Glass scanning detection method, device and equipment

Abstract

The present invention relates to the field of glass detection technology, and discloses a method, device and equipment for detecting glass after scanning. The method includes: collecting multi-source original sensor data of the glass to be detected, and performing signal denoising, synchronous transformation and multi-scale glass feature extraction on the multi-source original sensor data, and performing feature combination and feature dimension reduction compression, region segmentation and regional contour reconstruction on the multi-scale glass features to obtain an initial defect region set, and performing spatiotemporal feature combination of multi-source features on the initial defect region set to obtain an enhanced defect feature set; constructing a regional similarity matrix corresponding to the enhanced defect feature set, and determining the defect category and scoring the multi-dimensional glass quality on the regional similarity matrix to obtain glass quality evaluation data, and performing process association on the glass quality evaluation data to generate a final inspection report for the glass to be detected. The present application improves the accuracy of the results of quality inspection of glass after scanning.

Description

Technical Field

[0001] The present invention relates to the technical field of glass detection, and in particular to a method, device and equipment for detecting glass after scanning. Background Art

[0002] In the field of glass manufacturing, the inspection of glass surface quality after scanning is an important part of product development and quality control, and the real-time monitoring and feature recognition of surface defects is the most critical and challenging stage in quality inspection. Accurate detection of glass surface features such as scratches, pits and light transmittance is crucial for manufacturers, R&D institutions and testing centers, as it directly affects product quality evaluation, performance optimization and overall use value. Therefore, an effective inspection method to monitor these surface features is essential to ensure the quality stability and appearance quality of glass products.

[0003] Nowadays, machine vision technology is used to analyze collected data, build defect models to evaluate surface quality, or try to apply artificial intelligence technology to the defect recognition process to better evaluate defect types and distribution trends. However, these methods still face challenges in integrating multi-dimensional surface features, handling nonlinear defects, and adapting to dynamic lighting. In addition, these detection methods often ignore some unique features of the glass surface, such as texture changes, height fluctuations, and uneven light transmission, which may have a significant impact on defect judgment and positioning. That is, the existing glass scanning detection method ignores multi-dimensional surface features and some unique features of glass, resulting in low accuracy of the final quality detection results. Summary of the invention

[0004] The main purpose of the present invention is to solve the problem that the existing inspection method after glass scanning ignores the multi-dimensional surface characteristics and some unique characteristics of glass, resulting in low accuracy of the final quality inspection result.

[0005] A first aspect of the present invention provides a method for detecting glass after scanning, the method comprising: using a preset multi-sensor array to collect multi-source original sensor data of the glass to be detected, and performing signal denoising and synchronous transformation on the multi-source original sensor data to obtain pre-processed multi-source original sensor data; extracting multi-scale glass features corresponding to the glass to be detected in the pre-processed multi-source original sensor data, and performing feature combination and feature dimension reduction compression on the multi-scale glass features to obtain an optimized feature matrix; based on the optimized feature matrix, performing regional segmentation and regional contour reconstruction on the glass to be detected to obtain an initial defect region set, and performing spatiotemporal feature combination of multi-source features on the initial defect region set to obtain an enhanced defect feature set; constructing a regional similarity matrix corresponding to the enhanced defect feature set, and determining the defect category of the regional similarity matrix to generate a classification label set; performing multi-dimensional glass quality scoring on the classification label set to obtain glass quality evaluation data, and performing process association on the glass quality evaluation data to generate a final inspection report of the glass to be detected.

[0006] Optionally, in a first implementation method of the first aspect of the present invention, the signal denoising and synchronous transformation of the multi-source original sensor data to obtain preprocessed multi-source original sensor data includes: based on a preset glass acquisition speed, window segmenting the multi-source original sensor data to obtain a multi-source original sensor sub-sequence, and respectively extracting the size coefficient of each of the multi-source original sensor sub-sequences to obtain a multi-scale segmentation coefficient; performing threshold screening of the multi-scale segmentation coefficients corresponding to high-frequency interference components of glass detection to obtain a multi-scale noise reduction coefficient, and reconstructing coefficient data of the multi-scale noise reduction coefficient to obtain filtered multi-source original sensor data; based on the corresponding sampling timestamps of the multi-sensor array, time aligning the filtered multi-source original sensor data and normalizing the data stream to obtain preprocessed multi-source original sensor data.

[0007] Optionally, in a second implementation of the first aspect of the present invention, the multi-scale glass features corresponding to the glass to be detected in the pre-processed multi-source original sensor data are extracted, and the multi-scale glass features are subjected to feature combination and feature dimensionality reduction compression to obtain an optimized feature matrix, including: based on a preset glass partitioning block unit, the pre-processed multi-source original sensor data is subjected to regional segmentation of local blocks to obtain a regional segmentation matrix, and grayscale statistics are performed on the image sensor data in the regional segmentation matrix to obtain a glass texture feature matrix, and gradient operations are performed on the glass surface height data in the regional segmentation matrix to obtain a morphology feature matrix, and optical distribution features are extracted from the optical data in the regional segmentation matrix to obtain an optical feature matrix; The glass texture feature matrix, the morphology feature matrix and the optical feature matrix are normalized to obtain a normalized feature matrix, and a tensor superposition operation is performed on the normalized feature matrix to obtain a glass feature tensor; the glass feature tensor is subjected to singular value decomposition of the corresponding matrix dimension to obtain a tensor decomposition result, and the eigenvectors whose eigenvalues ​​on each dimension in the tensor decomposition result are greater than a preset feature threshold are extracted to obtain a modal feature vector; the variance contribution rate corresponding to each eigenvector in the modal feature vector is calculated, and the feature combination corresponding to the variance contribution rate greater than a preset contribution threshold is selected to obtain important features of glass defects; the important features of glass defects are subjected to matrix reconstruction corresponding to the glass defect features and an outer product operation of a two-dimensional structure to obtain an optimized feature matrix.

[0008] Optionally, in a third implementation of the first aspect of the present invention, the glass to be inspected is segmented and its regional contours are reconstructed based on the optimized feature matrix to obtain an initial defect region set, including: performing spatial density transformation on the defect region of the optimized feature matrix to obtain a density distribution map, and marking the density region in the density distribution map where the defect density is greater than a preset dense threshold to obtain an initial marked region; clustering and regional expansion growth of pixels with similar features on the initial marked region to obtain an initial segmented region, and extracting boundary features of the grayscale gradient and height difference of the glass surface on the initial segmented region to obtain a regional boundary contour; performing morphological corrosion and expansion and noise debris elimination on the regional boundary contour to obtain a smooth boundary region, and performing area threshold filtering on the smooth boundary region and merging it with adjacent regions to obtain a merged detection region; performing abnormality calculation and detection of multiple detection features on the merged detection region to obtain an abnormal detection region, and performing defect number marking and feature attribute association on the abnormal detection region to obtain an initial defect region set.

[0009] Optionally, in a fourth implementation of the first aspect of the present invention, the temporal and spatial feature combination of multi-source features of the initial defect area set to obtain an enhanced defect feature set includes: extracting the time series features of the initial defect area set corresponding to the multi-image acquisition moments to obtain a defect feature time series matrix, extracting feature change trends and marking key point sampling on the defect feature time series matrix to obtain key time series features; performing numerical sequence transformation of preset glass surface characteristics and defect state quantification on the key time series features to obtain a state change matrix, extracting and combining features of multi-dimensional change characteristics of the glass surface on the state change matrix to obtain a combined feature sequence; calculating defect feature significance and assigning importance to the combined feature sequence to obtain a weighted feature set, integrating multi-source features and enhancing glass defect features on the weighted feature set to obtain an enhanced feature vector; reorganizing defect position features and overall surface features on the enhanced feature vector to obtain a feature structure set, performing correlation analysis and combined integration of scanning defects on the feature structure set to obtain an enhanced defect feature set.

[0010] Optionally, in a fifth implementation of the first aspect of the present invention, a regional similarity matrix corresponding to the enhanced defect feature set is constructed, and defect categories are determined on the regional similarity matrix to generate a classification label set, including: performing numerical difference calculation and feature mapping on each glass defect feature in the enhanced defect feature set to obtain a multidimensional difference matrix, and performing defect type similarity quantification and threshold division on the multidimensional difference matrix to obtain a defect similarity matrix; performing correlation statistics and association path construction on the defect similarity matrix to obtain a defect association network, and performing type determination and classification on the glass defect features in the defect association network to obtain a preliminary classification result; performing statistics and category confirmation on the glass defect features on the preliminary classification result to obtain a classification statistics table, and performing defect classification label generation and attribute association on the classification statistics table to obtain a classification label set.

[0011] Optionally, in a sixth implementation method of the first aspect of the present invention, the multi-dimensional glass quality scoring of the classification label set to obtain glass quality evaluation data includes: performing feature statistics and data extraction on the corresponding surface quality parameters in the classification label set to obtain a quality feature table, and performing numerical grading and range division on each glass defect feature data in the quality feature table to obtain a grading standard table; based on the grading standard table, performing grade assessment and parameter grading on the quality feature table to obtain a grading result matrix, and performing weight assignment and coefficient calculation on the grading results corresponding to the surface quality parameters in the grading standard table to obtain a weight coefficient matrix, and performing weighted calculation and score summation on the grading result matrix and the weight coefficient matrix to obtain a comprehensive score value; dividing the glass surface area and performing defect distribution statistics on the comprehensive score value to obtain a regional score distribution map, and performing regional score integration and quality indicator generation on the regional score distribution map to obtain glass quality evaluation data.

[0012] Optionally, in a seventh implementation method of the first aspect of the present invention, the glass quality evaluation data is process-associated to generate a final inspection report of the glass to be inspected, including: extracting secondary distribution features and classifying defect features of corresponding surface quality parameters in the glass quality evaluation data to obtain a defect feature table, and associating and statistically analyzing the relationship between defect types in the defect feature table and preset scanning process parameters to obtain a process association table; performing data comparison and parameter influence calculation on the process association table to obtain an influence degree matrix, and ranking the influence degree matrix by importance and screening key defect parameters to obtain a key parameter table; integrating the key parameter table and each of the glass defect features and summarizing the inspection results to generate a final inspection report of the glass to be inspected.

[0013] The second aspect of the present invention provides a glass scanning detection device, the glass scanning detection device comprising: a preprocessing module, used to collect multi-source original sensor data of the glass to be detected by using a preset multi-sensor array, and perform signal denoising and synchronous transformation on the multi-source original sensor data to obtain preprocessed multi-source original sensor data; a feature optimization module, used to extract multi-scale glass features corresponding to the glass to be detected in the preprocessed multi-source original sensor data, and perform feature combination and feature dimensionality reduction compression on the multi-scale glass features to obtain an optimized feature matrix; a feature combination module, used to extract multi-scale glass features corresponding to the glass to be detected in the preprocessed multi-source original sensor data, and perform feature combination and feature dimensionality reduction compression on the multi-scale glass features to obtain an optimized feature matrix; and a feature combination module, used to extract multi-scale glass features corresponding to the glass to be detected in the preprocessed multi-source original sensor data. , performing regional segmentation and regional contour reconstruction on the glass to be inspected to obtain an initial defect region set, and performing spatiotemporal feature combination of multi-source features on the initial defect region set to obtain an enhanced defect feature set; a category determination module, used to construct a regional similarity matrix corresponding to the enhanced defect feature set, and determine the defect category of the regional similarity matrix to generate a classification label set; a report generation module, used to perform multi-dimensional glass quality scoring on the classification label set to obtain glass quality evaluation data, and perform process association on the glass quality evaluation data to generate a final inspection report for the glass to be inspected.

[0014] The third aspect of the present invention provides a glass post-scanning inspection device, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory to enable the glass post-scanning inspection device to perform each step of the above-mentioned glass post-scanning inspection method.

[0015] The above-mentioned detection method, device and equipment after scanning the glass. In the embodiment of the present invention, the multi-source original sensor data of the glass to be detected is collected by using a preset multi-sensor array, and the multi-source original sensor data is subjected to signal denoising and synchronous transformation to obtain the pre-processed multi-source original sensor data; the multi-scale glass features corresponding to the glass to be detected in the pre-processed multi-source original sensor data are extracted, and the multi-scale glass features are subjected to feature combination and feature dimension reduction compression to obtain an optimized feature matrix; based on the optimized feature matrix, the glass to be detected is subjected to regional segmentation and regional contour reconstruction to obtain an initial defect region set, and the initial defect region set is subjected to spatiotemporal feature combination of multi-source features to obtain an enhanced defect feature set; a regional similarity matrix corresponding to the enhanced defect feature set is constructed, and the defect category of the regional similarity matrix is ​​determined to generate a classification label set; the classification label set is subjected to multi-dimensional glass quality scoring to obtain glass quality evaluation data, and the glass quality evaluation data is subjected to process correlation to generate a final inspection report of the glass to be detected. Compared with the prior art, this application obtains basic surface features through multi-sensor data collection and feature extraction on the glass surface, and then combines these features for dimensionality reduction and regional segmentation to obtain defect area features; then obtains defect enhancement features and performs similarity analysis through multi-source feature fusion and spatiotemporal feature enhancement; finally, quality scoring and process association are performed based on category judgment, and a test report is output. Through hierarchical data processing and feature analysis, the problem of accurate identification of defects after glass scanning is solved, especially in terms of surface scratches, pits and transmittance, the multi-dimensional characteristics and defect characteristics of the glass surface are fully considered, and the detection accuracy is effectively improved; and the multi-level feature extraction and defect mapping strategy are adopted, which not only realizes the correlation analysis between surface parameters, but also enhances the reliability of defect identification; in addition, through regional contour reconstruction and defect location, the defect type and distribution are accurately identified, thereby realizing efficient and accurate detection of defects after glass scanning as a whole.

[0016] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description, claims and drawings.

[0017] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a schematic diagram of a first embodiment of a method for detecting glass after scanning according to an embodiment of the present invention;

[0019] Figure 2A schematic diagram of an embodiment of a detection device after glass scanning in an embodiment of the present invention;

[0020] Figure 3 It is a schematic diagram of an embodiment of a detection device after glass scanning in an embodiment of the present invention. DETAILED DESCRIPTION

[0021] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0022] The terms "including" and "having" and any variations thereof mentioned in the embodiments of the present invention are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device end including a series of steps or units is not limited to the listed steps or units, but may optionally include other steps or units that are not listed, or may optionally include other steps or units that are inherent to these processes, methods, products or device ends.

[0023] To facilitate understanding of this embodiment, the specific process of the embodiment of the present invention is described below. Figure 1 The first embodiment of the glass scanning detection method in the embodiment of the present invention includes:

[0024] 101. Collect multi-source original sensor data of the glass to be detected by using a preset multi-sensor array, and perform signal denoising and synchronous transformation on the multi-source original sensor data to obtain pre-processed multi-source original sensor data;

[0025] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Artificial Intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0026] AI basic technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics, etc. AI software technologies mainly include computer vision technology, robotics technology, biometrics technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0027] In this embodiment, the multi-sensor array here refers to an acquisition array composed of a high-resolution area array camera, a laser displacement sensor and a photoelectric sensor, wherein the high-resolution area array camera is used to collect image data of the glass surface (i.e., obtain visual features of surface defects) to capture two-dimensional image information of surface defects such as scratches and stains on the surface; the laser displacement sensor is used to scan the height data of the glass surface (i.e., obtain three-dimensional morphological features of the surface) to measure three-dimensional morphological information such as surface unevenness and roughness, and can accurately detect tiny surface changes; the photoelectric sensor is used to measure the light transmittance data of the glass (i.e., obtain the optical features of the material) to detect the light transmittance and optical uniformity of the glass, and can detect subtle changes in light transmittance; the glass to be inspected here refers to the glass that needs to be quality inspected after the scanning process is performed by the dual-module scanning loading and unloading device, and the items of glass inspection may include but are not limited to: physical features: the surface is scanned to form a flat and smooth surface, and the possible defect types are: scratches (linear damage on the surface), pits (local concave areas), bubbles (bubble defects inside or on the surface The invention relates to a method for detecting a plurality of glass surfaces, comprising: detecting a plurality of glass surfaces, ...

[0028] In practical applications, a multi-sensor array system consisting of an array camera, a laser sensor, and a photoelectric sensor is used to simultaneously collect multi-source data on the glass surface. The array camera acquires high-resolution image data of defects such as scratches and stains on the glass surface at an acquisition frequency of 25fps. The laser sensor scans the three-dimensional morphological data such as the bumps and roughness of the glass surface with a measurement accuracy of 0.1μm. The photoelectric sensor measures the transmittance distribution data of the glass surface at a sampling frequency of 100Hz. Since the glass moves at a constant speed during production, it is necessary to perform window segmentation processing on the collected data according to the transmission speed of the glass. Usually, a 10ms time window is used to segment the data to ensure that the complete glass surface feature information is obtained in each window. The segmented data sequences are then subjected to wavelet transform respectively, and the data collected by different sensors are decomposed into coefficients of multiple scales (where these coefficients reflect the information of the glass surface features at different frequency levels). Then, the high-frequency interference components of the multi-scale segmentation coefficients are removed. The high-frequency interference component mainly comes from factors such as equipment vibration, ambient light changes, and electromagnetic interference. By setting a reasonable threshold, the high-frequency coefficients are screened, the effective coefficients that characterize the true characteristics of the glass surface are retained, and the high-frequency components that represent noise are removed. The screened coefficients are reconstructed and the effective coefficients of different scales are recombined to obtain the multi-source sensor data after noise removal. Furthermore, due to the different sampling frequencies of the three sensors, the acquired data are misaligned in time. It is necessary to use the timestamp information recorded when the sensor collects data for data alignment. That is, first, the sensor data with the highest sampling frequency is used as the benchmark, and the other sensor data are resampled by the cubic spline interpolation method to make the data of the three sensors completely correspond on the time axis. Considering that the data collected by different sensors have different physical dimensions and numerical ranges, the data needs to be normalized so that all data are uniformly mapped to the same numerical range to eliminate the influence of the dimension, and finally obtain pre-processed data with time synchronization, noise elimination and unified dimension.

[0029] 102. Extracting multi-scale glass features corresponding to the glass to be detected from the pre-processed multi-source original sensor data, and performing feature combination and feature dimension reduction compression on the multi-scale glass features to obtain an optimized feature matrix;

[0030] In this embodiment, based on the preset glass division block unit, the pre-processed multi-source original sensor data is regionally segmented into local blocks to obtain a regional segmentation matrix, and grayscale statistics are performed on the image sensor data in the regional segmentation matrix to obtain a glass texture feature matrix, and gradient calculation is performed on the glass surface height data in the regional segmentation matrix to obtain a morphology feature matrix, and optical distribution features are extracted from the optical data in the regional segmentation matrix to obtain an optical feature matrix; the glass texture feature matrix, the morphology feature matrix and the optical feature matrix are normalized to obtain a normalized feature matrix , and perform tensor superposition operation on the normalized feature matrix to obtain a glass feature tensor; perform singular value decomposition of the corresponding matrix dimension on the glass feature tensor to obtain a tensor decomposition result, and extract the feature vectors whose eigenvalues ​​on each dimension in the tensor decomposition result are greater than the preset feature threshold to obtain a modal feature vector; calculate the variance contribution rate corresponding to each eigenvector in the modal feature vector, and select the feature combination corresponding to the variance contribution rate greater than the preset contribution threshold to obtain the important features of the glass defect; perform matrix reconstruction of the corresponding glass defect features and outer product operation of the two-dimensional structure on the important features of the glass defect to obtain an optimized feature matrix. The above-mentioned preset glass partitioning block unit refers to dividing the entire glass surface into uniform small blocks according to the size of 10×10 pixels; the above-mentioned preset feature threshold is a judgment standard (such as 85%) for screening the eigenvalues ​​obtained after tensor decomposition; the above-mentioned preset contribution threshold is a judgment standard (such as 90%) for screening the importance of modal feature vectors.

[0031] In practical applications, the glass surface is divided into uniform block units of 10×10 pixels (to ensure that local detail features are captured while maintaining computational efficiency), and the data in each block is analyzed and processed, that is, the grayscale value distribution characteristics in each block of the image data collected by the array camera are statistically analyzed, including statistics such as grayscale mean, standard deviation, skewness and kurtosis. These characteristics can reflect the defect characteristics such as scratches and stains on the glass surface, thereby forming a glass texture feature matrix; at the same time, the height data collected by the laser sensor is used to calculate the height gradient and curvature characteristics in each block. These characteristics can characterize the glass The surface undulations and roughness changes are calculated to obtain the morphological feature matrix. In addition, the transmittance mean and change rate of each block are extracted from the data collected by the photoelectric sensor. These features reflect the optical performance distribution of the glass and form an optical feature matrix. In order to eliminate the dimensional differences between different features, the three feature matrices are normalized respectively so that all eigenvalues ​​are mapped to a unified numerical range to obtain a normalized feature matrix. The three feature matrices are superimposed and combined in the form of tensors to construct a three-order tensor structure, in which the first dimension represents the block position, the second dimension represents the feature type, and the third dimension represents the feature type. Representing the eigenvalue, a complete glass feature tensor is obtained; then the feature tensor is tensor decomposed, that is, singular value decomposition is performed on three dimensions respectively to obtain the eigenvalue and eigenvector on each dimension, and by setting the feature threshold (such as 85%), the eigenvectors with eigenvalues ​​greater than the preset feature threshold are screened out. These eigenvectors represent the most important change mode in the data, thereby obtaining the modal eigenvector; then the variance contribution rate corresponding to each modal eigenvector is calculated, that is, the degree to which the eigenvector can explain the change of the original data, and by setting the contribution threshold (such as 90%), the eigenvector combinations with variance contribution rates greater than the contribution threshold are selected. These eigenvector combinations contain the most significant defect feature information on the glass surface; then the selected important eigenvectors are recombined into a matrix form, and the feature information of different dimensions is integrated together through the outer product operation to obtain the final optimized feature matrix, wherein this optimized feature matrix not only retains the most important feature information in the original data, but also greatly reduces the dimension of the data and improves the efficiency of subsequent processing. In addition, the entire feature extraction and optimization process makes full use of the complementarity of multi-source sensor data, and effectively extracts the key features of glass surface defects through feature dimensionality reduction and screening.

[0032] 103. Based on the optimized feature matrix, the glass to be inspected is segmented and the regional contour is reconstructed to obtain an initial defect region set, and the initial defect region set is combined with the spatiotemporal features of multi-source features to obtain an enhanced defect feature set;

[0033] In this embodiment, the optimized feature matrix is ​​subjected to spatial density transformation of the defect area to obtain a density distribution map, and the density area in the density distribution map where the defect density is greater than a preset density threshold is marked to obtain an initial marked area; the initial marked area is subjected to clustering of pixels with similar features and regional expansion growth to obtain an initial segmented area, and the boundary features of the grayscale gradient and height difference of the glass surface are extracted from the initial segmented area to obtain a regional boundary contour; the regional boundary contour is subjected to morphological corrosion expansion and noise debris elimination to obtain a smooth boundary area, and the smooth boundary area is subjected to area threshold filtering and merged with adjacent areas to obtain a merged detection area; the merged detection area is subjected to abnormality calculation and detection of multiple detection features to obtain an abnormal detection area, and the abnormal detection area is subjected to defect number labeling and feature attribute association to obtain an initial defect area set; the extraction of the The initial defect area set corresponds to the time series characteristics of the multi-image acquisition moment to obtain a defect feature time series matrix, and the defect feature time series matrix is ​​subjected to feature change trend extraction and key point sampling marking to obtain key time series features; the key time series features are subjected to numerical sequence transformation of preset glass surface characteristics and defect state quantification to obtain a state change matrix, and the state change matrix is ​​subjected to multi-dimensional change characteristics of the glass surface and feature combination to obtain a combined feature sequence; the defect feature significance is calculated and the importance is assigned to the combined feature sequence to obtain a weighted feature set, and the weighted feature set is subjected to multi-source feature integration and glass defect feature enhancement to obtain an enhanced feature vector; the enhanced feature vector is subjected to defect position feature and overall surface feature reorganization to obtain a feature structure set, and the feature structure set is subjected to correlation analysis and combined integration of scanning defects to obtain an enhanced defect feature set. The above-mentioned preset density threshold is an important parameter used to determine the degree of clustering of defect features in spatial distribution. Assuming that in an area of ​​10×10 pixels, the average feature density of normal glass is 0.2, the preset density threshold may be set to 0.5, that is, when the feature density of a certain area exceeds 0.5, the area is considered to be a defect clustering area; the above-mentioned preset glass surface characteristics are the changing characteristics of the grayscale value, height value and transmittance value of the defect area at different times.

[0034] In practical applications, the optimized feature matrix is ​​first transformed into a spatial density of the defect area, and the defect feature density of each position point on the glass surface is calculated (where this density value reflects the degree of clustering of defect features in the area), that is, a density distribution map is constructed by calculating the distribution of eigenvalues ​​in the neighborhood of each position point. This distribution map clearly shows the clustering area of ​​defects on the glass surface, and according to a pre-set density threshold (such as 0.5), the areas where the density value exceeds the threshold are marked. These areas are likely to be the concentrated areas of defects, so as to obtain the initial marked area; then the initial marked area is expanded and grown (that is, the marked area is used as the seed point, and the surrounding pixels with similar features are gradually included in the area), that is, by comparing the feature similarity of adjacent pixels, when the similarity exceeds the set threshold, the pixel is included in the current area, and the expansion is continued until there are no pixels that meet the conditions, thereby obtaining the initial segmented area, and then the boundary features of these segmented areas are extracted, which is mainly determined by using the grayscale gradient information of the image and the difference information of the height data. The boundary of the region is determined to obtain a relatively accurate regional boundary contour; the boundary contour is then morphologically processed, including corrosion and expansion operations, wherein the corrosion operation removes small protrusions on the boundary, while the expansion operation fills small depressions on the boundary. By combining the two, the noise fragments on the boundary are effectively eliminated to obtain a smooth boundary region, and the smoothed boundary region is screened by area, and the region with an area less than the set threshold is regarded as a noise region and deleted, and the adjacent regions are merged to obtain a more reasonable detection region; the merged detection region is then analyzed for abnormality, that is, the degree of abnormality of the region is determined by calculating the degree of deviation between multiple detection features (including shape features, texture features, and optical features, etc.) of each region and the features of the normal region, and for the region determined to be abnormal, it is necessary to mark it with a number and associate its characteristic attributes, including defect type, location, size, etc., to obtain a complete set of initial defect regions (wherein this set not only contains the spatial distribution information of the defect region, but also contains the characteristic attribute information of each defect region). Taking the scratch defect on the glass surface as an example, assume that a potential defect area is detected at the position (100,100): Calculate the local density: Calculate the local density of the eigenvalue in a 5×5 window around the point. If ρ(100,100)=0.85 is obtained, which is higher than the set density threshold of 0.5, then mark the point as the initial seed point; then perform regional growth: With the seed point as the center, calculate the characteristic distance D between the adjacent pixels and the current area. If D<0.3 (set threshold), then include the pixel in the area; then extract the boundary features: calculate the gradient intensity E of the boundary of the area. If E>0.5, it is confirmed as a defect boundary, and finally accurately locate a scratch defect area of ​​about 5mm long and 0.2mm wide.

[0035] Secondly, after obtaining the initial defect area set, the feature changes of each defect area at different times are extracted from the continuous image sequence captured by the area array camera. Since the camera captures images at a rate of 25fps, the continuous change information of the defect area in the time dimension can be obtained, which includes the gray value change, shape change and boundary change of the defect area. The image data, height data and light transmittance data of each defect area at n consecutive moments are tracked and calculated by using the time series feature extraction formula to extract the feature change sequence over time. The time series feature extraction formula is:

[0036] ;

[0037] in, is the time series characteristic value of position (i, j) at time t, is the time attenuation weight, I, H, L represent image, height, and light transmittance features respectively, α, β, γ are feature weight coefficients, λ is the time attenuation factor, It decays exponentially with time. For example, for a certain scratch defect area, the characteristic values ​​at five consecutive moments may be: t1: I=0.8, H=0.6, L=0.7, t2: I=0.82, H=0.58, L=0.72, t3: I=0.79, H=0.61, L=0.69. The time series characteristic value of the area can be calculated by the time series feature extraction formula. . Then, the changing trend of the eigenvalues ​​over time in the defect feature time series matrix is ​​extracted, and the key time points where the eigenvalues ​​change significantly are marked on the change curve. These key points can reflect the important change state of the defect features, thereby obtaining the key time series features; then, according to the characteristics of the glass surface, these key time series features are numerically transformed and quantified, that is, the time series data is converted into a numerical sequence that can characterize the defect state (wherein this sequence reflects the state change of the defect at different times), such as converting the change in scratch depth, the change in pit area, the change in transmittance, etc. into a standardized numerical sequence to form a state change matrix, and extracting the multi-dimensional change characteristics of the glass surface from this matrix, including the dynamic evolution characteristics of the defects, spatial distribution characteristics, and changes in optical characteristics, etc., and combining these characteristics to obtain a feature sequence that can comprehensively characterize the defect characteristics; then, the contribution of each feature in the combined feature sequence to defect recognition is calculated, and the contribution is assigned according to the contribution. Corresponding weights (wherein the setting of weights reflects the importance of different features in defect detection, in this way to highlight the role of key features and suppress the influence of secondary features) are used to obtain a weighted feature set, and the feature data collected by multiple sensors are integrated, and the image features, morphological features and optical features are comprehensively considered. The feature expression of defects is enhanced by feature fusion to form an enhanced feature vector; then the defect position information and the overall feature information of the glass surface in the enhanced feature vector are recombined to construct a feature structure set reflecting the defect distribution and surface state (wherein this structure set includes both the detailed features of local defects and the overall characteristics of the glass surface), and by analyzing the correlation between defect features, the related features are combined and integrated to finally obtain an enhanced defect feature set (wherein this feature set not only retains the important information of the original defect features, but also improves the feature expression capability through multi-source data fusion and feature enhancement).

[0038] 104. Construct a regional similarity matrix corresponding to the enhanced defect feature set, determine the defect category of the regional similarity matrix, and generate a classification label set;

[0039] In this embodiment, numerical difference calculation and feature mapping are performed on each glass defect feature in the enhanced defect feature set to obtain a multidimensional difference matrix, and defect type similarity quantization and threshold division are performed on the multidimensional difference matrix to obtain a defect similarity matrix; the defect similarity matrix is ​​statistically analyzed for correlations between each glass defect feature and constructed for correlation paths to obtain a defect correlation network, and the glass defect features in the defect correlation network are type-determined and classified to obtain a preliminary classification result; the preliminary classification result is statistically analyzed for glass defect features and category confirmation to obtain a classification statistics table, and defect classification labels are generated and attribute associations are performed on the classification statistics table to obtain a classification label set.

[0040] In practical applications, the numerical differences between various defect features are calculated, including the difference values ​​of features such as scratch depth, pit area, and transmittance. That is, the difference values ​​between each defect feature and other features (such as Euclidean distance, Mahalanobis distance, etc.) are calculated by using the feature difference calculation formula, and these difference values ​​are organized into a multidimensional matrix to form a multidimensional difference matrix. The feature difference calculation formula is:

[0041] ;

[0042] in, is the difference between the i-th and j-th defects, is the weight coefficient of the k-th dimension feature, and are the k-th eigenvalues ​​of the i-th and j-th defects, respectively. and are the spatial position coordinates of the two defects, σ ​​is the spatial scale parameter, for example, to calculate the difference between two scratch defects: assuming that there are two scratches with features [length, depth, width] of [2.5, 0.3, 0.1] and [2.3, 0.4, 0.15] respectively, set the weight to [0.4, 0.4, 0.2], the position distance is 5 pixels, σ = 10, then their difference can be calculated , and then perform similarity analysis on this difference matrix, that is, cluster similar defect features by setting a similarity threshold to preliminarily distinguish different types of defects, such as classifying scratches with similar depth and length into one category, and classifying pits with similar area and depth into one category, thereby obtaining a defect similarity matrix; then based on the defect similarity matrix, statistically analyze the frequency of simultaneous occurrence of different defect features, calculate the correlation strength between them, and construct a network structure reflecting this correlation relationship, in which each node in this network represents a defect feature, and the connection between nodes represents their degree of correlation. The greater the correlation strength, the greater the connection weight, so as to form a complete defect correlation network, and use this network structure to analyze the mutual influence and dependency between different types of defects; then after determining the correlation between the defects, the defect features in the network are typed, That is, based on the association relationship established previously, defects with similar characteristics and strong association relationships are classified into the same category. For example, scratches with similar depth and length characteristics and often appearing at the same time are classified into one category, and pits with similar area and depth characteristics and similar distribution patterns are classified into another category. In this way, a preliminary classification result is obtained; then the number, distribution characteristics and typical characteristic values ​​of each type of defects are calculated for the preliminary classification results to form a detailed classification statistics table, and a unique classification label is generated for each type of defect based on the classification statistics table, and these labels are associated with the specific attributes of the defects (wherein these attributes include the location information, geometric characteristics, optical characteristics, etc. of the defects), so that each defect is given a label containing complete information (wherein these labels not only identify the type of defect, but also record the key characteristic attributes of the defect) to form the final classification label set.

[0043] 105. Score the glass quality in multiple dimensions based on the classification label set to obtain glass quality evaluation data, and perform process correlation on the glass quality evaluation data to generate a final test report for the glass to be tested.

[0044] In this embodiment, feature statistics and data extraction are performed on the surface quality parameters corresponding to the classification label set to obtain a quality feature table, and each glass defect feature data in the quality feature table is numerically graded and ranged to obtain a grading standard table; based on the grading standard table, the quality feature table is graded and parameter graded to obtain a grading result matrix, and weight allocation and coefficient calculation are performed on the grading results corresponding to the surface quality parameters in the grading standard table to obtain a weight coefficient matrix, and weighted calculation and score summation are performed on the grading result matrix and the weight coefficient matrix to obtain a comprehensive score value; the comprehensive score value is divided into glass surface areas and defect distribution statistics are performed to obtain a regional score distribution map, The regional score distribution map is integrated with regional scores and quality indicators are generated to obtain glass quality evaluation data; the corresponding surface quality parameters in the glass quality evaluation data are extracted for secondary distribution characteristics and defect characteristics are classified to obtain a defect characteristic table, and the defect types in the defect characteristic table are associated and statistically analyzed with the preset scanning process parameters to obtain a process association table; data comparison and parameter impact calculation are performed on the process association table to obtain an influence degree matrix, and the influence degree matrix is ​​ranked by importance and key defect parameters are screened to obtain a key parameter table; the key parameter table and each of the glass defect characteristics are integrated and the test results are summarized to generate the final test report of the glass to be tested. The above-mentioned preset scanning process parameters refer to glass scanning process parameters (such as mechanical parameters, chemical parameters and environmental parameters, etc.).

[0045] In practical applications, firstly, the surface quality parameters contained therein are statistically analyzed. These parameters include specific numerical characteristics such as scratch depth, pit area, bubble diameter, and light transmittance. That is, by statistically calculating these characteristics, statistical data such as the number distribution, size distribution, and spatial distribution of various defects are obtained. These data are organized into a quality characteristic table, and the parameters in the quality characteristic table are graded according to the quality standard requirements of glass products. For example, the scratch depth is divided into three levels: slight, medium, and severe, and the light transmittance change is divided into uniform, slightly uneven, and severely uneven levels, thereby forming a standardized grading standard table; then, based on the formulated grading standard table, each parameter in the quality characteristic table is graded, and the specific numerical characteristics are mapped to the corresponding level, so that a grading result matrix reflecting the quality level of each parameter is obtained; at the same time, considering that different quality parameters have different degrees of influence on the overall quality of the product, different weight coefficients need to be assigned to each parameter. For example, deep scratches that affect the performance of the product may be more important than slight light transmittance unevenness, so they should be assigned a higher weight, and the weight coefficient matrix is ​​obtained in this way. By performing matrix multiplication operation on the grading result matrix and the weight coefficient matrix and summing the results, we can get a comprehensive score value reflecting the overall quality level of the glass; then we can perform spatial analysis on the obtained comprehensive score value, that is, divide the glass surface into multiple regions, count the defect distribution and quality score in each region, and generate a regional score distribution map that intuitively reflects the quality distribution (wherein, this distribution map not only shows the quality level of different regions, but also reflects the degree of defect aggregation and distribution law), and calculate the weighted average of the scores of each region on the regional score distribution map, taking into account the quality uniformity between regions and combining the specific application requirements of the product, to generate the final quality evaluation data to reflect the quality status of the glass product, which not only includes specific quality indicators, but also embodies the spatial distribution characteristics of quality, thereby providing a reliable basis for product quality judgment and process optimization.

[0046] Secondly, an in-depth distribution characteristic analysis is performed on the surface quality parameters corresponding to the glass quality evaluation data, that is, a secondary extraction is performed on the quality evaluation data obtained previously, focusing on the spatial and numerical distribution laws of defects, such as analyzing the directional distribution of scratches, the depth distribution of pits, the density distribution of bubbles and other characteristics, and classifying and arranging these defects with similar distribution characteristics to form a more systematic defect characteristic table, and comparing and analyzing this defect characteristic table with the process parameters corresponding to the product to be produced in the glass scanning process, wherein these process parameters include key process variables such as scanning pressure, scanning speed, and polishing liquid concentration. By statistically analyzing the defect characteristics corresponding to different process parameters, the corresponding relationship between defect type and process parameters is established, and a process association table is generated; then, based on the process association table, by comparing the defect characteristics under different process parameter values, the defect characteristics under different process parameter values ​​are analyzed. The change of defects, calculate the influence of parameter changes on defect characteristics, such as analyzing the influence of changes in scanning pressure on scratch depth, the influence of changes in scanning speed on surface roughness, etc., quantify these influence relationships into specific values, form an influence degree matrix, and rank the importance of each influencing factor in this matrix, screen out key process parameters that have a significant impact on defect formation, these parameters may be the main causes of specific types of defects, and record these key parameters and their influence in the key parameter table; then systematically integrate the key parameter table with all previously obtained defect characteristic information (wherein the integration process not only summarizes the basic information such as defect type, distribution, severity, etc., but also includes the correlation between these defects and process parameters, as well as possible causes and improvement suggestions), so as to finally generate a complete inspection report.

[0047] In the embodiment of the present invention, the basic surface features are obtained by multi-sensor data collection and feature extraction on the glass surface, and then these features are combined for dimensionality reduction and regional segmentation to obtain defect area features; then, through multi-source feature fusion and spatiotemporal feature enhancement, defect enhancement features are obtained and similarity analysis is performed; finally, quality scoring and process association are performed based on category judgment, and a test report is output. Through hierarchical data processing and feature analysis, the problem of accurate identification of defect detection after glass scanning is solved, especially in terms of surface scratches, pits and transmittance, the multi-dimensional characteristics and defect characteristics of the glass surface are fully considered, and the detection accuracy is effectively improved; and the multi-level feature extraction and defect mapping strategy are adopted, which not only realizes the correlation analysis between surface parameters, but also enhances the reliability of defect identification; in addition, through regional contour reconstruction and defect location, the defect type and distribution are accurately identified, thereby realizing efficient and accurate detection of defects after glass scanning as a whole.

[0048] The above describes the detection method after glass scanning in the embodiment of the present invention. The following describes the detection device after glass scanning in the embodiment of the present invention. Figure 2 , an embodiment of the glass scanning detection device in the embodiment of the present invention comprises:

[0049] The preprocessing module 201 is used to collect multi-source original sensor data of the glass to be detected by using a preset multi-sensor array, and perform signal denoising and synchronous transformation on the multi-source original sensor data to obtain preprocessed multi-source original sensor data;

[0050] A feature optimization module 202 is used to extract multi-scale glass features corresponding to the glass to be detected in the pre-processed multi-source original sensor data, and perform feature combination and feature dimension reduction compression on the multi-scale glass features to obtain an optimized feature matrix;

[0051] The feature combination module 203 is used to perform regional segmentation and regional contour reconstruction on the glass to be inspected based on the optimized feature matrix to obtain an initial defect region set, and to perform spatiotemporal feature combination of multi-source features on the initial defect region set to obtain an enhanced defect feature set;

[0052] The category determination module 204 is used to construct a regional similarity matrix corresponding to the enhanced defect feature set, and to determine the defect category of the regional similarity matrix to generate a classification label set;

[0053] The report generation module 205 is used to score the multi-dimensional glass quality of the classification label set to obtain glass quality evaluation data, and to perform process association on the glass quality evaluation data to generate a final test report for the glass to be tested.

[0054] In the embodiment of the present invention, the basic surface features are obtained by multi-sensor data collection and feature extraction on the glass surface, and then these features are combined for dimensionality reduction and regional segmentation to obtain defect area features; then, through multi-source feature fusion and spatiotemporal feature enhancement, defect enhancement features are obtained and similarity analysis is performed; finally, quality scoring and process association are performed based on category judgment, and a test report is output. Through hierarchical data processing and feature analysis, the problem of accurate identification of defect detection after glass scanning is solved, especially in terms of surface scratches, pits and transmittance, the multi-dimensional characteristics and defect characteristics of the glass surface are fully considered, and the detection accuracy is effectively improved; and the multi-level feature extraction and defect mapping strategy are adopted, which not only realizes the correlation analysis between surface parameters, but also enhances the reliability of defect identification; in addition, through regional contour reconstruction and defect location, the defect type and distribution are accurately identified, thereby realizing efficient and accurate detection of defects after glass scanning as a whole.

[0055] above Figure 2The detection device after glass scanning in the embodiment of the present invention is described in detail from the perspective of modular functional entities. The detection device after glass scanning in the embodiment of the present invention is described in detail from the perspective of hardware processing.

[0056] Figure 3 3 is a schematic diagram of the structure of a glass inspection device after scanning provided by an embodiment of the present invention. The glass inspection device 300 after scanning may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) 310 (for example, one or more processors) and a memory 320, and one or more storage media 330 (for example, one or more mass storage devices) storing application programs 333 or data 332. Among them, the memory 320 and the storage medium 330 may be temporary storage or permanent storage. The program stored in the storage medium 330 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations in the glass inspection device 300 after scanning. Furthermore, the processor 310 may be configured to communicate with the storage medium 330 to execute a series of instruction operations in the storage medium 330 on the glass inspection device 300 after scanning.

[0057] The glass scanning inspection device 300 may also include one or more power supplies 340, one or more wired or wireless network interfaces 350, one or more input and output interfaces 360, and / or one or more operating systems 331, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. It will be appreciated by those skilled in the art that Figure 3 The structure of the inspection device for glass after scanning is not intended to limit the inspection device for glass after scanning, and may include more or less components than shown in the figure, or a combination of certain components, or a different arrangement of components.

[0058] The present invention also provides a glass scanning inspection device, the computer device includes a memory and a processor, the memory stores computer-readable instructions, when the computer-readable instructions are executed by the processor, the processor executes each step of the glass scanning inspection method in the above embodiments.

[0059] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0060] 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 this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the whole 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, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program code.

[0061] The present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc. The present application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application can also be practiced in distributed computing environments, in which tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

[0062] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting glass after scanning, characterized in that: The detection method after the glass is scanned includes: Using a preset multi-sensor array to collect multi-source raw sensor data of the glass to be detected, and performing signal denoising and synchronous transformation on the multi-source raw sensor data to obtain pre-processed multi-source raw sensor data; Based on the preset glass division block unit, the preprocessed multi-source original sensor data is regionally segmented into local blocks to obtain a regional segmentation matrix, and the image sensor data in the regional segmentation matrix is ​​statistically analyzed in grayscale to obtain a glass texture feature matrix, and the glass surface height data in the regional segmentation matrix is ​​gradient operated to obtain a morphology feature matrix, and the optical data in the regional segmentation matrix is ​​extracted by optical distribution features to obtain an optical feature matrix; the glass texture feature matrix, the morphology feature matrix and the optical feature matrix are normalized to obtain a normalized feature matrix, and the The normalized feature matrix is ​​subjected to tensor superposition operation to obtain a glass feature tensor; the glass feature tensor is subjected to singular value decomposition of the corresponding matrix dimension to obtain a tensor decomposition result, and the feature vectors whose eigenvalues ​​in each dimension of the tensor decomposition result are greater than a preset feature threshold are extracted to obtain a modal feature vector; the variance contribution rate corresponding to each eigenvector in the modal feature vector is calculated, and the feature combination corresponding to the variance contribution rate greater than a preset contribution threshold is selected to obtain important features of glass defects; the important features of glass defects are subjected to matrix reconstruction corresponding to the glass defect features and an outer product operation of a two-dimensional structure to obtain an optimized feature matrix; Based on the optimized feature matrix, the glass to be inspected is segmented and the regional contour is reconstructed to obtain an initial defect region set, and the time series features of the initial defect region set corresponding to the multi-image acquisition moment are extracted to obtain a defect feature time series matrix, and the defect feature time series matrix is ​​extracted for feature change trends and key point sampling marks to obtain key time series features; the key time series features are transformed into a numerical sequence of preset glass surface characteristics and defect state quantification to obtain a state change matrix, and the state change matrix is ​​extracted and features are combined to obtain a combined feature sequence; the defect feature significance is calculated and the importance is assigned to the combined feature sequence to obtain a weighted feature set, and the weighted feature set is integrated with multi-source features and the glass defect features are enhanced to obtain an enhanced feature vector; the enhanced feature vector is reorganized with defect position features and surface overall features to obtain a feature structure set, and the feature structure set is subjected to correlation analysis and combined integration of scanning defects to obtain an enhanced defect feature set; Constructing a regional similarity matrix corresponding to the enhanced defect feature set, determining defect categories on the regional similarity matrix, and generating a classification label set; The classification label set is scored for multi-dimensional glass quality to obtain glass quality evaluation data, and the glass quality evaluation data is process-associated to generate a final test report for the glass to be tested.

2. The method for detecting glass after scanning according to claim 1, characterized in that: The performing signal denoising and synchronous transformation on the multi-source raw sensor data to obtain pre-processed multi-source raw sensor data includes: Based on a preset glass acquisition speed, the multi-source original sensor data is window segmented to obtain a multi-source original sensor subsequence, and the size coefficient of each of the multi-source original sensor subsequences is extracted to obtain a multi-scale segmentation coefficient; Performing threshold screening of high-frequency interference components corresponding to glass detection on the multi-scale segmented coefficients to obtain multi-scale noise reduction coefficients, and reconstructing coefficient data on the multi-scale noise reduction coefficients to obtain filtered multi-source original sensing data; Based on the corresponding sampling timestamps of the multi-sensor array, the filtered multi-source raw sensor data is time-aligned and the data stream is normalized to obtain the pre-processed multi-source raw sensor data.

3. The method for detecting glass after scanning according to claim 1, characterized in that: The step of performing region segmentation and region contour reconstruction on the glass to be inspected based on the optimized feature matrix to obtain an initial defect region set includes: Performing spatial density transformation of defect regions on the optimized feature matrix to obtain a density distribution map, and marking density regions in the density distribution map where the defect density is greater than a preset density threshold to obtain an initial marked region; Aggregating pixels with similar features and performing regional expansion growth on the initial marked area to obtain an initial segmented area, and extracting boundary features of the grayscale gradient and height difference of the glass surface on the initial segmented area to obtain a regional boundary contour; Performing morphological corrosion and dilation and noise debris elimination on the boundary contour of the region to obtain a smooth boundary region, and performing area threshold filtering on the smooth boundary region and merging it with adjacent regions to obtain a merged detection region; The abnormality degree of multiple detection features is calculated and detected for the combined detection area to obtain the abnormal detection area, and the defect number labeling and feature attribute association are performed on the abnormal detection area to obtain an initial defect area set.

4. The method for detecting glass after scanning according to claim 1, characterized in that: The step of constructing a regional similarity matrix corresponding to the enhanced defect feature set, determining defect categories on the regional similarity matrix, and generating a classification label set includes: Performing numerical difference calculation and feature mapping on each glass defect feature in the enhanced defect feature set to obtain a multidimensional difference matrix, and performing defect type similarity quantification and threshold division on the multidimensional difference matrix to obtain a defect similarity matrix; The defect similarity matrix is ​​used to perform correlation statistics and correlation path construction between each glass defect feature to obtain a defect correlation network, and the glass defect features in the defect correlation network are type-determined and classified to obtain a preliminary classification result; The preliminary classification results are subjected to statistics of glass defect features and category confirmation to obtain a classification statistics table, and defect classification labels are generated and attribute associations are performed on the classification statistics table to obtain a classification label set.

5. The method for detecting glass after scanning according to claim 1, characterized in that: The multi-dimensional glass quality scoring of the classification label set is performed to obtain glass quality evaluation data, including: Performing feature statistics and data extraction on the surface quality parameters corresponding to the classification label set to obtain a quality feature table, and performing numerical grading and range division on each glass defect feature data in the quality feature table to obtain a grading standard table; Based on the grading standard table, the quality characteristic table is graded and parameter graded to obtain a grading result matrix, and weights are assigned and coefficients are calculated for the grading results corresponding to the surface quality parameters in the grading standard table to obtain a weight coefficient matrix, and weighted calculation and score summation are performed on the grading result matrix and the weight coefficient matrix to obtain a comprehensive score value; The comprehensive score value is used to divide the glass surface area and perform defect distribution statistics to obtain a regional score distribution map, and the regional score distribution map is used to perform regional score integration and quality index generation to obtain glass quality evaluation data.

6. The method for detecting glass after scanning according to claim 5, characterized in that: The step of performing process association on the glass quality evaluation data to generate a final test report for the glass to be tested includes: Extracting secondary distribution features and classifying defect features of the surface quality parameters corresponding to the glass quality evaluation data to obtain a defect feature table, and correlating and statistically analyzing the defect types in the defect feature table with preset scanning process parameters to obtain a process correlation table; Performing data comparison and parameter influence calculation on the process association table to obtain an influence degree matrix, and performing importance sorting and key defect parameters screening on the influence degree matrix to obtain a key parameter table; The key parameter table and each of the glass defect characteristics are integrated and the test results are summarized to generate a final test report for the glass to be tested.

7. A glass scanning detection device, characterized in that: The glass scanning detection device adopts the glass scanning detection method according to any one of claims 1 to 6, and the glass scanning detection device comprises: A preprocessing module, used to collect multi-source original sensor data of the glass to be detected by using a preset multi-sensor array, and perform signal denoising and synchronous transformation on the multi-source original sensor data to obtain preprocessed multi-source original sensor data; A feature optimization module, used for extracting multi-scale glass features corresponding to the glass to be detected in the pre-processed multi-source original sensor data, and performing feature combination and feature dimension reduction compression on the multi-scale glass features to obtain an optimized feature matrix; A feature combination module, used to perform regional segmentation and regional contour reconstruction on the glass to be inspected based on the optimized feature matrix to obtain an initial defect region set, and perform spatiotemporal feature combination of multi-source features on the initial defect region set to obtain an enhanced defect feature set; A category determination module, used to construct a regional similarity matrix corresponding to the enhanced defect feature set, and determine the defect category of the regional similarity matrix to generate a classification label set; The report generation module is used to score the multi-dimensional glass quality of the classification label set to obtain glass quality evaluation data, and to perform process association on the glass quality evaluation data to generate a final test report for the glass to be tested.

8. A glass scanning inspection device, characterized in that: The glass scanning detection device comprises: a memory and at least one processor, wherein the memory stores instructions; The at least one processor calls the instructions in the memory to enable the glass post-scanning inspection device to perform each step of the glass post-scanning inspection method according to any one of claims 1-6.

Images