Detection Method, Device, Equipment and Storage Medium for Curved Flexible Screen

Abstract

The present application relates to the technical field of display screen detection, and discloses a detection method, device, equipment and storage medium for a curved flexible screen. The method includes: measuring physical parameters of a curved flexible screen sample to obtain an initial parameter data set of the sample, and performing three-dimensional laser scanning on the curved flexible screen sample based on the initial parameter data set of the sample to obtain sample three-dimensional point cloud data; calculating multi-scale curvatures of the sample three-dimensional point cloud data to obtain curved surface curvature distribution data; performing stress analysis and solving equilibrium equations on the sample three-dimensional point cloud data to obtain curved surface stress distribution data; performing feature extraction and feature vector coding fusion on the curved surface curvature distribution data and the curved surface stress distribution data to obtain a curvature-stress distribution relationship matrix; inputting the curvature-stress distribution relationship matrix into a pre-set curved surface defect detection model for curved surface defect detection to obtain a curved surface defect detection result. The present application improves the detection accuracy of the curved flexible screen.

Description

Technical Field

[0001] The present application relates to the technical field of display screen detection, and particularly to a detection method, device, equipment and storage medium for a curved flexible screen. Background Art

[0002] With the rapid development of display technology, curved flexible screens have become an important trend in the field of electronic devices. This new display technology not only provides a better visual experience but also greatly increases the flexibility of device design. However, the manufacturing and quality control of curved flexible screens face unprecedented challenges. Traditional flat display screen detection methods are difficult to adapt to the complex geometric shapes and material properties of curved flexible screens, resulting in low detection efficiency and insufficient accuracy.

[0003] In addition, complex stress distributions will occur in curved flexible screens in a bent state, and these stresses may cause minor structural changes or potential defects, which are often difficult to detect by conventional methods. Existing detection technologies often separate curvature analysis and stress analysis and cannot comprehensively reflect the actual state of curved flexible screens, thus affecting the evaluation of product quality and reliability. Summary of the Invention

[0004] The present application provides a detection method, device, equipment and storage medium for a curved flexible screen, which is used to improve the detection accuracy of curved flexible screens.

[0005] In a first aspect, the present application provides a detection method for a curved flexible screen, and the detection method for the curved flexible screen includes:

[0006] Measure the physical parameters of a curved flexible screen sample to obtain an initial parameter dataset of the sample, and perform three-dimensional laser scanning on the curved flexible screen sample based on the initial parameter dataset of the sample to obtain sample three-dimensional point cloud data;

[0007] Perform multi-scale curvature calculation on the sample three-dimensional point cloud data to obtain curved surface curvature distribution data;

[0008] Perform stress analysis and balance equation solution on the sample three-dimensional point cloud data to obtain curved surface stress distribution data;

[0009] Perform feature extraction and feature vector encoding fusion on the curved surface curvature distribution data and the curved surface stress distribution data to obtain a curvature-stress distribution relationship matrix;

[0010] Input the curvature-stress distribution relationship matrix into a preset curved surface defect detection model for curved surface defect detection to obtain a curved surface defect detection result.

[0011] In a second aspect, the present application provides a detection device for a curved flexible screen, and the detection device for the curved flexible screen includes:

[0012] A scanning module, configured to measure physical parameters of a curved flexible screen sample to obtain an initial parameter data set of the sample, and perform three-dimensional laser scanning on the curved flexible screen sample based on the initial parameter data set of the sample to obtain sample three-dimensional point cloud data;

[0013] A calculation module, configured to perform multi-scale curvature calculation on the sample three-dimensional point cloud data to obtain curved surface curvature distribution data;

[0014] An analysis module, configured to perform stress analysis and solve equilibrium equations on the sample three-dimensional point cloud data to obtain curved surface stress distribution data;

[0015] A fusion module, configured to perform feature extraction and feature vector coding fusion on the curved surface curvature distribution data and the curved surface stress distribution data to obtain a curvature-stress distribution relationship matrix;

[0016] A detection module, configured to input the curvature-stress distribution relationship matrix into a pre-set curved surface defect detection model for curved surface defect detection to obtain a curved surface defect detection result.

[0017] In a third aspect of the present application, there is provided a detection device for a curved flexible screen, including: a memory and at least one processor, wherein instructions are stored in the memory; the at least one processor invokes the instructions in the memory so that the detection device for the curved flexible screen executes the above-mentioned detection method for the curved flexible screen.

[0018] In a fourth aspect of the present application, there is provided a computer-readable storage medium, wherein instructions are stored in the computer-readable storage medium, and when the instructions are run on a computer, the computer is enabled to execute the above-mentioned detection method for the curved flexible screen.

[0019] In the technical solution provided by this application, through comprehensive physical parameter measurement and three-dimensional laser scanning of the sample, rich initial data is obtained. By using Gaussian filtering and k-d tree index structure, multi-scale accurate description of the surface geometric features is realized, and the curvature changes at different scales are effectively captured. Combining the finite element method and non-linear solution technology, accurate calculation of the internal stress state of the surface is realized. By extracting features and encoding and fusing the curvature and stress data, a unique curvature-stress distribution relationship matrix is constructed, which comprehensively reflects the geometric and mechanical properties of the surface. By using a model based on multiple high-frequency projections and inverse transfer learning, combined with the self-attention mechanism, the accuracy and robustness of defect detection are greatly improved. Through non-maximum suppression and feature contrast analysis, accurate identification and positioning of the defect type are realized, and it can adapt to the curved flexible screens with different curvatures and stress distributions, thus improving the detection accuracy of the curved flexible screens. Brief Description of the Drawings

[0020] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can also be obtained based on these drawings.

[0021] Figure 1 It is a schematic diagram of an embodiment of the detection method for the curved flexible screen in the embodiment of this application;

[0022] Figure 2 It is a schematic diagram of an embodiment of the detection device for the curved flexible screen in the embodiment of this application. Detailed Embodiments

[0023] The embodiments of this application provide a detection method, device, equipment and storage medium for a curved flexible screen. The terms "first", "second", "third", "fourth", etc. (if any) in the description and claims of this application and the above accompanying drawings are used to distinguish similar objects, and do not necessarily need to be used to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments described here can be implemented in an order other than those illustrated or described here. In addition, the terms "comprising" or "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or equipment comprising a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these process, method, product or equipment.

[0024] For easy understanding, the specific process of the embodiments of this application will be described below. Please refer to Figure 1, an embodiment of the detection method of the curved flexible screen in the embodiments of the present application includes:

[0025] Step S101: Measure the physical parameters of the curved flexible screen sample to obtain the initial parameter dataset of the sample, and perform three-dimensional laser scanning on the curved flexible screen sample based on the initial parameter dataset of the sample to obtain the three-dimensional point cloud data of the sample;

[0026] It can be understood that the execution subject of the present application can be a detection device for a curved flexible screen, or a terminal or a server. Specifically, it is not limited here. In the embodiments of the present application, the server is taken as the execution subject for illustration.

[0027] Specifically, measure the size of the curved flexible screen sample to obtain the length, width, and thickness data of the sample. These data are basic physical parameters used to analyze the geometric and quality characteristics of the sample. According to the length, width, and thickness data, calculate the density of the curved flexible screen sample to obtain the density distribution data, which reflects the material distribution of the sample in different regions. Normalize the length, width, and thickness data to eliminate the influence of the data dimension, making the data more suitable for subsequent calculations and analyses, and obtain the sample size feature vector. Calculate the sample mass feature map according to the density distribution data. The mass feature map can visually display the mass distribution of the sample, facilitating the identification of potential defect areas. Generate the initial parameter dataset of the sample according to the sample size feature vector and the sample mass feature map. Generate multiple sets of candidate scanning parameters according to the initial parameter dataset of the sample. These candidate scanning parameters represent different scanning schemes. In order to determine the best scanning scheme, perform a quality assessment on the candidate scanning parameters, and obtain the optimal three-dimensional laser scanning parameters through the assessment. These parameters can improve the scanning efficiency and quality while ensuring the scanning accuracy. Perform three-dimensional laser scanning on the curved flexible screen sample through the three-dimensional laser scanning parameters to obtain the three-dimensional point cloud data of the sample. The three-dimensional point cloud data is an accurate description of the surface geometry of the sample, containing all the detailed information on the surface of the sample.

[0028] Step S102: Perform multi-scale curvature calculation on the three-dimensional point cloud data of the sample to obtain the curved surface curvature distribution data;

[0029] Specifically, Gaussian filtering is performed on the three-dimensional point cloud data of the sample to reduce the noise in the point cloud data, and the denoised point cloud data is obtained. Based on the denoised point cloud data, a k-d tree index structure is constructed, which helps to quickly search for neighborhoods and form a fast neighborhood search tree. The multi-scale neighborhood radius is set for the fast neighborhood search tree. By setting different neighborhood radii, neighborhood point sets of different scales are obtained. Each scale of neighborhood point set represents a different spatial resolution, which helps to comprehensively analyze the geometric characteristics of the surface. The principal curvature is estimated for each point according to the neighborhood point sets of different scales. The principal curvature estimation calculates the curvature change of each point in its neighborhood to obtain the point-level principal curvature vector, and these vectors describe the curvature characteristics of each point at different scales. A tensor field is constructed for the point-level principal curvature vector to obtain the curvature tensor field. The tensor field synthesizes the principal curvature information of all points and provides the overall curvature distribution of the surface. According to the curvature tensor field, the Gaussian curvature and the mean curvature are calculated to obtain the multi-scale curvature feature map. Wavelet transform is performed on the multi-scale curvature feature map to obtain a multi-resolution representation of the curvature. The wavelet transform can decompose the curvature feature map into different resolution representation forms. Feature points are extracted according to the multi-resolution representation of the curvature to obtain a curvature feature point set, and the points with significant curvature changes on the surface are found. These points are usually the key positions of surface deformation or defects. Spatial clustering analysis is performed on the curvature feature point set. By analyzing the spatial distribution of these feature points, regions with significant curvature changes are obtained. These regions usually correspond to the important features or potential defect positions of the surface. Interpolation fitting is performed according to the regions with significant curvature changes and the multi-scale curvature feature map to convert the discrete feature point information into continuous curvature distribution data, and the surface curvature distribution data is obtained.

[0030] Step S103: Perform stress analysis and solve the equilibrium equation for the three-dimensional point cloud data of the sample to obtain the surface stress distribution data;

[0031] Specifically, for the three-dimensional point cloud data of the sample, surface reconstruction is performed to convert the discrete point cloud data into a continuous geometric surface model. A finite element mesh is constructed based on the geometric surface model, and the finite element mesh discretizes the geometric model to form a computational domain. Material property assignment is carried out for the discretized computational domain, and the material properties of the sample such as elastic modulus and Poisson's ratio are assigned to each node of the mesh to obtain a node-level material parameter matrix. Based on the node-level material parameter matrix, a local stiffness matrix is constructed, and the local stiffness matrix describes the ability of the material to respond to external forces at the node level. All local stiffness matrices are assembled to obtain a global stiffness matrix, and the global stiffness matrix describes the stiffness characteristics of the entire computational domain. Boundary conditions are applied to the global stiffness matrix, and actual physical constraint conditions such as fixed supports and external loads are applied to the model to obtain a modified stiffness equation. Based on the modified stiffness equation, a system of equilibrium equations is constructed, and the system of equilibrium equations is a mathematical expression that describes the equilibrium state of the system under the action of external forces. The Newton-Raphson iteration method is used to solve the system of equilibrium equations. This method continuously corrects the node displacement values through iterative calculations until convergence is achieved, and a node displacement field is obtained. The node displacement field describes the displacement changes of each node under the action of external forces. Based on the node displacement field, the strain tensor is calculated, and the strain tensor describes the strain state of the material during the deformation process to obtain a continuous strain distribution field. Constitutive equation conversion is performed on the continuous strain distribution field. The constitutive equation is an equation in material mechanics that describes the relationship between stress and strain. Through constitutive equation conversion, the strain distribution field is converted into an initial stress distribution field. The initial stress distribution field describes the stress distribution of the material in the initial state. Stress equilibrium correction and singularity point processing are performed based on the initial stress distribution field. Stress equilibrium correction is to ensure that the entire system reaches an equilibrium state under the action of external forces, and singularity point processing is to correct the abnormal points in the stress distribution to eliminate errors in numerical calculations, and finally the surface stress distribution data is obtained.

[0032] Step S104: Extract features and encode and fuse eigenvectors for the surface curvature distribution data and the surface stress distribution data to obtain a curvature-stress distribution relationship matrix;

[0033] Specifically, perform principal component analysis on the surface curvature distribution data to extract the main features in the curvature data and obtain the curvature principal eigenvectors. At the same time, calculate the principal stress and equivalent stress based on the surface stress distribution data to obtain the stress feature field. The stress feature field is a key description of the stress distribution and can reflect the stress conditions of the material at different positions. Perform singular value decomposition on the stress feature field, decompose the complex stress feature field into a set of principal eigenvectors, and obtain the stress principal eigenvectors. These principal eigenvectors are the main features of the stress distribution and can effectively simplify the representation and analysis of stress data. Construct a joint feature space based on the curvature principal eigenvectors and stress principal eigenvectors. The joint feature space synthesizes the main features of curvature and stress to obtain the initial feature matrix. Perform non-linear dimensionality reduction on the initial feature matrix to map the high-dimensional feature data to a low-dimensional space and obtain the low-dimensional feature representation. The low-dimensional feature representation can retain the main information of the original data while simplifying the data complexity. Construct an autoencoder neural network based on the low-dimensional feature representation. The autoencoder neural network includes a feature encoder and a decoder and can automatically learn and extract the latent features of the data. Optimize the sparse constraint on the hidden layer output of the feature encoder to enhance the feature recognition ability and obtain the sparse feature encoding. Calculate the curvature-stress correlation coefficient based on the sparse feature encoding. The correlation coefficient can quantify the relationship between curvature and stress to obtain the correlation intensity matrix. The correlation intensity matrix reflects the coupling intensity of curvature and stress. Perform spectral clustering analysis on the correlation intensity matrix. Spectral clustering analysis can discover the latent patterns in the data to obtain the curvature-stress coupling pattern. The curvature-stress coupling pattern is an important description of the surface characteristics and helps to understand the deformation behavior of the surface under stress conditions. Perform a tensor product operation based on the curvature-stress coupling pattern and the sparse feature encoding to synthesize the multi-dimensional feature data and obtain the curvature-stress distribution relationship matrix.

[0034] Step S105: Input the curvature-stress distribution relationship matrix into a pre-set surface defect detection model for surface defect detection to obtain the surface defect detection result.

[0035] Specifically, perform multiple high-frequency projection transformations on the curvature-stress distribution relationship matrix. Through high-frequency projection transformations, extract the hidden high-frequency features in the matrix to obtain multiple high-frequency feature maps. Construct a multi-scale feature pyramid based on the multiple high-frequency feature maps, and organically combine feature maps of different scales to form a hierarchical feature representation. Perform inverse transfer feature extraction on the hierarchical feature representation. Inverse transfer feature extraction is a method of extracting features adapted to the target domain from source domain features to obtain target domain adapted features. Perform dense connection operations based on the target domain adapted features to fuse features of different scales and levels to obtain a fused feature map. Perform self-attention mechanism analysis on the fused feature map, automatically weight important regions according to the information weights in the feature map to obtain weighted important region features. Based on the weighted important region features, construct a defect candidate region network. The defect candidate region network can initially identify potential defect regions to obtain potential defect regions. Align and refine the features of the potential defect regions. Through feature alignment, eliminate the deviation between features to make the defect representation more accurate. Calculate the defect confidence score based on the accurate defect representation. The defect confidence score reflects the reliability and possibility of the detected defect to obtain a defect probability map. Perform non-maximum suppression processing on the defect probability map to remove redundant detection results and retain the most representative defect regions to obtain defect bounding boxes. Classify the defect types based on the defect bounding boxes and the curvature-stress distribution relationship matrix. The classification process can determine the specific type of the detected defect, and finally obtain the surface defect detection result.

[0036] Perform connected component labeling on the defect probability map to identify possible defect regions. Then perform non-maximum suppression processing on the connected components to remove redundant detection results and ensure that the most prominent defect regions are retained to obtain defect bounding boxes. Perform regional cropping on the curvature-stress distribution relationship matrix according to the defect bounding boxes, and extract the regions within these bounding boxes to obtain defect feature sub-matrices. Transform the defect feature sub-matrices to convert the original matrix data into high-dimensional defect feature vectors. Analyze the high-dimensional defect feature vectors, and extract the target feature representation through feature analysis. Calculate the similarity with a preset defect template based on the target feature representation. By comparing the similarity between the target feature representation and the preset defect template, obtain the probability distribution of each defect type. Perform softmax function operation on the defect type probability distribution to normalize the probability distribution to obtain normalized defect type scores. These scores reflect the relative possibility of each defect type. Based on the normalized defect type scores, select the defect type with the highest probability as the final defect type, and combine it with the corresponding defect bounding box information to obtain the surface defect detection result.

[0037] In the embodiments of the present application, through comprehensive physical parameter measurement and three-dimensional laser scanning of the sample, rich initial data is obtained. By using Gaussian filtering and k-d tree index structure, multi-scale accurate description of the surface geometric features is achieved, effectively capturing the curvature changes at different scales. Combining the finite element method and nonlinear solution technology, accurate calculation of the internal stress state of the surface is realized. Through feature extraction and coding fusion of the curvature and stress data, a unique curvature-stress distribution relationship matrix is constructed, comprehensively reflecting the geometric and mechanical properties of the surface. By using a model based on multiple high-frequency projections and inverse transfer learning, combined with the self-attention mechanism, the accuracy and robustness of defect detection are greatly improved. Through non-maximum suppression and feature contrast analysis, accurate identification and positioning of the defect types are realized, and it can adapt to the curved flexible screens with different curvatures and stress distributions, thereby improving the detection accuracy of the curved flexible screens.

[0038] In a specific embodiment, the process of executing step S101 may specifically include the following steps:

[0039] (1) Measure the size of the curved flexible screen sample to obtain the length, width and thickness data of the sample, and calculate the density distribution data of the curved flexible screen sample according to the length, width and thickness data;

[0040] (2) Normalize the length, width and thickness data to obtain the sample size feature vector, and calculate the sample mass feature map according to the density distribution data;

[0041] (3) Generate the sample initial parameter dataset according to the sample size feature vector and the sample mass feature map;

[0042] (4) Generate multiple groups of candidate scanning parameters according to the sample initial parameter dataset, and perform quality evaluation according to the multiple groups of candidate scanning parameters to obtain the three-dimensional laser scanning parameters;

[0043] (5) Perform three-dimensional laser scanning on the curved flexible screen sample through the three-dimensional laser scanning parameters to obtain the sample three-dimensional point cloud data.

[0044] Specifically, a precision measuring device such as a laser rangefinder or a coordinate measuring machine is used to precisely measure each key position of the curved flexible screen, obtaining the length, width, and thickness data of the sample. For example, for a specific curved flexible screen sample, its length is 200 millimeters, width is 150 millimeters, and the average thickness at different positions is 0.5 millimeters. Based on these measurement data, the volume of the sample is calculated. The density of the curved flexible screen sample is calculated according to the measurement data to obtain density distribution data. The length, width, and thickness data are normalized to eliminate the influence between data of different scales, obtaining the size feature vector of the sample. The mass feature map of the sample is calculated based on the density distribution data, and the mass feature map can show the mass distribution of the sample at different positions. An initial parameter dataset of the sample is generated through the sample size feature vector and the sample mass feature map, and the initial parameter dataset of the sample includes the geometric and physical characteristics of the sample. Multiple groups of candidate scanning parameters are generated according to the initial parameter dataset of the sample. The selection of scanning parameters directly affects the accuracy and efficiency of 3D laser scanning. The candidate scanning parameters include scanning resolution, scanning angle, laser intensity, etc. For example, three groups of candidate scanning parameters can be generated: low resolution (0.1 millimeter), medium resolution (0.05 millimeter), and high resolution (0.01 millimeter), and each group of parameters corresponds to different scanning times and data accuracies. The quality of multiple groups of candidate scanning parameters is evaluated, and the evaluation criteria include scanning accuracy, data integrity, and scanning time, etc. Through comprehensive evaluation, the optimal 3D laser scanning parameters are selected. For example, it is found through evaluation that the parameters of medium resolution (0.05 millimeter) can ensure scanning accuracy while completing scanning within a reasonable time, so this parameter is selected as the final scanning parameter. The curved flexible screen sample is scanned three-dimensionally with the selected 3D laser scanning parameters. 3D laser scanning irradiates the surface of the sample with a laser beam and receives the reflected light through a sensor to record the point cloud data of the sample surface. The point cloud data consists of a large number of points and accurately describes the three-dimensional shape and surface characteristics of the sample. After scanning, the three-dimensional point cloud data of the sample is obtained, and these data are used for subsequent stress analysis, curvature calculation, and defect detection.

[0045] In a specific embodiment, the process of executing step S102 may specifically include the following steps:

[0046] (1) Gaussian filtering is performed on the three-dimensional point cloud data of the sample to obtain the denoised point cloud data, and a k-d tree index structure is constructed based on the denoised point cloud data to obtain a fast neighborhood search tree;

[0047] (2) Multiscale neighborhood radius settings are performed on the fast neighborhood search tree to obtain neighborhood point sets of different scales, and the principal curvature of each point is estimated based on the neighborhood point sets of different scales to obtain a point-level principal curvature vector;

[0048] (3) Construct a tensor field for the discrete principal curvature vectors to obtain a curvature tensor field, and calculate the Gaussian curvature and the mean curvature based on the curvature tensor field to obtain a multi-scale curvature feature map;

[0049] (4) Perform wavelet transform on the multi-scale curvature feature map to obtain a multi-resolution representation of the curvature, and extract feature points based on the multi-resolution representation of the curvature to obtain a set of curvature feature points;

[0050] (5) Conduct spatial clustering analysis on the set of curvature feature points to obtain regions with significant curvature changes, and perform interpolation fitting based on the regions with significant curvature changes and the multi-scale curvature feature map to obtain surface curvature distribution data.

[0051] Specifically, the three-dimensional point cloud data is obtained by laser scanning. These data usually contain noise and need to be processed by Gaussian filtering to remove the noise. Gaussian filtering is a smoothing process that reduces the impact of noise by performing weighted averaging on the neighborhood of each point. For a point cloud data set , where , the calculation formula for Gaussian filtering is:

[0052] ;

[0053] where, is the filtered point, is the point 's neighborhood, is the standard deviation of the Gaussian kernel. Through this formula, the denoised point cloud data is obtained. Construct a k-d tree index structure based on the denoised point cloud data. A k-d tree is a tree-shaped data structure used to organize k-dimensional space data and can perform efficient nearest neighbor searches. Through the k-d tree, the neighborhood points of each point can be quickly found to obtain a fast neighborhood search tree. Set multi-scale neighborhood radii for the fast neighborhood search tree. Different neighborhood radii represent different scales. By setting multi-scale neighborhood radii, sets of neighborhood points at different scales are obtained. Assume that at scale , the set of neighborhood points of point is , then the principal curvature of each point can be estimated based on the sets of neighborhood points at different scales. Principal curvature estimation describes the local geometric characteristics of each point by calculating the curvature of the set of neighborhood points of each point. For each point , its principal curvature can be calculated through the covariance matrix of the set of neighborhood points . The calculation formula for the covariance matrix is:

[0054] ;

[0055] where, is the point The centroid of the neighborhood point set, and the principal curvature vector can be obtained by performing eigenvalue decomposition on the covariance matrix Performing tensor field construction on the point-level principal curvature vectors to obtain a curvature tensor field. The tensor field is obtained by spatially interpolating the principal curvature vectors of each point, and the curvature tensor field can comprehensively describe the geometric characteristics of the surface. According to the curvature tensor field, the Gaussian curvature and the mean curvature are calculated. The Gaussian curvature and the mean curvature are important parameters for describing the shape of the surface. The Gaussian curvature and the mean curvature The calculation formulas are as follows:

[0056] ;

[0057] where, and are the principal curvature values. Through calculation, a multi-scale curvature feature map is obtained. The multi-scale curvature feature map can describe the curvature characteristics of the surface at different scales. Performing wavelet transform on the multi-scale curvature feature map to decompose the signal into different frequency components, and obtaining a multi-resolution representation of the curvature. The multi-resolution representation can describe the detailed characteristics of the surface at different resolutions. Through the multi-resolution representation, feature point extraction is performed. Feature points are points on the surface where the curvature changes significantly, and these points usually correspond to the key features or defects of the surface. Performing spatial clustering analysis on the curvature feature point set. Through clustering analysis, regions with significant curvature changes can be found, and these regions are usually the parts with the most significant surface characteristics. Spatial clustering analysis can use clustering algorithms such as DBSCAN or K-means to obtain regions with significant curvature changes through clustering analysis. According to the regions with significant curvature changes and the multi-scale curvature feature map, interpolation fitting is performed to convert the discrete feature point data into continuous curvature distribution data, and the surface curvature distribution data is obtained. Interpolation fitting can use interpolation algorithms such as Kriging interpolation or spline interpolation to obtain high-precision surface curvature distribution data.

[0058] In a specific embodiment, the process of executing step S103 may specifically include the following steps:

[0059] (1) Reconstructing the surface of the sample three-dimensional point cloud data to obtain a geometric surface model, and constructing a finite element mesh according to the geometric surface model to obtain a discretized computational domain;

[0060] (2) Assigning material properties to the discretized computational domain to obtain a node-level material parameter matrix, and constructing a local stiffness matrix according to the node-level material parameter matrix to obtain a global stiffness matrix;

[0061] (3) Applying boundary conditions to the global stiffness matrix to obtain a modified stiffness equation, and constructing a balance equation system according to the modified stiffness equation;

[0062] (4) Perform Newton - Raphson iterative solution on the equilibrium equations to obtain the nodal displacement field, and calculate the strain tensor based on the nodal displacement field to obtain a continuous strain distribution field;

[0063] (5) Perform constitutive equation conversion on the continuous strain distribution field to obtain the initial stress distribution field, and perform stress equilibrium correction and singular point treatment based on the initial stress distribution field to obtain the surface stress distribution data.

[0064] Specifically, perform surface reconstruction on the 3D point cloud data of the sample. For example, use the Poisson surface reconstruction algorithm or the Delaunay triangulation method to convert the discrete point cloud data into a continuous and smooth geometric surface model. Construct a finite - element mesh based on the geometric surface model. The finite - element mesh is a collection of finite elements obtained by discretizing the continuous geometric surface. Each element consists of several nodes, and these nodes form the computational domain. Common finite - element mesh construction methods include tetrahedral meshes and hexahedral meshes. Through these methods, convert the geometric surface model into a finite - element mesh to form a discretized computational domain. Assign material properties to the discretized computational domain. Each finite - element unit needs to be assigned corresponding material properties, such as elastic modulus and Poisson's ratio, etc. These properties determine the mechanical behavior of the material under external forces. Material property assignment can be achieved by establishing a nodal - level material parameter matrix. Construct the local stiffness matrix based on the nodal - level material parameter matrix. The local stiffness matrix describes the stiffness characteristics of each finite - element unit at the nodal level. The calculation formula for the local stiffness matrix is:

[0065] ;

[0066] where, is the local stiffness matrix, is the strain - displacement matrix, is the material property matrix, is the unit volume. By assembling the local stiffness matrices of all elements, the global stiffness matrix is obtained. The global stiffness matrix describes the stiffness characteristics of the entire computational domain. Apply boundary conditions to the global stiffness matrix, and apply actual physical constraint conditions such as fixed supports and external loads to the model to obtain the modified stiffness equation. The form of the modified stiffness equation is:

[0067] ;

[0068] where, is the nodal displacement vector, is the external load vector. According to the modified stiffness equation, an equilibrium equation set is constructed, which describes the equilibrium state of the system under external forces. The Newton-Raphson iteration method is used to solve the equilibrium equation set. The Newton-Raphson method is a method for solving nonlinear equations. Through iterative calculations, the node displacement values are continuously corrected until convergence is achieved. The result of the solution is the node displacement field , and the node displacement field describes the displacement changes of each node under external forces. According to the node displacement field, the strain tensor is calculated, and the strain tensor describes the strain state of the material during the deformation process. The calculation formula for the strain tensor is:

[0069] ;

[0070] where is the strain tensor, B is the strain-displacement matrix, is the node displacement vector. By calculating the strain tensor, a continuous strain distribution field is obtained, and the continuous strain distribution field describes the strain state of the entire calculation domain. The constitutive equation transformation is performed on the continuous strain distribution field. The constitutive equation is an equation in material mechanics that describes the relationship between stress and strain. Through the constitutive equation transformation, the strain distribution field is transformed into the initial stress distribution field. The calculation formula for the initial stress distribution field is:

[0071] ;

[0072] where is the stress tensor, is the material property matrix, is the strain tensor. Through calculation, the initial stress distribution field is obtained, and the initial stress distribution field describes the stress distribution of the material in the initial state. Stress equilibrium correction and singularity point treatment are performed based on the initial stress distribution field. Stress equilibrium correction is to ensure that the entire system reaches an equilibrium state under external forces, and singularity point treatment is to correct the abnormal points in the stress distribution to eliminate errors in numerical calculations and obtain the surface stress distribution data. Stress equilibrium correction and singularity point treatment can be achieved through numerical optimization methods such as the Lagrange multiplier method and the least squares method, and finally the surface stress distribution data is obtained.

[0073] In a specific embodiment, the process of executing step S104 may specifically include the following steps:

[0074] (1) Perform principal component analysis on the surface curvature distribution data to obtain the curvature principal eigenvector, and calculate the principal stress and equivalent stress according to the surface stress distribution data to obtain the stress characteristic field;

[0075] (2) Perform singular value decomposition on the stress characteristic field to obtain the stress principal eigenvector;

[0076] (3) Construct a joint feature space based on the principal curvature eigenvector and the principal stress eigenvector to obtain an initial feature matrix;

[0077] (4) Perform non-linear dimensionality reduction on the initial feature matrix to obtain a low-dimensional feature representation, and construct an autoencoder neural network based on the low-dimensional feature representation. The autoencoder neural network includes a feature encoder and a decoder;

[0078] (5) Perform sparse constraint optimization on the hidden layer output of the feature encoder to obtain a sparse feature encoding, and calculate the curvature-stress correlation coefficient based on the sparse feature encoding to obtain a correlation strength matrix;

[0079] (6) Perform spectral clustering analysis on the correlation strength matrix to obtain a curvature-stress coupling pattern;

[0080] (7) Perform a tensor product operation based on the curvature-stress coupling pattern and the sparse feature encoding to obtain a curvature-stress distribution relationship matrix.

[0081] Specifically, perform principal component analysis on the surface curvature distribution data. Principal component analysis is a dimensionality reduction technique that extracts the main eigenvectors by performing a linear transformation on the data, reducing the dimensionality of the data while retaining the main information. For the surface curvature distribution data , the calculation formula for principal component analysis is:

[0082] ;

[0083] where is the covariance matrix, is the curvature distribution data matrix, is the number of samples. By performing eigenvalue decomposition on the covariance matrix , the principal eigenvectors are obtained, and these vectors represent the main change directions of the curvature data. Calculate the principal stress and equivalent stress based on the surface stress distribution data. The stress distribution data is usually obtained through finite element analysis and contains the stress information of the surface at different positions. The principal stress is calculated as follows:

[0084] ;

[0085] where and are the plane stresses, is the shear stress. The equivalent stress is calculated as follows:

[0086] ;

[0087] Through these formulas, the stress feature field is obtained, and the stress feature field describes the stress state of the surface at different positions. Perform singular value decomposition on the stress feature field. Singular value decomposition is a matrix decomposition technique that can decompose a matrix into the product of three matrices, expressed as:

[0088] ;

[0089] where, is the stress feature field matrix, is the left singular vector matrix, is the singular value matrix, is the right singular vector matrix. Through singular value decomposition, the principal stress eigenvectors are obtained, and these vectors represent the main change directions of the stress data. Construct a joint feature space based on the principal curvature eigenvectors and the principal stress eigenvectors. The joint feature space combines the principal feature vectors of curvature and stress to form a new feature space. Perform non-linear dimensionality reduction on the initial feature matrix. Non-linear dimensionality reduction is a method of mapping high-dimensional data to a low-dimensional space and is achieved through an autoencoder neural network. The autoencoder neural network includes a feature encoder and a decoder. The feature encoder encodes the high-dimensional data into a low-dimensional representation, and the decoder restores the low-dimensional representation to high-dimensional data. Through training the autoencoder neural network, a low-dimensional feature representation is obtained. Optimize the hidden layer output of the feature encoder with a sparsity constraint. Sparsity constraint optimization is a technique that makes the feature representation more sparse and can be achieved by adding a sparsity regularization term. The objective function of the optimization is expressed as:

[0090] ;

[0091] where, is the input data, is the dictionary matrix, is the sparse coding, is the regularization parameter. Through optimization, sparse feature coding is obtained. Calculate the curvature-stress correlation coefficient based on the sparse feature coding. The correlation coefficient quantifies the relationship between curvature and stress and is expressed as:

[0092] ;

[0093] where, is the curvature feature, is the stress feature, and is its mean value. Through calculation, a correlation strength matrix is obtained. Spectral clustering analysis is performed on the correlation strength matrix. Spectral clustering is a clustering method based on graph theory, and clustering analysis is carried out through the eigenvectors of the Laplacian matrix to obtain the curvature-stress coupling mode. According to the curvature-stress coupling mode and sparse feature coding, a tensor product operation is performed. The tensor product operation is a method for fusing multi-dimensional data, which is expressed as:

[0094] ;

[0095] where, is the curvature-stress distribution relationship matrix, is the joint feature matrix, is the coupling mode. Through the tensor product operation, the final curvature-stress distribution relationship matrix is obtained.

[0096] In a specific embodiment, the process of executing step S105 may specifically include the following steps:

[0097] (1) Perform multiple high-frequency projection transformations on the curvature-stress distribution relationship matrix to obtain multiple high-frequency feature maps, and construct a multi-scale feature pyramid based on the multiple high-frequency feature maps to obtain a hierarchical feature representation;

[0098] (2) Perform inverse transfer feature extraction on the hierarchical feature representation to obtain target domain adaptation features, and perform a dense connection operation based on the target domain adaptation features to obtain a fused feature map;

[0099] (3) Perform self-attention mechanism analysis on the fused feature map to obtain weighted features of important regions, and construct a defect candidate region network based on the weighted features of important regions to obtain potential defect regions;

[0100] (4) Align and refine the features of the potential defect regions to obtain accurate defect representations, and calculate defect confidence scores based on the accurate defect representations to obtain a defect probability map;

[0101] (5) Perform non-maximum suppression processing on the defect probability map to obtain defect bounding boxes, and classify defect types based on the defect bounding boxes and the curvature-stress distribution relationship matrix to obtain surface defect detection results.

[0102] Specifically, perform multiple high-frequency projection transformations on the curvature-stress distribution relationship matrix. The high-frequency projection transformation is a signal processing technique that extracts high-frequency components in the matrix through a filter. For example, the discrete Fourier transform (DFT) is used for high-frequency projection, and the transformation formula is:

[0103] ;

[0104] where, is the transformed high-frequency feature, is the data in the original relationship matrix, is the number of data points, is the frequency component. By performing inverse transforms on different frequency components, multiple high-frequency feature maps are obtained. Based on the high-frequency feature maps, a multi-scale feature pyramid is constructed. A multi-scale feature pyramid is an image processing technique that extracts features at different levels by processing feature maps at different scales. By performing downsampling and filtering operations on the high-frequency feature maps, a multi-scale feature pyramid is constructed to obtain a hierarchical feature representation. Inverse transfer feature extraction is performed on the hierarchical feature representation. Inverse transfer feature extraction is a domain adaptation technique that maps source domain features to the target domain to obtain features adapted to the target domain. For example, a deep convolutional neural network is used for feature extraction. By training the deep convolutional neural network model, the hierarchical feature representation can be adapted to the target domain to obtain target domain adapted features. Suppose a pre-trained ResNet model is used. By performing convolution and pooling operations on the hierarchical features, target domain adapted features are obtained. Dense connection operations are performed based on the target domain adapted features. Dense connection operations are a method of fusing features at different levels. By connecting feature maps at different levels, a fused feature map can be obtained. For example, a dense convolutional network is used for feature fusion. By performing convolution operations at each level and connecting the output with the input, a fused feature map is obtained. Self-attention mechanism analysis is performed on the fused feature map. The self-attention mechanism is a method of improving the expressive power of feature maps. By calculating the importance weights of each position in the feature map, important regions are highlighted. For example, a self-attention mechanism module is used. By calculating the weights of the fused feature map, important region weighted features are obtained. A defect candidate region network is constructed based on the important region weighted features. A defect candidate region network is a neural network used to preliminarily identify potential defect regions. By performing convolution and pooling operations on the important region weighted features, potential defect regions are obtained. Feature alignment and refinement are performed on the potential defect regions. Feature alignment is a method of aligning different features. By adjusting the position and scale of the features, the features become more consistent. Refinement processing further improves the accuracy of the features. By alignment and refinement, an accurate defect representation is obtained. The defect confidence score is calculated based on the accurate defect representation. The defect confidence score is a measure reflecting the reliability of the defect detection result. By classifying the accurate defect representation, a defect probability map is obtained. Non-maximum suppression processing is performed on the defect probability map. Non-maximum suppression is a method of removing redundant detection results. By retaining the detection result with the highest probability, a defect bounding box is obtained. For example, threshold processing is performed on the defect probability map and non-maximum suppression is performed on each detection result to obtain a defect bounding box.

[0105] In a specific embodiment, the process of performing non-maximum suppression on the defect probability map to obtain defect bounding boxes and classifying defect types according to the defect bounding boxes and the curvature-stress distribution relationship matrix to obtain the surface defect detection result may specifically include the following steps:

[0106] (1) Perform connected component labeling and non-maximum suppression on the defect probability map to obtain defect bounding boxes, and perform regional cropping on the curvature-stress distribution relationship matrix according to the defect bounding boxes to obtain defect feature sub-matrices;

[0107] (2) Transform the defect feature sub-matrices to obtain high-dimensional defect feature vectors, and analyze the high-dimensional defect feature vectors to obtain target feature representations;

[0108] (3) Calculate the similarity with a preset defect template according to the target feature representation to obtain a defect type probability distribution, and perform a softmax function operation on the defect type probability distribution to obtain a normalized defect type score;

[0109] (4) Select the type with the highest probability according to the normalized defect type score, and combine it with the corresponding defect bounding box information to obtain the surface defect detection result.

[0110] Specifically, perform connected component labeling on the defect probability map to identify possible defect regions. Connected component labeling is achieved by traversing the pixels in the image to identify adjacent pixel blocks with the same label and marking them as the same region. Perform non-maximum suppression on the connected components. Non-maximum suppression is a method for removing redundant detection results. By retaining the detection result with the highest probability, a more accurate defect bounding box is obtained. For each connected component, find its maximum probability value and retain that region while removing other low-probability regions. The bounding box can precisely enclose each detected defect region. According to the defect bounding box, perform regional cropping on the curvature-stress distribution relationship matrix. The curvature-stress distribution relationship matrix contains curvature and stress information at different positions on the surface. By cropping these matrices, the characteristic data corresponding to the defect region is extracted to obtain a defect characteristic submatrix. Transform the defect characteristic submatrix to obtain a high-dimensional defect feature vector. Feature transformation is a method for converting matrix data into vector representation, and principal component analysis or convolutional neural network can be used for transformation. Analyze the high-dimensional defect feature vector to obtain the target feature representation. Feature analysis can be performed through convolutional neural network or deep learning model. By further processing and analyzing the high-dimensional defect feature vector, the target feature representation is extracted. Calculate the similarity with the preset defect template based on the target feature representation to obtain the defect type probability distribution. Similarity calculation can use cosine similarity or Euclidean distance. By comparing the similarity between the target feature representation and the preset defect template, the probability of each defect type is obtained. Assume there is a preset defect template , the similarity score can be calculated by cosine similarity as follows:

[0111] ;

[0112] where is the similarity score, z is the target feature representation, is the preset defect template, represents the norm of the vector. By calculating the similarity score, the defect type probability distribution is obtained. Perform the softmax function operation on the defect type probability distribution to obtain the normalized defect type score. The softmax function is a method for converting multiple similarity scores into a probability distribution, and the formula is:

[0113] ;

[0114] where is the probability that the defect type is , It is the similarity score. Through the operation of the softmax function, the normalized defect type scores are obtained. According to the normalized defect type scores, the type with the highest probability is selected and combined with the corresponding defect bounding box information to obtain the curved surface defect detection result.

[0115] The detection method of the curved surface flexible screen in the embodiment of the present application is described above. Next, the detection device of the curved surface flexible screen in the embodiment of the present application will be described. Please refer to Figure 2 , an embodiment of the detection device of the curved surface flexible screen in the embodiment of the present application includes:

[0116] A scanning module 201, configured to perform physical parameter measurement on a curved surface flexible screen sample to obtain an initial parameter dataset of the sample, and perform three-dimensional laser scanning on the curved surface flexible screen sample based on the initial parameter dataset of the sample to obtain sample three-dimensional point cloud data;

[0117] A calculation module 202, configured to perform multi-scale curvature calculation on the sample three-dimensional point cloud data to obtain curved surface curvature distribution data;

[0118] An analysis module 203, configured to perform stress analysis and equilibrium equation solution on the sample three-dimensional point cloud data to obtain curved surface stress distribution data;

[0119] A fusion module 204, configured to perform feature extraction and feature vector encoding fusion on the curved surface curvature distribution data and the curved surface stress distribution data to obtain a curvature-stress distribution relationship matrix;

[0120] A detection module 205, configured to input the curvature-stress distribution relationship matrix into a preset curved surface defect detection model for curved surface defect detection to obtain a curved surface defect detection result.

[0121] Through the collaborative cooperation of the above-mentioned various components, by comprehensively measuring the physical parameters of the sample and performing three-dimensional laser scanning, rich initial data is obtained. Using Gaussian filtering and k-d tree index structure, multi-scale accurate description of the curved surface geometric features is realized, and the curvature changes at different scales are effectively captured. Combining the finite element method and nonlinear solution technology, accurate calculation of the internal stress state of the curved surface is realized. By performing feature extraction and coding fusion on the curvature and stress data, a unique curvature-stress distribution relationship matrix is constructed, which comprehensively reflects the geometric and mechanical characteristics of the curved surface. Using a model based on multiple high-frequency projections and inverse transfer learning, combined with self-attention mechanism, the accuracy and robustness of defect detection are greatly improved. Through non-maximum suppression and feature contrast analysis, accurate identification and positioning of defect types are realized, and it can adapt to curved surface flexible screens with different curvatures and stress distributions, thereby improving the detection accuracy of curved surface flexible screens.

[0122] The present application also provides a detection device for a curved flexible screen. The detection device for the curved flexible screen includes a memory and a processor. Computer-readable instructions are stored in the memory. When the computer-readable instructions are executed by the processor, the processor is caused to execute the steps of the detection method for the curved flexible screen in the above respective embodiments.

[0123] The present application also provides a computer-readable storage medium. The computer-readable storage medium can be a non-volatile computer-readable storage medium, or can also be a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions run on a computer, the computer is caused to execute the steps of the detection method for the curved flexible screen.

[0124] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the above-described system, system, and unit can refer to the corresponding processes in the foregoing method embodiments, and will not be described herein again.

[0125] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods in the various embodiments of the present application. The foregoing storage medium includes: various media such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disc that can store program codes.

[0126] As described above, the above embodiments are only used to illustrate the technical solution of the present application, and are not intended to limit it; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the various embodiments of the present application.

Claims

1. A detection method for a curved flexible screen, characterized in that, The detection method of the curved flexible screen includes: Performing physical parameter measurement on the curved flexible screen sample to obtain an initial parameter dataset of the sample, and performing three-dimensional laser scanning on the curved flexible screen sample based on the initial parameter dataset of the sample to obtain sample three-dimensional point cloud data; Performing multi-scale curvature calculation on the sample three-dimensional point cloud data to obtain curved surface curvature distribution data; Performing stress analysis and solving equilibrium equations on the sample three-dimensional point cloud data to obtain curved surface stress distribution data; Performing feature extraction and feature vector encoding fusion on the curved surface curvature distribution data and the curved surface stress distribution data to obtain a curvature-stress distribution relationship matrix; specifically including: performing principal component analysis on the curved surface curvature distribution data to obtain a curvature principal eigenvector, calculating principal stress and equivalent stress according to the curved surface stress distribution data to obtain a stress feature field; performing singular value decomposition on the stress feature field to obtain a stress principal eigenvector; constructing a joint feature space according to the curvature principal eigenvector and the stress principal eigenvector to obtain an initial feature matrix; performing non-linear dimensionality reduction processing on the initial feature matrix to obtain a low-dimensional feature representation, and constructing an autoencoder neural network according to the low-dimensional feature representation, the autoencoder neural network including a feature encoder and a decoder; performing sparse constraint optimization on the hidden layer output of the feature encoder to obtain a sparse feature code, and calculating a curvature-stress correlation coefficient according to the sparse feature code to obtain a correlation strength matrix; performing spectral clustering analysis on the correlation strength matrix to obtain a curvature-stress coupling mode; performing tensor product operation according to the curvature-stress coupling mode and the sparse feature code to obtain a curvature-stress distribution relationship matrix; Inputting the curvature-stress distribution relationship matrix into a preset curved surface defect detection model for curved surface defect detection to obtain a curved surface defect detection result.

2. The detection method of the curved flexible screen according to claim 1, wherein, The performing physical parameter measurement on the curved flexible screen sample to obtain an initial parameter dataset of the sample, and performing three-dimensional laser scanning on the curved flexible screen sample based on the initial parameter dataset of the sample to obtain sample three-dimensional point cloud data includes: Performing size measurement on the curved flexible screen sample to obtain the length, width and thickness data of the sample, and calculating the density distribution data of the curved flexible screen sample according to the length, width and thickness data; Performing normalization processing on the length, width and thickness data to obtain a sample size feature vector, and calculating a sample mass feature map according to the density distribution data; Generating an initial parameter dataset of the sample according to the sample size feature vector and the sample mass feature map; Generating multiple groups of candidate scanning parameters according to the initial parameter dataset of the sample, and performing quality evaluation according to the multiple groups of candidate scanning parameters to obtain three-dimensional laser scanning parameters; Performing three-dimensional laser scanning on the curved flexible screen sample through the three-dimensional laser scanning parameters to obtain sample three-dimensional point cloud data.

3. The detection method of the curved flexible screen according to claim 1, wherein, The performing multi-scale curvature calculation on the sample three-dimensional point cloud data to obtain curved surface curvature distribution data includes: Perform Gaussian filtering on the three-dimensional point cloud data of the sample to obtain the denoised point cloud data, and construct a k-d tree index structure based on the denoised point cloud data to obtain a fast neighborhood search tree; Set multi-scale neighborhood radii for the fast neighborhood search tree to obtain neighborhood point sets at different scales, and estimate the principal curvature for each point based on the neighborhood point sets at different scales to obtain a point-level principal curvature vector; Construct a tensor field for the point-level principal curvature vector to obtain a curvature tensor field, and calculate the Gaussian curvature and mean curvature based on the curvature tensor field to obtain a multi-scale curvature feature map; Perform wavelet transform on the multi-scale curvature feature map to obtain a multi-resolution representation of the curvature, and extract feature points based on the multi-resolution representation of the curvature to obtain a curvature feature point set; Perform spatial clustering analysis on the curvature feature point set to obtain regions with significant curvature changes, and perform interpolation fitting based on the regions with significant curvature changes and the multi-scale curvature feature map to obtain surface curvature distribution data.

4. The detection method of the curved flexible screen according to claim 1, characterized in that The stress analysis and balance equation solution for the three-dimensional point cloud data of the sample to obtain surface stress distribution data include: Perform surface reconstruction on the three-dimensional point cloud data of the sample to obtain a geometric surface model, and construct a finite element mesh based on the geometric surface model to obtain a discretized computational domain; Assign material properties to the discretized computational domain to obtain a node-level material parameter matrix, and construct a local stiffness matrix based on the node-level material parameter matrix to obtain a global stiffness matrix; Apply boundary conditions to the global stiffness matrix to obtain a modified stiffness equation, and construct a balance equation system based on the modified stiffness equation; Perform Newton-Raphson iterative solution on the balance equation system to obtain a node displacement field, and calculate the strain tensor based on the node displacement field to obtain a continuous strain distribution field; Perform constitutive equation conversion on the continuous strain distribution field to obtain an initial stress distribution field, and perform stress balance correction and singular point processing based on the initial stress distribution field to obtain surface stress distribution data.

5. The detection method of the curved flexible screen according to claim 1, wherein The input of the curvature-stress distribution relationship matrix into a pre-set surface defect detection model for surface defect detection to obtain surface defect detection results includes: Perform multiple high-frequency projection transformations on the curvature-stress distribution relationship matrix to obtain multiple high-frequency feature maps, and construct a multi-scale feature pyramid based on the multiple high-frequency feature maps to obtain a hierarchical feature representation; Extract inverse transfer features from the hierarchical feature representation to obtain target domain adaptation features, and perform dense connection operations based on the target domain adaptation features to obtain a fused feature map; Perform self-attention mechanism analysis on the fused feature map to obtain weighted features of important regions, and construct a defect candidate region network based on the weighted features of important regions to obtain potential defect regions; Align and refine the features of the potential defect regions to obtain accurate defect representations, and calculate defect confidence scores based on the accurate defect representations to obtain a defect probability map; Perform non-maximum suppression on the defect probability map to obtain defect bounding boxes, and classify the defect types based on the defect bounding boxes and the curvature-stress distribution relationship matrix to obtain the curved surface defect detection result.

6. The detection method of the curved flexible screen according to claim 5, wherein The performing non-maximum suppression on the defect probability map to obtain defect bounding boxes, and classifying the defect types based on the defect bounding boxes and the curvature-stress distribution relationship matrix to obtain the curved surface defect detection result includes: Perform connected region labeling and non-maximum suppression on the defect probability map to obtain defect bounding boxes, and perform region cropping on the curvature-stress distribution relationship matrix according to the defect bounding boxes to obtain a defect feature sub-matrix; Transform the defect feature sub-matrix to obtain a high-dimensional defect feature vector, and analyze the high-dimensional defect feature vector to obtain a target feature representation; Calculate the similarity with a preset defect template based on the target feature representation to obtain a defect type probability distribution, and perform a softmax function operation on the defect type probability distribution to obtain a normalized defect type score; Select the type with the highest probability according to the normalized defect type score, and combine it with the corresponding defect bounding box information to obtain the curved surface defect detection result.

7. A detection device for a curved flexible screen, characterized in that, For implementing the detection method of the curved surface flexible screen according to any one of claims 1-6, the detection device for the curved surface flexible screen includes: A scanning module, configured to measure physical parameters of a curved surface flexible screen sample to obtain an initial parameter data set of the sample, and perform three-dimensional laser scanning on the curved surface flexible screen sample based on the initial parameter data set of the sample to obtain sample three-dimensional point cloud data; A calculation module, configured to perform multi-scale curvature calculation on the sample three-dimensional point cloud data to obtain curved surface curvature distribution data; An analysis module, configured to perform stress analysis and balance equation solving on the sample three-dimensional point cloud data to obtain curved surface stress distribution data; A fusion module, configured to perform feature extraction and feature vector encoding fusion on the curved surface curvature distribution data and the curved surface stress distribution data to obtain a curvature-stress distribution relationship matrix; A detection module, configured to input the curvature-stress distribution relationship matrix into a preset curved surface defect detection model for curved surface defect detection to obtain a curved surface defect detection result.

8. A detection device for a curved flexible screen, characterized in that, The detection device for the curved surface flexible screen includes: a memory and at least one processor, and instructions are stored in the memory; The at least one processor calls the instructions in the memory so that the detection device for the curved surface flexible screen executes the detection method of the curved surface flexible screen according to any one of claims 1-6.

9. A computer-readable storage medium having instructions stored thereon, characterized in that, When the instructions are executed by the processor, the detection method of the curved surface flexible screen according to any one of claims 1-6 is implemented.

Images