Method and device for detecting etching thickness of key glass

Abstract

The invention provides a key glass etching thickness detection method and device, and the method comprises the steps: collecting sample images of all key glass calibration samples under the conditions of fixed illumination intensity and imaging; based on the sample image, determining a corresponding etching area and a target pixel feature vector thereof; establishing a target etching thickness calibration model based on the pixel feature vector of each etching area and the corresponding real thickness value; under the conditions of fixed illumination intensity and imaging, extracting to-be-detected pixel feature vectors of the etching areas from the to-be-detected key glass image; inputting the to-be-measured pixel feature vector into the target etching thickness calibration model to obtain a predicted thickness value of each etching area; and comparing the predicted thickness value with a preset process tolerance interval to obtain a detection judgment result of each etching region. Therefore, non-contact quantitative measurement of the etching thickness is realized, and meanwhile, the requirements of quick, non-contact and quantitative detection of the etching thickness of the key glass and multi-key batch online application are met.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electronic product glass processing and quality detection, and particularly relates to a detection method and device for etching thickness of a key glass. BACKGROUND

[0002] With the popularity of smart phones, tablet computers, vehicle-mounted central control and smart home electronic products, key glass with pattern etching and functional identification is widely used in touch key, decorative panel and backlight display scenes. The surface of the key glass is usually formed with a recessed structure through chemical etching or laser etching to realize light transmission, light guiding or tactile feedback functions, and the etching thickness is directly related to the brightness uniformity of key characters, mechanical strength and service life. Therefore, accurate detection and process control of the etching thickness in the production process are crucial, and are a core link for guaranteeing product quality, improving production efficiency and optimizing process parameters.

[0003] At present, the detection of glass thickness or etching thickness in industrial sites mainly adopts two categories of methods, namely contact type mechanical measurement and non-contact type optical measurement, and defect detection technology based on machine vision, but all have significant limitations. The contact type mechanical measurement method (such as micrometer, dial gauge, mechanical thickness gauge) obtains thickness data through the relative displacement of the measuring head and the measured surface. This kind of method has simple structure and low cost, but needs manual point-by-point operation, and the measurement efficiency is low; and the measuring head directly contacts the glass surface, which is easy to scratch or bruise, and cannot meet the demand of online full detection of batch key glass.

[0004] The non-contact type optical measurement method (such as devices based on optical interference or confocal principle, including Michelson interferometer modification, white light interferometer, three-dimensional optical profilometer, etc.) can realize the measurement of transparent material thickness or step height in micrometer / sub-micrometer level. However, such devices are generally complex in structure, expensive in price, and high in requirements for vibration and environmental stability; and most of them adopt point measurement or small field scanning mode, which is difficult to simultaneously and quickly measure multiple etching regions on a piece of key glass on the production line, and is not suitable for large-scale online quality detection and machine optimization.

[0005] Machine vision defect detection technology can replace manual visual inspection through image recognition to realize automatic identification and classification of surface defects such as scratches, edge collapse and color difference, but it mainly outputs detection results or defect type information, and the accurate thickness or etching depth of the etching area still needs to be measured separately by contact or interference instruments. Although the existing micro-imaging combined with image processing technology can evaluate geometric dimensions, it mainly targets two-dimensional dimensions of opaque / semi-transparent materials and cannot solve the problem of quantitative measurement of three-dimensional thickness of etching areas of transparent key glass; and the feature extraction and measurement results are mainly simple geometric relationships, which are difficult to adapt to the nonlinear mapping relationship between image gray level / texture and etching thickness under the conditions of transparent substrate, complex backlight structure and multiple processes.

[0006] In summary, the existing technology cannot simultaneously meet the requirements of rapid, non-contact and quantitative detection of key glass etching thickness and batch online application of multiple key positions, and there is an urgent need for a detection method that can establish a reliable mapping relationship between image pixel features and etching thickness under ordinary imaging conditions. SUMMARY

[0007] The present application provides a detection method and device for key glass etching thickness, which can realize non-contact quantitative measurement of etching thickness under ordinary optical imaging conditions, and simultaneously meet the requirements of rapid, non-contact and quantitative detection of key glass etching thickness and batch online application of multiple key positions.

[0008] In a first aspect, the present application provides a detection method for key glass etching thickness, comprising: Under fixed illumination intensity and imaging conditions, sample images of all key glass calibration samples are collected; Based on the sample images, corresponding etching areas and their target pixel feature vectors are determined; Based on the pixel feature vectors of each etching area and the corresponding true thickness values, a target etching thickness calibration model is established; Under the fixed illumination intensity and imaging conditions, the test pixel feature vectors of each etching area are extracted from the images of the key glass to be tested; The test pixel feature vectors are input into the target etching thickness calibration model to obtain the predicted thickness values of each etching area; The predicted thickness values are compared with a preset process tolerance interval to obtain the detection judgment results of each etching area.

[0009] Optionally, based on the sample images, the corresponding etching areas and their target pixel feature vectors are determined, comprising: Using image algorithms, the sample images are segmented based on the key glass outer contour and / or glass printing pattern and / or a preset template to determine the etching areas and their pixel sets; extracting the target pixel feature vector of all the etching areas from the pixel set.

[0010] Optionally, after collecting the sample images of all the key glass calibration samples under the fixed light intensity and imaging conditions, the method further comprises: preprocessing the sample images; the preprocessing step comprises: performing geometric distortion correction on the sample images based on the obtained camera calibration parameters; performing brightness normalization processing on the sample images to unify the gray scale range; performing background compensation and filtering denoising operations on the sample images.

[0011] Optionally, the pixel feature vector comprises a gray scale statistical feature and a texture feature; extracting the target pixel feature vector of all the etching areas from the pixel set comprises: calculating the mean and variance of the gray scale of the pixels in the pixel set to determine the gray scale statistical feature; constructing a gray scale co-occurrence matrix based on the pixel set and calculating at least one of contrast, homogeneity and energy from the gray scale co-occurrence matrix as the texture feature; combining the gray scale statistical feature and the texture feature to form the target pixel feature vector.

[0012] Optionally, after comparing the predicted thickness value with the preset process tolerance interval to obtain the detection judgment result of each etching area, the method further comprises: generating an etching thickness distribution map of the to-be-tested key glass according to the predicted thickness values of all the etching areas, and combining the detection judgment result of each etching area to obtain a comprehensive quality grade evaluation of the to-be-tested key glass.

[0013] Optionally, after comparing the predicted thickness value with the preset process tolerance interval to obtain the detection judgment result of each etching area, the method further comprises: adaptively updating the target etching thickness calibration model; the adaptive updating step comprises: extracting samples from the actual production of key glasses according to a preset frequency; collecting sample images of the extracted samples under the fixed light intensity and imaging conditions to extract sample pixel features, and combining the true thickness of part of the etching areas in the extracted samples to form a new sample set; updating the parameters of the target etching thickness calibration model using the new sample set, and the updated model is used for subsequent to-be-tested glass detection.

[0014] Optionally, when the target etching thickness calibration model is a multiple linear regression model, the etching thickness calibration model is specifically: ; wherein, is a predicted etching thickness, is an average gray level, is a gray level standard deviation, is an edge density, is a texture energy, , , , and is a regression coefficient.

[0015] Optionally, when the target etching thickness calibration model is a feedforward neural network model; based on the pixel feature vectors of each etching region and the corresponding true thickness values, a target etching thickness calibration model is established, including: inputting the pixel feature vectors into an initially established etching thickness calibration model to obtain corresponding thickness prediction values; determining training errors according to the thickness prediction values and the true thickness values; based on the training errors, optimizing network parameters of the etching thickness calibration model until a training termination condition is met, to obtain optimal network parameters; adopting the optimal network parameters to generate the target etching thickness calibration model.

[0016] In a second aspect, the present application provides a detection device for etching thickness of key glass, comprising: a sample image acquisition module for acquiring sample images of all key glass calibration samples under fixed illumination intensity and imaging conditions; a pixel feature vector determination module for determining corresponding etching regions and their target pixel feature vectors based on the sample images; a model establishment module for establishing a target etching thickness calibration model based on the pixel feature vectors of each etching region and the corresponding true thickness values; a pixel feature vector extraction module for extracting the pixel feature vectors of each etching region from the image of the key glass to be detected under the fixed illumination intensity and imaging conditions; a prediction module for inputting the pixel feature vectors to be detected into the target etching thickness calibration model to obtain the predicted thickness values of each etching region; a comparison module for comparing the predicted thickness values with a preset process tolerance interval to obtain the detection judgment results of each etching region.

[0017] In a third aspect, the present application provides an electronic device comprising a processor and a memory, wherein the memory stores computer readable instructions which, when executed by the processor, perform the steps of the method according to the first aspect.

[0018] In a fourth aspect, the present application provides a storage medium having stored thereon a computer program which, when executed by a processor, performs the steps of the method according to the first aspect.

[0019] In a fifth aspect, the present application provides a computer program product comprising a computer program which, when executed by a processor, performs the steps of the method according to the first aspect.

[0020] From the above technical solutions, the present application has the following advantages: The present application provides a method and device for detecting the etching thickness of key glass, comprising: under fixed light intensity and imaging conditions, collecting sample images of all key glass calibration samples; based on the sample images, determining corresponding etching regions and target pixel feature vectors thereof; based on the pixel feature vectors of each etching region and corresponding true thickness values, establishing a target etching thickness calibration model; under the fixed light intensity and imaging conditions, extracting test pixel feature vectors of each etching region from a to-be-tested key glass image; inputting the test pixel feature vectors into the target etching thickness calibration model to obtain predicted thickness values of each etching region; and comparing the predicted thickness values with a preset process tolerance interval to obtain detection determination results of each etching region. The target etching thickness calibration model is used to predict the thickness of the to-be-tested key glass, realizing non-contact quantitative measurement of the etching thickness, and meeting the requirements of rapid, non-contact, quantitative detection of the etching thickness and batch online application of multiple key positions. BRIEF DESCRIPTION OF DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments or prior art description will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor.

[0022] Figure 1 Flow chart of the first embodiment of the method for detecting the etching thickness of key glass according to the present application; Figure 2 Flow chart of the second embodiment of the method for detecting the etching thickness of key glass according to the present application; Figure 3It is a structure schematic view of a key glass etching thickness detection system of a key glass etching thickness detection method embodiment two of the application. Figure 4 It is a structure block diagram of a key glass etching thickness detection device embodiment of the application. DETAILED DESCRIPTION

[0023] The embodiment of the application provides a key glass etching thickness detection method and device, which can realize non-contact quantitative measurement of etching thickness under ordinary optical imaging conditions, and meets the requirements of rapid, non-contact, quantitative detection and batch online application of multiple key positions.

[0024] In order to make the application purpose, features and advantages of the application more obvious and easy to understand, the technical solutions in the embodiments of the application will be clearly and completely described below in combination with the drawings in the embodiments of the application. Obviously, the following described embodiments are only a part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the application.

[0025] Embodiment one Please refer to Figure 1 , Figure 1 It is a flow step diagram of a key glass etching thickness detection method embodiment one of the application, and the method comprises the following steps. Step S101, under fixed light intensity and imaging conditions, sample images of all key glass calibration samples are collected. In the embodiment of the application, within the thickness range allowed by the etching process, N groups of key glass calibration samples with different etching thicknesses are prepared, and the etching patterns of each group of samples are the same as or equivalent to the key layout of the actual product. A high-precision measuring device (such as a step meter or an interferometer) is used to measure the actual thickness of each etching area on each calibration sample, which is used as a thickness true value label.

[0026] Step S102, based on the sample image, the corresponding etching area and its target pixel feature vector are determined. In the embodiment of the application, the preprocessed image is segmented based on the key glass outer contour, the printed pattern or the preset template. The center and the boundary of each etching area are located by using template matching, shape detection or Hough transformation algorithm, and the pixel set corresponding to each etching area is obtained, and the target pixel feature vector is determined based on the pixel combination.

[0027] Step S103, based on the pixel feature vector of each etching area and the corresponding actual thickness value, a target etching thickness calibration model is established. In the implementation of the present application, the etching thickness calibration model can be a multiple linear regression model, a support vector regression model or a feedforward neural network model, etc.

[0028] Step S104, under the fixed light intensity and imaging conditions, a to-be-measured pixel feature vector of each etching region is extracted from the to-be-measured key glass image; In the implementation of the present application, the to-be-measured key glass is placed on a sample platform, and an image is collected under the imaging conditions consistent with the calibration stage. The pixel set of each etching region is determined by using the same method as that in the calibration stage. For each etching region, the same pixel features as those in the calibration stage are extracted to form a to-be-measured pixel feature vector.

[0029] Step S105, the to-be-measured pixel feature vector is input into the target etching thickness calibration model to obtain a predicted thickness value of each etching region; In the implementation of the present application, the to-be-measured pixel feature vector of each etching region is input into the trained target etching thickness calibration model, and the model outputs the predicted thickness value of each etching region.

[0030] Step S106, the predicted thickness value is compared with a preset process tolerance interval to obtain a detection judgment result of each etching region.

[0031] In the implementation of the present application, according to the process requirements, a qualified tolerance interval of etching thickness is set. For each etching region, the predicted thickness is compared with the tolerance interval to obtain a corresponding detection judgment result.

[0032] The detection method for the etching thickness of the key glass provided by the embodiment of the present application comprises the following steps: collecting sample images of all key glass calibration samples under fixed light intensity and imaging conditions; determining corresponding etching regions and target pixel feature vectors based on the sample images; establishing a target etching thickness calibration model based on the pixel feature vectors of each etching region and corresponding true thickness values; extracting to-be-measured pixel feature vectors of each etching region from a to-be-measured key glass image under the fixed light intensity and imaging conditions; inputting the to-be-measured pixel feature vectors into the target etching thickness calibration model to obtain predicted thickness values of each etching region; and comparing the predicted thickness values with a preset process tolerance interval to obtain detection judgment results of each etching region. The thickness of the to-be-measured key glass is predicted by using the target etching thickness calibration model, non-contact quantitative measurement of the etching thickness is realized, and the requirements of fast, non-contact, quantitative detection and batch online application of the etching thickness of the key glass are met.

[0033] Embodiment two Please refer to Figure 2 , Figure 2 Fig. 2 is a flow step diagram of the detection method for the etching thickness of the key glass according to the embodiment two of the present application, which comprises the following steps: Step S201, under fixed light intensity and imaging conditions, sample images of all key glass calibration samples are collected; The key glass etching thickness detection method of the embodiment is applied to Figure 3 The key glass etching thickness detection system, which comprises a support platform 1, a light unit 2, an imaging unit 3, a computing processing unit 4 and a key glass, wherein the support platform 1 is a vibration-proof optical platform, used for fixing and placing calibration samples and key glasses to be detected, so as to ensure that the positions of the samples are stable and can be repositioned during the detection process. The light unit 2 comprises a ring-shaped LED cold light source 21 and a backlight plate 22. The ring-shaped LED cold light source 21 is installed around the imaging unit 3, and the incident angle thereof is preferably about 45°, used for providing oblique incident reflected light. The backlight plate 22 is arranged below the key glass, used for providing uniform transmitted light. Through the combination of oblique incidence and backlight illumination, the etching area and the non-etching area form obvious brightness and texture differences in the image. The imaging unit 3 adopts a 5 million pixel CCD industrial camera, with a typical resolution of 2448x2048 pixels, and is equipped with a 50 mm fixed-focus lens. The optical axis of the camera is perpendicular to the surface of the support platform 1, and the imaging field of view covers the whole key glass and all etching areas thereof. After the imaging distance and the magnification are reasonably set, each pixel corresponds to an actual size of about 10 μm, so that the changes in etching depth in the range of 20-80 μm can be distinguished. The computing processing unit 4 is an industrial computer, which is configured with an image preprocessing module, an etching area positioning module, a pixel feature extraction module, an etching thickness calibration model training module and an etching thickness calculation module, and is electrically connected with the imaging unit 3 through an image acquisition card, used for receiving digital images collected by the camera and executing image preprocessing, etching area positioning, pixel feature extraction, thickness calibration model establishment and thickness calculation algorithms, and simultaneously displaying or outputting the detection results to a production line control system.

[0034] In the embodiment of the present application, under the premise that the geometric positions of the light unit 2 and the imaging unit 3, the light intensity and the exposure time and other parameters are fixed, the calibration samples are placed on the support platform one by one, and the corresponding original images, i.e. sample images, are collected by the imaging unit 3.

[0035] Step S202, the sample images are preprocessed; In the embodiment of the present application, the preprocessing step comprises: performing geometric distortion correction on the sample images based on the obtained camera calibration parameters; performing brightness normalization processing on the sample images to unify the gray scale range; and performing background compensation and filter denoising operations on the sample images.

[0036] In a specific implementation, the image preprocessing module pre-processes the sample image, including 1. grayscale processing, distortion correction, ROI extraction, brightness normalization and background compensation and denoising, contrast enhancement, and threshold segmentation.

[0037] The grayscale processing includes: converting a color image into a grayscale image to improve the efficiency of subsequent processing.

[0038] The distortion correction includes: based on camera calibration parameters, performing geometric distortion correction on the original image to obtain a corrected image.

[0039] The brightness normalization includes: using linear stretching or gamma correction to normalize the grayscale of the geometric distortion corrected image to a unified range, which can be represented as: wherein, is the brightness normalized image, is one of the normalization coefficients obtained according to histogram statistics, is the geometric distortion corrected image, is the second normalization coefficient obtained according to histogram statistics, is the image coordinate.

[0040] The ROI extraction includes: according to the predetermined position of the key glass shape and etching pattern and cropping, extracting a region of interest (ROI) sub-image containing the etching region.

[0041] The background compensation and denoising includes: suppressing uneven illumination by Gaussian filtering, median filtering or background fitting subtraction, which can be represented as: wherein, is the filtered image, is the background compensation and denoising operator.

[0042] The contrast enhancement includes: performing histogram equalization processing on the image to obtain an enhanced image, which improves the brightness contrast between the etching region and the non-etching region. The threshold segmentation includes: using the Otsu algorithm to perform adaptive threshold segmentation on the item to obtain a binary image, wherein the etching region is the foreground and the non-etching region is the background.

[0043] In step S203, an image algorithm is used to segment the sample image based on the key glass outer contour and / or the glass printing pattern and / or a preset template to determine the etching region and its pixel set. In the embodiments of the present application, the etching region positioning module segments the pre-processed sample image based on one or more of the key glass outer contour, the printing pattern and the preset template to obtain the pixel set corresponding to different etching regions in each sample.​​

[0044] In a specific implementation, a template matching, shape detection or Hough transform algorithm can be used to locate the center and boundary of each etching region, and obtain a pixel set.

[0045] In step S204, the target pixel feature vector of each etching region is extracted from the pixel set; the pixel feature vector includes a gray scale statistical feature and a texture feature. In the embodiments of the present application, the gray scale mean and the gray scale variance of the pixels in the pixel set are calculated to determine the gray scale statistical feature; a gray scale co-occurrence matrix is constructed based on the pixel set, and at least one of the contrast, homogeneity and energy is calculated from the gray scale co-occurrence matrix as the texture feature; the gray scale statistical feature and the texture feature are combined to form the target pixel feature vector.

[0046] In a specific implementation, for each etching region of the calibration sample, the pixel feature extraction module calculates a multi-dimensional feature vector from the pixel set, and the typical components of the multi-dimensional feature vector include the gray scale statistical feature, the texture feature, the gradient and edge feature, and the frequency domain or filter response feature. The gray scale statistical feature is determined by the gray scale mean and the gray scale variance of the pixels, and the texture feature includes at least one of the contrast, homogeneity and energy calculated from the gray scale co-occurrence matrix.

[0047] The gray scale mean is determined by the following expression: ; wherein, is the pixel set, is the gray scale mean.

[0048] The gray scale variance is determined by the following expression: ; wherein, is the gray scale variance.

[0049] The contrast is determined by the following expression: ; wherein, is the gray scale value of the first pixel, is the gray scale value of the second pixel, is the gray scale co-occurrence matrix, is the contrast.

[0050] The homogeneity is determined by the following expression: ; wherein, is the homogeneity.

[0051] The energy is determined by the following expression: ; wherein, is the energy.

[0052] The gradient and edge features are obtained by performing gradient operation on the processed wafer image, denoted as: , , Then the average gradient amplitude is defined within the pixel set, i.e.: ; wherein, is the gradient component of the image in the horizontal direction, is the gradient component of the image in the vertical direction, is the gradient amplitude at the pixel point , and is the average gradient amplitude.

[0053] The frequency domain or filter response feature can perform two-dimensional fast Fourier transform or band-pass filtering on the sub-image extracted from the pixel set, i.e.: ; wherein, is the two-dimensional discrete Fourier transform operator, is the frequency domain representation, is the horizontal spatial frequency, is the vertical spatial frequency.

[0054] Then the energy within the specific frequency band is counted, i.e.: .

[0055] wherein, is the energy within the specific frequency band, and is the selected frequency band region.

[0056] Finally, the above various features are combined into the feature vector of the first etching region of the wafer wafer: ; wherein, is the real number set, is the feature vector, is the dimension of the feature vector.

[0057] Step S205, based on the pixel feature vector of each etching region and the corresponding true thickness value, a target etching thickness calibration model is established; ​In this embodiment of the application, the pixel feature vectors and actual thickness values ​​of each etched region in the sample image constitute a training sample set: By establishing a functional mapping relationship between pixel features and etching thickness. ;in, This corresponds to the predicted etching thickness. The parameters are used to train the model, enabling non-contact quantitative estimation of the etching thickness.

[0058] In practice, the calibration sample of the button glass uses the same glass material and etching process as the actual button glass, forming etched areas at the same pattern locations. Within the allowable range of the etching process, by controlling the etching time to 10s, 30s, 60s, and 90s, four sets of button glass calibration samples with etching depths of approximately 20μm, 40μm, 60μm, and 80μm were obtained.

[0059] For the etched area of ​​each keypad glass calibration sample, a contact profilometer was first used to scan along the cross-section containing the etched area to obtain the etch groove profile curve; then, a white light interferometric 3D measuring instrument was used to measure the surface morphology of the area. The thickness difference between the two measurement results was controlled within ±1 μm, and their average value was taken as the true value of the etch thickness of the calibration sample.

[0060] For example, the true etching thickness values ​​obtained for the four calibration samples are 20.0 μm, 39.8 μm, 60.3 μm, and 79.7 μm, respectively. These true thickness values ​​are used as the actual thickness values ​​and correspond one-to-one with the calibration sample numbers.

[0061] In an optional embodiment, when the target etching thickness calibration model is a multiple linear regression model, the etching thickness calibration model is: ; in, For the predicted etching thickness, For bias terms, To act on the feature vector The There are basis functions, which can be linear terms, quadratic terms, interaction terms, radial basis functions, etc. The number of basis functions.

[0062] The etching thickness calibration model in this application embodiment is specifically as follows: ; For the predicted etching thickness, The average gray level, The standard deviation of grayscale For edge density, For texture energy, , , , and is a regression coefficient.

[0063] In a specific implementation, the parameter vector is determined by least squares, i.e., minimizing a loss function: .

[0064] The optimal parameters are obtained by matrix operations, and the fitting performance of the model is evaluated by cross-validation.

[0065] In the experiment of the present embodiment, the fitted example parameters are: , , , , .

[0066] Therefore, the etching thickness calibration model of the present embodiment can be specifically represented as: .

[0067] The training data is fitted, and the model determination coefficient is greater than 0.98, indicating that the linear model can well describe the relationship between the pixel features and the etching thickness.

[0068] In an optional embodiment, when the target etching thickness calibration model is a feedforward neural network model; based on the pixel feature vectors of each etching region and the corresponding true thickness values, the target etching thickness calibration model is established, including: inputting the pixel feature vector into the initially established etching thickness calibration model to obtain a corresponding thickness prediction value; determining a training error according to the thickness prediction value and the true thickness value; based on the training error, optimizing the network parameters of the etching thickness calibration model until the training termination condition is met, to obtain optimal network parameters; using the optimal network parameters to generate the target etching thickness calibration model.

[0069] In the embodiments of the present application, for the training sample set, the optimal parameters are determined by minimizing the loss function. The preferred loss function is a weighted least squares form with a regularization term: ; wherein, is the loss function based on the optimal parameters, is the optimal parameter, is the sample weight, and , is a regularization coefficient, and , is a regularization term.

[0070] In a specific implementation, the optimization process can be implemented by using numerical optimization algorithms such as gradient descent, conjugate gradient, L-BFGS, etc. The parameters obtained through training are fixed as the target etching thickness calibration model, and a functional mapping from pixel features to etching thickness is realized.

[0071] Step S206, under the fixed illumination intensity and imaging conditions, a to-be-measured pixel feature vector of each etching region is extracted from the to-be-measured key glass image; In the actual detection stage of the present application, the to-be-measured key glass is placed on the object platform 1, and the to-be-measured key glass image is collected under the imaging conditions consistent with the calibration stage by the illumination unit 2 and the imaging unit 3, and the distortion correction, brightness normalization and background compensation are sequentially performed to obtain the preprocessed image.

[0072] The etching region positioning module determines the pixel set of each etching region of the to-be-measured glass according to the same method as in the calibration stage, and extracts a to-be-measured pixel feature vector for each region.

[0073] Step S207, inputting the to-be-measured pixel feature vector into the target etching thickness calibration model to obtain a predicted thickness value of each etching region; In the present application, the feature vector of each etching region is input into the target etching thickness calibration model to obtain the thickness prediction value of each etching region.

[0074] Step S208, comparing the predicted thickness value with a preset process tolerance interval to obtain a detection judgment result of each etching region; In the present application, the computing processing unit further compares the predicted thickness of each etching region with the thickness tolerance interval required by the process to obtain the detection judgment result of each etching region.

[0075] Step S209, generating an etching thickness distribution map of the to-be-measured key glass according to the predicted thickness values of all etching regions, and combining the detection judgment results of each etching region to obtain a comprehensive quality grade evaluation of the to-be-measured key glass; In the present application, the predicted thickness values of all etching regions are summarized to generate a thickness distribution map, which can be colorized and visualized. According to the judgment results of each region, the qualified rate is calculated or the weight is set, and the quality grade evaluation of the whole glass is output.

[0076] Step S210, adaptively updating the target etching thickness calibration model; In the embodiment of the present application, the adaptive updating step comprises: periodically selecting a plurality of key glass in actual production according to a preset frequency, and performing spot checks on the partial etching areas thereof to obtain the true thickness and simultaneously extract the corresponding pixel feature vectors to form a new sample set. The new sample set and the original calibration data set are merged or weightedly fused, and the updated model parameters are obtained through incremental learning or re-optimization of the loss function. Through this updating mechanism, the pixel feature-etching thickness calibration model maintains high measurement accuracy and stability in long-term operation to cope with the situations of light source attenuation, camera characteristic change and process parameter slow drift.

[0077] In a specific implementation, a plurality of key glasses in actual production are periodically selected according to a preset frequency, and the partial etching areas thereof are spot checked to obtain the true thickness, and simultaneously the corresponding pixel feature vectors are extracted to form a new sample set. The new sample set and the original calibration data set are merged or weightedly fused, and the updated model parameters are obtained through incremental learning or re-optimization of the loss function. Through this updating mechanism, the pixel feature-etching thickness calibration model maintains high measurement accuracy and stability in long-term operation to cope with the situations of light source attenuation, camera characteristic change and process parameter slow drift.

[0078] The detection method for etching thickness of key glass provided by the embodiment of the present application comprises the following steps: collecting sample images of all key glass calibration samples under fixed illumination intensity and imaging conditions; determining corresponding etching areas and target pixel feature vectors based on the sample images; establishing a target etching thickness calibration model based on the pixel feature vectors of each etching area and the corresponding true thickness values; extracting the pixel feature vectors of each etching area from the images of the key glass to be measured under the fixed illumination intensity and imaging conditions; inputting the pixel feature vectors to be measured into the target etching thickness calibration model to obtain the predicted thickness values of each etching area; and comparing the predicted thickness values with a preset process tolerance interval to obtain the detection judgment results of each etching area. The target etching thickness calibration model is used to predict the thickness of the key glass to be measured, realizing non-contact quantitative measurement of the etching thickness, and meeting the requirements of rapid, non-contact, quantitative detection of the etching thickness of the key glass and batch online application of multiple key positions. At the same time, by generating an etching thickness distribution map, quality visual analysis and process tracing are realized; a comprehensive quality level evaluation is given combined with the multi-region judgment results, supporting more comprehensive quality decision and closed-loop control.

[0079] In order to facilitate the understanding of the beneficial effects of the present application by those skilled in the art, a detection method for etching thickness of key glass is exemplarily described below.

[0080] This example verifies a plurality of test samples that do not participate in modeling. For each test sample, image acquisition, preprocessing, feature extraction and thickness prediction are performed according to the steps mentioned in the above embodiment, and the true thickness measured by a profilometer or an interferometer is compared.

[0081] Table 1 gives example data of four test samples, including feature values, predicted thickness, true thickness, and error.

[0082] Table 1 Example of thickness detection model test data

[0083] As can be seen from Table 1, in this example, the thickness range of the test samples is about 20-80 μm, the predicted values of each sample are highly close to the true values, the absolute error is within ±3 μm, and the relative error is not more than 5%. The prediction error of T3 sample is only -0.2 μm, and the relative error is about -0.3%.

[0084] Therefore, the key glass etching thickness detection method based on the pixel feature-etching thickness calibration model can realize non-contact quantitative measurement of etching thickness under ordinary optical imaging conditions, has the advantages of high detection precision, high efficiency, and batch detection of multiple etching regions, and is suitable for online quality monitoring and process control on the production line of key glass and other optical glass products.

[0085] Example Three Please refer to Figure 4 , Figure 4 is a structure block diagram of an embodiment of a key glass etching thickness detection device of the present application. The device comprises: A sample image acquisition module 301 is configured to acquire sample images of all key glass calibration samples under fixed illumination intensity and imaging conditions. A pixel feature vector determination module 302 is configured to determine corresponding etching regions and target pixel feature vectors thereof based on the sample images. A model establishment module 303 is configured to establish a target etching thickness calibration model based on pixel feature vectors of each etching region and corresponding true thickness values. A pixel feature vector extraction module 304 is configured to extract test pixel feature vectors of each etching region from a to-be-detected key glass image under the fixed illumination intensity and imaging conditions. A prediction module 305 is configured to input the test pixel feature vectors into the target etching thickness calibration model to obtain predicted thickness values of each etching region. A comparison module 306 is configured to compare the predicted thickness values with a preset process tolerance interval to obtain detection determination results of each etching region.

[0086] In an optional embodiment, the pixel feature vector determination module 302 comprises: The segmentation sub-module is configured to segment the sample image based on the key glass outer contour and / or the glass printing pattern and / or a preset template by using an image algorithm, so as to determine the etching area and a pixel set thereof; The target vector extraction sub-module is configured to extract the target pixel feature vector of all the etching areas from the pixel set.

[0087] In an optional embodiment, the method further comprises: The pre-processing module is configured to pre-process the sample image; The pre-processing module comprises: The correction sub-module is configured to perform geometric distortion correction on the sample image based on the obtained camera calibration parameters; The normalization processing sub-module is configured to perform brightness normalization processing on the sample image to unify the gray scale range; The filtering sub-module is configured to perform background compensation and filtering denoising operations on the sample image.

[0088] In an optional embodiment, the pixel feature vector comprises a gray scale statistical feature and a texture feature; and the target vector extraction sub-module comprises: The gray scale statistical feature determination unit is configured to calculate the mean value and variance of the gray scale of the pixels in the pixel set, so as to determine the gray scale statistical feature; The texture feature determination unit is configured to construct a gray scale co-occurrence matrix based on the pixel set, and calculate at least one of contrast, homogeneity and energy from the gray scale co-occurrence matrix as the texture feature; The target vector formation unit is configured to combine the gray scale statistical feature and the texture feature to form the target pixel feature vector.

[0089] In an optional embodiment, the method further comprises: The grade evaluation module is configured to generate an etching thickness distribution map of the to-be-tested key glass according to the predicted thickness values of all the etching areas, and obtain a comprehensive quality grade evaluation of the to-be-tested key glass in combination with the detection and judgment results of the etching areas.

[0090] In an optional embodiment, the method further comprises: The adaptive updating module is configured to adaptively update the target etching thickness calibration model; The adaptive updating module comprises: The sample extraction sub-module is configured to extract samples from the actually produced key glass at a preset frequency; A new sample construction submodule is configured to collect sample images of the extracted samples under the fixed illumination intensity and imaging conditions to extract sample pixel features, and to construct a new sample set in combination with the real thickness of part of the etching area in the extracted samples. An updating submodule is configured to update parameters of the target etching thickness calibration model using the new sample set, and the updated model is used for subsequent detection of the to-be-detected glass.

[0091] In an optional embodiment, when the target etching thickness calibration model is a multiple linear regression model, the etching thickness calibration model is specifically: ; wherein, is a predicted etching thickness, is an average gray scale, is a gray scale standard deviation, is an edge density, is a texture energy, 、 、 、 and is a regression coefficient.

[0092] In an optional embodiment, when the target etching thickness calibration model is a feedforward neural network model, the model establishing module 303 includes: A training prediction submodule is configured to input the pixel feature vector into an initially established etching thickness calibration model to obtain a corresponding thickness prediction value; A training error determination submodule is configured to determine a training error according to the thickness prediction value and the real thickness value; An optimization submodule is configured to optimize network parameters of the etching thickness calibration model based on the training error until a training termination condition is met, to obtain optimal network parameters; A model generation submodule is configured to generate the target etching thickness calibration model using the optimal network parameters.

[0093] Embodiment Four The embodiments of the present application also provide an electronic device including a memory and a processor, the memory storing a computer program, and the computer program being executed by the processor to make the processor execute the steps of the detection method of the etching thickness of the key glass according to any one of the embodiments.

[0094] Embodiment Five The embodiments of the present application also provide a computer storage medium storing a computer program, and the computer program being executed by the processor to implement the steps of the detection method of the etching thickness of the key glass according to any one of the embodiments.

[0095] Embodiment six The embodiment of the present application further provides a computer program product, which stores a computer program, and the computer program is executed by a processor to realize the steps of the key glass etching thickness detection method according to any one of the embodiments.

[0096] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the system, device and unit described above can refer to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0097] In several embodiments provided in the present application, it should be understood that the disclosed method, device, electronic equipment and storage medium can be implemented by other manners. For example, the foregoing device embodiments are only schematic, for example, the division of the units is only a logical function division, and actual implementation can have another division manner, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some interfaces, devices or units, and can be electrical, mechanical or other forms.

[0098] The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, that is, they can be located in one place, or can be distributed on a plurality of network units. According to actual needs, part or all of the units can be selected to achieve the purpose of the embodiment scheme.

[0099] In addition, each functional unit in each embodiment of the present application can be integrated in one processing unit, or each unit can be physically present separately, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of a software functional unit.

[0100] The integrated unit, if implemented in the form of a software function unit and sold or used as an independent product, can be stored in a computer-readable storage medium. Based on such understanding, the technical solutions of the present application or all or part of the technical solutions that essentially contribute to the prior art can be embodied in the form of a software product. The computer software product is stored in a readable storage medium, including a number of 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 described in the various embodiments of the present application. The aforementioned readable storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0101] The above-described embodiments are only used to illustrate the technical solutions of the present application, rather than limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacements for some technical features; and these modifications or replacements do not make the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for detecting the etching thickness of button glass, characterized in that, include: Under fixed illumination intensity and imaging conditions, sample images of all key glass calibration samples were acquired; Based on the sample image, determine the corresponding etched area and its target pixel feature vector; A target etching thickness calibration model is established based on the pixel feature vectors of each etched region and their corresponding true thickness values. Under the fixed illumination intensity and imaging conditions, the feature vectors of the pixels to be tested in each etched area are extracted from the image of the button glass to be tested. The feature vector of the pixel to be tested is input into the target etching thickness calibration model to obtain the predicted thickness value of each etching region. The predicted thickness value is compared with the preset process tolerance range to obtain the detection and judgment results of each etched area.

2. The method for detecting the etching thickness of button glass according to claim 1, characterized in that, Based on the sample image, the corresponding etched region and its target pixel feature vector are determined, including: Using image algorithms, the sample image is segmented based on the outer contour of the button glass and / or the glass printing pattern and / or the preset template to determine the etched area and its pixel set; Extract the feature vectors of the target pixels for all etched regions from the set of pixels.

3. The method for detecting the etching thickness of button glass according to claim 1, characterized in that, After acquiring sample images of all keypad glass calibration samples under fixed illumination and imaging conditions, the process also includes: The sample image is preprocessed; the preprocessing steps include: Geometric distortion correction is performed on the sample images based on the acquired camera calibration parameters; The sample image is subjected to brightness normalization processing to unify the grayscale range; Background compensation and filtering / denoising operations are performed on the sample image.

4. The method for detecting the etching thickness of button glass according to claim 2, characterized in that, The pixel feature vector includes: grayscale statistical features and texture features; extracting the target pixel feature vector of all etched regions from the pixel set includes: Calculate the mean gray level and the variance of the pixels in the pixel set to determine the gray level statistical features; Based on the pixel set, a gray-level co-occurrence matrix is ​​constructed, and at least one of contrast, homogeneity, and energy is calculated from the gray-level co-occurrence matrix as the texture feature; The grayscale statistical features and the texture features are combined to form the target pixel feature vector.

5. The method for detecting the etching thickness of button glass according to claim 1, characterized in that, After comparing the predicted thickness value with the preset process tolerance range to obtain the detection judgment result for each etched area, the following steps are also included: Based on the predicted thickness values ​​of all etched areas, an etching thickness distribution map of the key glass to be tested is generated. Combined with the detection and judgment results of each etched area, a comprehensive quality grade evaluation of the key glass to be tested is obtained.

6. The method for detecting the etching thickness of button glass according to claim 1 or 5, characterized in that, After comparing the predicted thickness value with the preset process tolerance range to obtain the detection judgment result for each etched area, the following steps are also included: The target etching thickness calibration model is adaptively updated; the adaptive update step includes: Samples are extracted from the button glass produced in actual production at a preset frequency. Under the fixed illumination intensity and imaging conditions, sample images of the extracted samples are acquired to extract sample pixel features, and combined with the true thickness of some etched areas in the sampled samples to form a new sample set; The parameters of the target etching thickness calibration model are updated using the newly added sample set, and the updated model is used for subsequent testing of the glass to be tested.

7. The method for detecting the etching thickness of button glass according to claim 1, characterized in that, When the target etching thickness calibration model is a multiple linear regression model, the etching thickness calibration model is specifically as follows: ； in, For the predicted etching thickness, The average gray level, The standard deviation of grayscale For edge density, For texture energy, , , , and is the regression coefficient.

8. The method for detecting the etching thickness of button glass according to claim 1, characterized in that, When the target etching thickness calibration model is a feedforward neural network model; based on the pixel feature vectors of each etched region and the corresponding true thickness values, the target etching thickness calibration model is established, including: The pixel feature vector is input into the initially established etching thickness calibration model to obtain the corresponding thickness prediction value; The training error is determined based on the predicted thickness value and the actual thickness value. Based on the training error, the network parameters of the etching thickness calibration model are optimized until the training termination condition is met, and the optimal network parameters are obtained. The target etching thickness calibration model is generated using the optimal network parameters.

9. A device for detecting the etching thickness of button glass, characterized in that, include: The sample image acquisition module is used to acquire sample images of all button glass calibration samples under fixed illumination intensity and imaging conditions. The pixel feature vector determination module is used to determine the corresponding etched area and its target pixel feature vector based on the sample image. The model building module is used to build a target etching thickness calibration model based on the pixel feature vectors of each etched region and the corresponding true thickness values. The pixel feature vector extraction module is used to extract the pixel feature vectors of each etched area from the key glass image under the fixed illumination intensity and imaging conditions. The prediction module is used to input the feature vector of the pixel to be tested into the target etching thickness calibration model to obtain the predicted thickness value of each etching region. The comparison module is used to compare the predicted thickness value with the preset process tolerance range to obtain the detection and judgment results of each etched area.

10. An electronic device, characterized in that, It includes a processor and a memory, the memory storing computer-readable instructions that, when executed by the processor, perform the method as described in any one of claims 1-8.

Images