and the correction coefficient.

[0098] The scattered light distribution data is weighted and adjusted using the combination of correction coefficients to obtain a corrected signal dataset;

[0099] The peak offset of the frequency distribution is extracted from the corrected signal dataset, and the offset distribution pattern is determined by a support vector machine classifier to obtain the offset classification result;

[0100] The offset classification results are reduced in dimensionality using principal component analysis to obtain a set of feature vectors.

[0101] Specifically, in step S102, the system matches the most suitable correction coefficients in a pre-established expert model based on the spectral characteristics obtained in the preceding steps, and uses these coefficients to weight and adjust the original scattered light distribution data, ultimately generating a corrected signal with stronger anti-interference capabilities. This process is a key step in achieving accurate measurement, aiming to eliminate or reduce signal distortion caused by high surface reflectivity and contamination.

[0102] The specific implementation path is as follows: First, the system will conduct in-depth statistical analysis on the acquired spectral features. By calculating statistical characteristic values ​​such as mean and variance, the complex spectral information is transformed into a set of statistical features that can quantify its distribution characteristics, laying a data foundation for subsequent accurate matching.

[0103] Next, the algorithm focuses on the frequency components most sensitive to surface conditions. It extracts the amplitude information corresponding to significant frequencies from the statistical feature set and filters out significant amplitude points based on a preset threshold, thus condensing the core amplitude feature set. This step ensures that subsequent corrections target the most critical signal-to-noise ratio points.

[0104] Subsequently, the system initiates the core matching calculation. It compares the currently measured amplitude feature set with the standard amplitude data in a standard model library established in advance using massive experimental data, and calculates the deviation value between the two. Each deviation value is transformed into a specific correction coefficient through a preset mapping function, ultimately forming a set of correction coefficients for the current surface state.

[0105] After obtaining the combination of correction coefficients, the system applies them to the original scattered light distribution data. This process is an adaptive weighted adjustment, where each data point is assigned a different weight based on its corresponding frequency characteristics, thereby generating a precisely calibrated corrected signal dataset. This effectively suppresses interference signals and enhances the representation of true surface morphology information.

[0106] To further ensure the quality of the correction, the system performs a closed-loop test on the corrected results. It extracts features such as the peak offset of the frequency distribution from the corrected signal dataset again, and uses a pre-trained support vector machine (SVM) classifier to intelligently judge the distribution pattern of these offsets, obtaining a classification result on the offset pattern to evaluate the correction effect.

[0107] Finally, to make the data more suitable for subsequent processing, the system uses Principal Component Analysis (PCA) to reduce the dimensionality of the above offset classification results, remove redundant information, condense the core features, and output a refined feature vector set. This set not only represents the core information of the corrected signal, but also serves as a high-quality input for subsequent 3D reconstruction and geometric parameter calculation.

[0108] S103: Extract the initial measured values ​​of the geometric parameters from the corrected signal, and adjust the initial measured values ​​through smoothing processing to obtain geometric data;

[0109] The initial measured values ​​of geometric parameters extracted from the corrected signal are then adjusted through smoothing processing to obtain geometric data, including:

[0110] The spectral distribution characteristics of the geometric parameters are obtained through the modified signal, and the spectral distribution is decomposed by Fast Fourier Transform to obtain the spectral feature set.

[0111] Significant frequency components related to geometric parameters are extracted from the spectral feature set, and frequency significance is filtered according to a preset threshold to obtain the frequency feature set;

[0112] The frequency deviation value is calculated based on the frequency feature set and the standard frequency data in the preset constraints. The adjustment coefficient is determined by using the mapping relationship between the deviation value and the geometric parameters.

[0113] The initial measurements are weighted and adjusted using the adjustment coefficients to obtain geometric data;

[0114] If the deviation between the geometric data and the preset constraints exceeds a preset threshold, the geometric data is smoothed using a Gaussian filtering method to obtain the new geometric data.

[0115] Statistical features of the distribution are extracted from the geometric data, and a set of statistical features is obtained by using the mean and standard deviation calculation method.

[0116] In step S103, the system extracts initial measurement values ​​reflecting the bevel geometry (such as angle and blunt edge thickness) from the interference-resistant corrected signal, and optimizes these initial values ​​through smoothing technology, ultimately outputting high-precision and high-stability geometric data. The core purpose of this step is to transform the corrected signal into reliable three-dimensional topographic information.

[0117] The specific implementation path is as follows: First, the system performs frequency domain analysis on the corrected signal to gain a deeper understanding of its contained geometric information. The signal's spectral distribution is decomposed using a Fast Fourier Transform (FFT) to obtain a spectral feature set containing various frequency components. Changes in geometric shape, such as the inclination of a bevel surface or the edge of a blunt edge, will exhibit obvious characteristics at specific frequencies.

[0118] The algorithm then focuses on identifying key frequency components that are strongly correlated with the geometric parameters. It extracts these significant frequency components from the spectral feature set and filters them according to a preset significance threshold, thus obtaining a set of frequency features that most effectively characterizes the current geometry. This step ensures that subsequent calculations are based on the most relevant information.

[0119] Next, the system initiates a calibration process. It compares the currently extracted frequency feature set with standard frequency data (derived from an ideal model or calibration sample) under preset constraints, and calculates the frequency deviation value. Each frequency deviation value is transformed into an adjustment coefficient for a specific geometric parameter through a pre-established mapping model.

[0120] After obtaining the adjustment coefficients, the system applies them to the initial measurements directly derived from the signal. This is a weighted adjustment process designed to systematically correct biases caused by residual noise or model mismatch, thereby generating a set of pre-optimized geometric data.

[0121] To ensure the reliability and smoothness of the data, the system performs a verification operation. It checks whether the geometric data meets preset constraints (such as angle range and thickness rationality). If the deviation between the data and the constraints exceeds a preset tolerance threshold, Gaussian filtering is used to smooth the geometric data. Gaussian filtering effectively suppresses high-frequency noise and abnormal fluctuations in the data while better preserving the true geometric trend, resulting in a smoother and more reasonable set of geometric data.

[0122] Finally, to quantitatively assess the quality of this final set of geometric data, the system extracts statistical features of its distribution, such as calculating the mean and standard deviation of all angle or thickness measurements. The resulting set of statistical features provides important reference information about the confidence level and dispersion of the current data for subsequent steps, such as 3D modeling.

[0123] S104: Separate the boundary feature signal from the corrected signal, process the boundary feature signal using an edge enhancement algorithm, and generate boundary enhancement data;

[0124] The step of separating the boundary feature signal from the corrected signal and processing the boundary feature signal using an edge enhancement algorithm to generate boundary enhancement data includes:

[0125] The boundary feature signal is separated by signal decomposition using the corrected signal, and then frequency domain analysis is performed on the boundary feature signal using wavelet transform technology to obtain a frequency domain feature set.

[0126] Significant boundary frequency components are extracted from the frequency domain feature set, and frequency significance is filtered according to a preset threshold to obtain the boundary frequency set;

[0127] The boundary frequency set is convolved using a two-dimensional Gaussian kernel to obtain the boundary enhancement set;

[0128] Based on the boundary enhancement set and the boundary feature signal, principal component analysis is used to reduce the dimensionality of the enhanced data to obtain feature-reduced data.

[0129] If the deviation between the reduced feature data and the preset boundary constraints exceeds a preset threshold, the reduced feature data is smoothed using a Gaussian filtering method to obtain a smoothed feature set.

[0130] Boundary distribution statistical features are extracted from the smoothed feature set, and the mean and variance are calculated to obtain the statistical feature set.

[0131] In step S104, the system addresses the common boundary ambiguity problem in blunt edge thickness measurement. Its core task is to accurately separate the feature signals representing the physical boundary from the information-rich correction signal, process them using advanced edge enhancement algorithms, and ultimately generate a dataset with clear boundaries and sharpened features, laying the foundation for subsequent precise quantization calculations.

[0132] The specific implementation involves a multi-stage, meticulous processing procedure. First, the system utilizes signal decomposition methods to extract the key feature signal components carrying boundary information from the complex modified signal. Next, wavelet transform technology is employed to perform multi-resolution frequency domain analysis on this boundary feature signal. Wavelet transform provides excellent localization characteristics in both the time and frequency domains, thereby accurately capturing the characteristics of boundary abrupt changes at different scales, forming a frequency domain feature set containing multi-scale information.

[0133] Subsequently, the algorithm intelligently extracts the most salient frequency components that are most sensitive to the boundaries from this frequency domain feature set. By filtering through a preset saliency threshold, the system is able to focus on the most critical information, resulting in a pure boundary frequency set that enhances boundary features.

[0134] To further sharpen edges and suppress noise, the system performs convolution operations on the boundary frequency set. This operation uses a two-dimensional Gaussian kernel as the convolution kernel, which enhances edge continuity while smoothing out high-frequency noise and unnecessary details, thereby generating a boundary enhancement set with a significantly improved signal-to-noise ratio.

[0135] Subsequently, the system combines the edge-enhanced data with the original boundary feature signals and uses Principal Component Analysis (PCA) to reduce the dimensionality of this fused high-dimensional data. PCA effectively eliminates information redundancy and noise, condensing the low-dimensional features that best represent the essential structure of the boundary, thus obtaining feature-reduced data.

[0136] To ensure that the generated boundary data is physically reasonable and consistent, the system compares it with preset boundary constraints (such as expected boundary positions and orientations). If the deviation exceeds the allowable threshold, the Gaussian filtering method is used again to smooth the feature reduction data, eliminating possible outliers and non-smooth transitions, thereby obtaining a more robust and smooth feature set.

[0137] Finally, the system extracts statistical features of the boundary distribution from this processed, smoothed feature set, such as quantifying the location and sharpness of the boundaries by calculating the mean and variance. The resulting statistical feature set not only provides a quantitative evaluation of the processing effect of this step but also offers high-quality boundary information input for the subsequent accurate construction of the 3D contour model.

[0138] S105: Construct a three-dimensional contour model using the boundary enhancement data and the geometric data, and perform quantization calculations on the three-dimensional contour model using segmentation technology to obtain quantized data;

[0139] The process involves constructing a three-dimensional contour model using the boundary enhancement data and the geometric data, and then using segmentation techniques to quantize the three-dimensional contour model to obtain quantized data, including:

[0140] Using the boundary enhancement data and the geometric data, the target region is spatially discretized using triangulation mesh generation technology to generate a discrete mesh set;

[0141] Curvature values ​​of each grid vertex are extracted from the discrete grid set to form surface curvature features;

[0142] Gaussian filtering is used to smooth the curvature features to obtain a curvature feature set.

[0143] For the curvature feature set, principal component analysis is used to compress the feature dimension to three dimensions, generating a dimensionality-reduced feature set;

[0144] If the standard deviation of the dimensionality-reduced feature set exceeds a preset threshold, the dimensionality-reduced feature set is smoothed by mean filtering to obtain a smoothed feature set.

[0145] Based on the smoothed feature set, the feature points are grouped using the K-means clustering algorithm to generate grouped feature sets.

[0146] In step S105, the system fuses the boundary enhancement data and geometric data obtained in the previous steps to construct a digital model that accurately reflects the three-dimensional morphology of the pipe bevel. Segmentation technology is then used to perform quantitative analysis on this model, ultimately outputting key dimensional quantitative data. This step is the core component in converting the raw signal into the final measurement index.

[0147] The specific implementation process is as follows: First, the system combines high-resolution boundary data with data representing macroscopic geometric morphology, and uses triangulation mesh generation technology to spatially discretize the target bevel region. This process transforms the continuous curved surface into a discrete geometric set composed of a large number of triangular meshes, establishing a digital framework for the three-dimensional contour model and providing a basic topological structure for subsequent detailed analysis.

[0148] Subsequently, the algorithm extracts geometric properties characterizing the surface curvature from the discrete mesh set. By calculating the curvature value of each mesh vertex, the system forms a set of surface curvature feature data that can finely describe the concavity, convexity, and turning features of the bevel surface. This data is key to identifying bevel angles and blunt edge regions.

[0149] To suppress potential noise fluctuations in curvature calculations and improve data stability, the system employs Gaussian filtering to smooth the original curvature features. After processing, a set of smoothed curvature features that better reflects the true shape trend is obtained.

[0150] Faced with potentially multidimensional and redundant curvature data, the system uses principal component analysis (PCA) to compress its features, projecting the core features of the data into a three-dimensional space to generate a dimensionality-reduced feature set that retains the main information and is easy to process.

[0151] The system continuously monitors data quality. If the standard deviation of the feature set after dimensionality reduction exceeds a preset threshold, it indicates abnormal fluctuations or outliers in the data. In this case, the system will smooth the data through mean filtering to further eliminate uncertainties and obtain a more robust and smooth feature set.

[0152] Finally, based on this processed high-quality feature data, the system uses the K-means clustering algorithm to intelligently group all feature points. This algorithm automatically aggregates points with similar geometric features (such as belonging to the same bevel face or blunt edge), thereby generating clear grouped feature sets. This process essentially segments the 3D model into different feature regions, providing a direct basis for the final calculation of quantitative parameters such as bevel angle and blunt edge thickness.

[0153] S106: Obtain the original image of the projection data, and perform denoising on the original image through filtering to generate a filtered image;

[0154] The step of extracting continuous features from the filtered image and processing the continuous features using a repair algorithm to generate repaired data includes:

[0155] Continuity features are extracted from the filtered image, and continuous quantization data is obtained by calculating the ratio of continuous length to spacing.

[0156] If the continuity quantization data is lower than a preset threshold, a morphological reconstruction algorithm is used to perform expansion and erosion operations on the fractured area to generate preliminary repair data.

[0157] Using the preliminary repaired data, the boundary point set of features is extracted using the Canny edge detection algorithm to obtain boundary feature data;

[0158] For the boundary feature data, the linear distribution characteristics are detected by Hough linear transform to generate a linear distribution feature set;

[0159] The distribution density of the fracture region is extracted from the linear distribution feature set, and the distribution density is smoothed by median filtering to obtain smoothed density data;

[0160] Based on the smooth density data, the K-nearest neighbor algorithm is used to classify the feature points of the fracture region and generate a set of classified feature points.

[0161] Repair data is obtained by matching the set of classification feature points with a pre-established template database.

[0162] In step S106, the system addresses the issue of stripe projection breakage caused by localized corrosion or deformation of the pipe surface during structured light scanning. By preprocessing and intelligently repairing the original projection image, complete stripe data is generated, providing reliable input for subsequent high-precision 3D measurements.

[0163] The specific implementation process is a sophisticated, step-by-step processing flow. First, the system extracts features representing the continuity of stripes from the filtered and denoised image. By calculating the ratio between the continuous length of the stripes and the break spacing, the system obtains a set of quantifiable continuity index data, which provides a basis for objectively judging the severity of stripe breakage.

[0164] When the system detects that the continuous quantization data is below a preset safety threshold, it indicates that there is a significant break in the stripes, requiring the initiation of a repair procedure. At this point, the system uses a morphological reconstruction algorithm to process the identified broken areas. This algorithm, through a series of precise dilation and erosion operations, can intelligently fill the gaps in the broken areas and eliminate noise interference based on the surrounding intact stripe structure features, thereby generating a preliminary repaired image.

[0165] To accurately locate the edge contours of the repaired stripes, the system performs Canny edge detection on the initial repair data. The Canny algorithm, with its excellent signal-to-noise ratio and accurate single-edge response characteristics, can extract clear and coherent stripe boundary point sets, providing precise boundary feature data for subsequent geometric analysis.

[0166] Since structured light stripes typically exhibit strong linear distribution characteristics, the system then employs the Hough line transform to process the boundary feature data. The Hough transform can effectively detect linear components in the image and associate broken edge segments that belong to the same line, thereby generating a feature set describing the linear distribution characteristics of all stripes, such as direction and position.

[0167] Next, the system analyzes the distribution of the fracture region from the linear distribution feature set and extracts its distribution density information. To eliminate potential impulse noise and abnormal fluctuations in the density data, the system uses median filtering to smooth it, resulting in smooth density data that more accurately reflects the distribution of the fracture region.

[0168] S107: Extract continuous features from the filtered image, process the continuous features, and generate repair data;

[0169] In step S107, the system focuses on solving the problem of structured light stripes breaking due to surface corrosion or deformation. By extracting and intelligently repairing the stripe continuity features in the filtered image, complete stripe data that can be used for accurate measurement is generated.

[0170] First, the continuity features of the stripes are extracted from the filtered image. The system generates quantified continuity index data by calculating the connectivity between adjacent stripe segments, the width of the gaps between breaks, and the coherence of the stripe direction. These data objectively reflect the integrity and quality of the striped image.

[0171] When the system detects that the continuity index is below a preset threshold, it will automatically trigger the repair process. Morphological reconstruction algorithms are used to process the identified fracture areas. By designing reasonable structural elements to perform dilation and erosion operations, the broken stripe segments are intelligently connected while maintaining the original stripe morphological characteristics, generating preliminary repaired image data.

[0172] To accurately obtain the geometric features of the repaired stripes, the system uses the Canny edge detection algorithm to process the preliminary repair data. The Canny algorithm extracts a precise and coherent set of stripe edge feature points through steps such as Gaussian filtering, gradient magnitude calculation, non-maximum suppression, and double threshold detection.

[0173] To address the linear characteristics typically found in structured light stripes, the system further employs the Hough transform to analyze the set of edge feature points. The Hough transform can detect linear components in an image and connect discrete edge points into complete linear features, generating a linearly distributed feature set describing parameters such as stripe direction, position, and length.

[0174] The system analyzes the distribution characteristics of fracture regions from linear distribution feature sets, paying particular attention to the density distribution of fracture points. Median filtering is used to smooth the density distribution data, effectively eliminating outliers and discontinuities, resulting in smoothed density data that better reflects the true fracture mode.

[0175] Based on the smoothed density data, the system uses the K-nearest neighbor algorithm to classify the feature points in the fracture region. The algorithm divides the feature points into different categories based on their similarity in spatial distribution, density characteristics, and geometric attributes, generating a set of categorized feature points.

[0176] Finally, the system matches the classified feature point set with a pre-established stripe template database. By calculating feature similarity and spatial distribution consistency, the most suitable repair template is selected, and the corresponding repair strategy is applied. This ultimately outputs high-quality structured light stripe repair data, providing accurate and reliable input for subsequent 3D measurements.

[0177] S108: Calculate the deviation based on the repaired data and the triangulation principle, and process the deviation using an outlier removal algorithm to obtain the deviation data;

[0178] The step involves calculating the deviation based on the repaired data and the triangulation principle, and then processing the deviation using an outlier removal algorithm to obtain deviation data, including:

[0179] The location and spatial information are extracted from the repaired data, and the initial deviation value is calculated using triangulation to generate the initial deviation value data.

[0180] For the initial deviation data, the isolated forest algorithm is used to detect outliers, outlier points are determined by calculating outlier scores, and deviation data is generated after removing outliers.

[0181] The spatial distribution characteristics of the deviation values ​​are obtained from the deviation data, and the distribution data is smoothed using a mean filter to obtain smoothed deviation distribution data.

[0182] Based on the smoothed deviation distribution data, the DBSCAN algorithm is used to cluster the feature points in the deviation region to generate a clustered feature point set.

[0183] The clustered feature point set is matched with a pre-established distribution template database, and the distance deviation between each feature point and the template is calculated using Euclidean distance to determine the optimized deviation distribution data.

[0184] Local density data is obtained from the optimized deviation distribution data, and the density data is smoothed by a median filter to obtain the final deviation value data.

[0185] In step S108, the system accurately calculates the misalignment in the pipe assembly based on the repaired high-quality stripe data and the optical triangulation principle, and ensures the accuracy and reliability of the final measurement results through advanced outlier processing technology.

[0186] The specific implementation process is as follows: First, the system accurately extracts the position and spatial information of the stripes from the repair data. Utilizing the principle of structured light triangulation, the system calculates the three-dimensional coordinates of each measurement point by calculating the geometric relationship between the projector, camera, and the measured surface, thereby obtaining the initial deviation data characterizing the misalignment of the pipeline.

[0187] Considering the potential presence of residual noise or local interference in the actual measurement environment, the system employs the Isolation Forest algorithm to detect anomalies in the initial deviation data. This algorithm calculates anomaly scores for each data point by randomly partitioning the data space, efficiently identifying and removing outliers that deviate from the main distribution, thus generating cleaned deviation data.

[0188] To further improve the distribution characteristics of the data, the system analyzes the spatial distribution pattern of the deviation data. A mean filter is used to smooth the spatial distribution data of the deviation, effectively suppressing random fluctuations while preserving the true deviation trend, resulting in smoothed deviation distribution data.

[0189] Based on the smoothed distribution data, the system uses the DBSCAN clustering algorithm to intelligently group feature points in the deviation region. This algorithm, based on the concept of density reachability, can automatically identify sets of points with similar deviation characteristics and aggregate them into different clusters, generating cluster feature point sets.

[0190] The system matches the clustered feature point set with a pre-established database of typical deviation distribution templates. By calculating the Euclidean distance between each feature point and the template, the similarity is quantified, thereby determining the optimal deviation distribution pattern and obtaining optimized deviation distribution data.

[0191] Finally, the system extracts local density features from the optimized deviation distribution data and smooths the density data using a median filter. Median filtering effectively eliminates impulse noise in the density data while preserving the edge characteristics of the density distribution, ultimately outputting high-precision final deviation values, providing a reliable quantitative basis for pipeline quality assessment.

[0192] S109: Obtain depth map data, process the depth map data through classification and segmentation, adjust the calculation range by combining the deviation data and the quantization data, and generate quantization data.

[0193] In step S109, the system performs the final data fusion and precise calculation.

[0194] First, the system acquires raw depth map data collected by a 3D sensor. This data contains depth information for each pixel in the measured area, forming a basic point cloud describing the 3D topography of the surface.

[0195] Subsequently, the system performs classification and segmentation processing on this depth map data. This process typically employs semantic segmentation algorithms based on machine learning or deep learning, aiming to classify each point or region in the depth map into different semantic categories, such as: bevel slopes, blunt edge planes, pipe base material surfaces, and potential areas of weld spatter or corrosion. Through this step, the system accurately identifies and isolates the bevel target areas that require focused analysis, defining the scope for subsequent precise calculations.

[0196] Next, the system initiates a crucial data fusion and adaptive adjustment process. It uses the deviation data obtained in the previous step (such as the misalignment amount calculated in S108) and the quantization data (such as the bevel angle and blunt edge thickness calculated in S105) as important prior information and constraints.

[0197] Adjusting based on deviation data: The system uses misalignment information to correct the relative position of the calculation area. For example, if misalignment between upper and lower pipe openings is detected, the system will correspondingly translate or rotate the calculation areas of the bevels of the two pipe openings to ensure that subsequent dimensional measurements are performed under the correct spatial relationship, thereby eliminating the coupling effect of misalignment on the bevel's own dimensional measurement.

[0198] Adjustments based on quantitative data: The system utilizes preliminarily calculated geometric parameters (such as the estimated range of angles and thicknesses) to define a more precise and reasonable calculation range. For example, the angle search range is constrained to within ±5° of the initial value, and the measurement area for the blunt edge thickness is precisely limited to the classified blunt edge area, avoiding calculations in irrelevant areas, thereby effectively improving computational efficiency and anti-interference capabilities.

[0199] After the aforementioned dynamic adjustment of the calculation range based on prior information, the system performs final refined quantitative calculations on the classified target regions (such as blunt edges and bevel surfaces) within the optimized precise region. This includes using the least squares method to fit the plane equation of the bevel surface to calculate the precise angle, and accurately calculating the thickness distribution of the blunt edge region in the normal direction.

[0200] Finally, the system outputs a set of final quantitative data that has undergone multiple data verifications and optimizations. This set of data not only includes the weld's own dimensions such as bevel angle and blunt edge thickness, but may also include comprehensive parameters related to the assembly status. Its reliability and accuracy are significantly improved due to the data fusion and range optimization throughout the entire process, providing the final basis for judging welding quality.

[0201] The processor 201 can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor.

[0202] The memory 202 can be an internal storage unit of the electronic device 200, such as a hard disk or RAM of the electronic device 200. The memory 202 can also be an external storage device of the electronic device 200, such as a plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card, or Flash Card equipped on the electronic device 200. Furthermore, the memory 202 can include both internal and external storage units of the electronic device 200. The memory 202 is used to store the computer program 203 and other programs and data required by the electronic device. The memory 202 can also be used to temporarily store data that has been output or will be output.

[0203] In the embodiments provided in this disclosure, it should be understood that the disclosed devices / electronic devices and methods can be implemented in other ways. For example, the device / electronic device embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. 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, and the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0204] If an integrated module / unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program may include computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. A computer-readable medium may include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in a computer-readable medium may be appropriately added to or subtracted according to the requirements of legislation and patent practice in a jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media may not include electrical carrier signals and telecommunication signals.

[0205] The above embodiments are only used to illustrate the technical solutions of this disclosure, and are not intended to limit it. Although this disclosure has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this disclosure, and should all be included within the protection scope of this disclosure.

Claims

1. A method for measuring weld parameters against interference, characterized in that, include: The scattered light distribution data of a highly reflective surface is obtained by laser scattering technology, and the scattered light distribution data is subjected to Fourier transform processing to obtain spectral characteristics; Based on the spectral characteristics, correction coefficients are matched in a pre-established model, and the scattered light distribution data are weighted and adjusted using the correction coefficients to generate a corrected signal. The initial measured values ​​of the geometric parameters are extracted from the corrected signal, and the initial measured values ​​are adjusted through smoothing to obtain the geometric data; The boundary feature signal is separated from the corrected signal, and the boundary feature signal is processed by an edge enhancement algorithm to generate boundary enhancement data; A three-dimensional contour model is constructed using the boundary enhancement data and the geometric data. The three-dimensional contour model is then quantized using segmentation technology to obtain the first quantized data. The original image of the projection data is acquired, and the original image is denoised by filtering to generate a filtered image; Continuous features are extracted from the filtered image, and the continuous features are processed to generate repaired data; The deviation is calculated based on the repaired data and the principle of triangulation. The deviation is then processed using an outlier removal algorithm to obtain the deviation data. Depth map data is acquired, processed by classification and segmentation, and the relative position of the calculation area is corrected by using edge misalignment information. The angle search range is constrained to within ±5° of the initial value for quantization calculation to generate second quantized data.

2. The method according to claim 1, characterized in that, The process involves acquiring scattered light distribution data from a highly reflective surface using laser scattering technology, performing Fourier transform on the scattered light distribution data to obtain spectral characteristics, including: Initial light distribution data of highly reflective surfaces are collected using laser scattering technology, and the distribution pattern of scattered light at different angles is recorded to obtain an initial light distribution dataset. Scattered light intensity information is extracted from the initial light distribution dataset, and the light intensity value at each angle is normalized to obtain scattered light intensity sequence data. The scattered light intensity sequence data is processed using a fast Fourier transform algorithm to generate a spectral data set; Spectral peaks and frequency distribution features are extracted from the spectral data set, and significant feature points are determined based on a preset threshold to obtain a spectral feature set; The spectral feature set is reduced in dimensionality using principal component analysis algorithm to obtain a dimensionality-reduced feature vector set. The reduced feature vector set is trained using a support vector machine classifier to generate a classification model and obtain surface characteristic determination data.

3. The method according to claim 1, characterized in that, The step of matching correction coefficients in a pre-established model based on the spectral characteristics, and then using the correction coefficients to weight and adjust the scattered light distribution data to generate a corrected signal includes: The statistical feature values ​​of the distribution data are obtained through the spectral features, and the statistical feature set is obtained by using the mean and variance calculation methods. The amplitude information corresponding to significant frequencies is extracted from the statistical feature set, and the amplitude significance is filtered according to a preset threshold to obtain the amplitude feature set; The deviation value is calculated based on the amplitude feature set and the standard amplitude data in the pre-established model. The combination of correction coefficients is determined by the mapping relationship between the deviation value and the correction coefficient. The scattered light distribution data is weighted and adjusted using the combination of correction coefficients to obtain a corrected signal dataset; The peak offset of the frequency distribution is extracted from the corrected signal dataset, and the offset distribution pattern is determined by a support vector machine classifier to obtain the offset classification result; The offset classification results are reduced in dimensionality using principal component analysis to obtain a set of feature vectors.

4. The method according to claim 1, characterized in that, The initial measured values ​​of geometric parameters extracted from the corrected signal are then adjusted through smoothing processing to obtain geometric data, including: The spectral distribution characteristics of the geometric parameters are obtained through the modified signal, and the spectral distribution is decomposed by Fast Fourier Transform to obtain the spectral feature set. Significant frequency components related to geometric parameters are extracted from the spectral feature set, and frequency significance is filtered according to a preset threshold to obtain the frequency feature set; The frequency deviation value is calculated based on the frequency feature set and the standard frequency data in the preset constraints. The adjustment coefficient is determined by using the mapping relationship between the deviation value and the geometric parameters. The initial measurements are weighted and adjusted using the adjustment coefficients to obtain geometric data; If the deviation between the geometric data and the preset constraints exceeds a preset threshold, the geometric data is smoothed using a Gaussian filtering method to obtain the new geometric data. Statistical features of the distribution are extracted from the geometric data, and a set of statistical features is obtained by using the mean and standard deviation calculation method.

5. The method according to claim 1, characterized in that, The step of separating the boundary feature signal from the corrected signal and processing the boundary feature signal using an edge enhancement algorithm to generate boundary enhancement data includes: The boundary feature signal is separated by signal decomposition using the corrected signal, and then frequency domain analysis is performed on the boundary feature signal using wavelet transform technology to obtain a frequency domain feature set. Significant boundary frequency components are extracted from the frequency domain feature set, and frequency significance is filtered according to a preset threshold to obtain the boundary frequency set; The boundary frequency set is convolved using a two-dimensional Gaussian kernel to obtain the boundary enhancement set; Based on the boundary enhancement set and the boundary feature signal, principal component analysis is used to reduce the dimensionality of the enhanced data to obtain feature-reduced data. If the deviation between the reduced feature data and the preset boundary constraints exceeds a preset threshold, the reduced feature data is smoothed using a Gaussian filtering method to obtain a smoothed feature set. Boundary distribution statistical features are extracted from the smoothed feature set, and the mean and variance are calculated to obtain the statistical feature set.

6. The method according to claim 1, characterized in that, The process involves constructing a three-dimensional contour model using the boundary enhancement data and the geometric data, and then performing quantization calculations on the three-dimensional contour model using segmentation techniques to obtain first quantized data, including: Using the boundary enhancement data and the geometric data, the target region is spatially discretized using triangulation mesh generation technology to generate a discrete mesh set; Curvature values ​​of each grid vertex are extracted from the discrete grid set to form surface curvature features; Gaussian filtering is used to smooth the curvature features to obtain a curvature feature set. For the curvature feature set, principal component analysis is used to compress the feature dimension to three dimensions, generating a dimensionality-reduced feature set; If the standard deviation of the dimensionality-reduced feature set exceeds a preset threshold, the dimensionality-reduced feature set is smoothed by mean filtering to obtain a smoothed feature set. Based on the smoothed feature set, the feature points are grouped using the K-means clustering algorithm to generate grouped feature sets.

7. The method according to claim 1, characterized in that, The step of extracting continuous features from the filtered image and processing the continuous features using a repair algorithm to generate repaired data includes: Continuity features are extracted from the filtered image, and continuous quantization data is obtained by calculating the ratio of continuous length to spacing. If the continuity quantization data is lower than a preset threshold, a morphological reconstruction algorithm is used to perform expansion and erosion operations on the fractured area to generate preliminary repair data. Using the preliminary repaired data, the boundary point set of features is extracted using the Canny edge detection algorithm to obtain boundary feature data; For the boundary feature data, the linear distribution characteristics are detected by Hough linear transform to generate a linear distribution feature set; The distribution density of the fracture region is extracted from the linear distribution feature set, and the distribution density is smoothed by median filtering to obtain smoothed density data; Based on the smooth density data, the K-nearest neighbor algorithm is used to classify the feature points of the fracture region and generate a set of classified feature points. Repair data is obtained by matching the set of classification feature points with a pre-established first template database.

8. The method according to claim 1, characterized in that, The step involves calculating the deviation based on the repaired data and the triangulation principle, and then processing the deviation using an outlier removal algorithm to obtain deviation data, including: The location and spatial information are extracted from the repaired data, and the initial deviation value is calculated using triangulation to generate the initial deviation value data. For the initial deviation data, the isolated forest algorithm is used to detect outliers, outlier points are determined by calculating outlier scores, and deviation data is generated after removing outliers. The spatial distribution characteristics of the deviation values ​​are obtained from the deviation data, and the distribution data is smoothed using a mean filter to obtain smoothed deviation distribution data. Based on the smoothed deviation distribution data, the DBSCAN algorithm is used to cluster the feature points in the deviation region to generate a clustered feature point set. The clustered feature point set is matched with a pre-established second template database, and the distance deviation between each feature point and the template is calculated using Euclidean distance to determine the optimized deviation distribution data. Local density data is obtained from the optimized deviation distribution data, and the density data is smoothed by a median filter to obtain the final deviation value data.

9. The method according to claim 1, characterized in that, After generating the second quantized data, the method further includes: The second quantified data is compared with the preset bevel size standard range. If it exceeds the standard range, an alarm signal is generated and an adjustment suggestion is output.

10. An electronic device, characterized in that, include: One or more processors; A storage unit for storing one or more programs that, when executed by one or more processors, enable the one or more processors to implement the method according to any one of claims 1 to 9.

Images