Transparent film online thickness detection method and device and storage medium

Abstract

This application discloses an online thickness detection method, apparatus, and storage medium for transparent thin films, relating to the field of thin film detection technology. The method includes: acquiring interference spectral data of scanning points formed during online thin film scanning, determining the original thickness value and confidence level of each scanning point; determining the predicted thickness value of each scanning point based on historical original thickness value data and historical confidence level data; generating a fused thickness value of each scanning point based on the original thickness value, predicted thickness value, and confidence level; classifying each scanning point based on the original thickness value, predicted thickness value, and spatial distribution characteristics between scanning points, including real defect points, random noise points, and normal points; correcting the fused thickness value and confidence level based on the classification; and determining the final confidence level and final thickness value of each scanning point. This application solves the technical problem of simultaneously effectively suppressing random noise and accurately preserving real defects in online thin film thickness detection, reducing the missed detection rate of minute defects.

Description

Technical Field

[0001] This application relates to the field of transparent film detection technology, and in particular to a method, apparatus and storage medium for online thickness detection of transparent films. Background Technology

[0002] In the online production of transparent films, line-scan white light spectral interferometry is commonly used for thickness measurement. However, production line vibrations can cause severe fluctuations in the interference signal, resulting in the calculated thickness value being mixed with a large amount of random noise.

[0003] Existing methods generally use fixed parameter filtering (such as Gaussian smoothing) to process these noises. However, due to the lack of a dynamic evaluation mechanism for data reliability, this indiscriminate processing method cannot effectively distinguish between "random noise caused by vibration" and "real defects such as scratches and pinholes" - either the noise removal is not thorough, resulting in large fluctuations in thickness data, which cannot be used for precise process control; or the smoothing is too excessive, causing tiny defects to be erased, resulting in missed detections.

[0004] Therefore, how to effectively suppress random noise and accurately preserve real defects under strong vibration interference is a key technical problem that needs to be solved in online film thickness detection.

[0005] The above content is only used to help understand the technical solution of this application and does not represent an admission that the above content is prior art. Summary of the Invention

[0006] The main objective of this application is to provide a method, apparatus, and storage medium for online thickness detection of transparent films, aiming to solve the technical problem of effectively suppressing random noise while accurately preserving the true defect characteristics when detecting the thickness of transparent films online under the interference of production line vibration.

[0007] To achieve the above objectives, this application proposes an online thickness detection method for transparent films, the method comprising: Acquire interference spectral data of scanning points formed during online scanning of transparent thin films; Based on the interference spectral data of the scanning points, the original thickness value of the scanning points and the confidence level of the scanning points are determined; Based on the collected historical raw thickness data and historical confidence data, the predicted thickness value of the scan point is determined; The fused thickness value of the scan point is generated based on the original thickness value, the predicted thickness value, and the confidence level of the scan point. Based on the original thickness value, the predicted thickness value, and the spatial distribution characteristics of the scan point and other scan points, the classification of the scan point is determined, including real defect points, random noise points, and normal points. Based on the classification of the scan points, the fusion thickness value and confidence level of the scan points are corrected to determine the final confidence level and final thickness value of the scan points.

[0008] In one embodiment, the step of determining the predicted thickness value of the scan point based on the collected historical raw thickness value data and historical confidence data includes: Based on the collected historical raw thickness data, the substrate component of the transparent film is determined, and the substrate component is used to characterize the amount of macroscopic curl change of the transparent film substrate; The difference between the historical original thickness value data and the substrate component is calculated as the historical high-frequency variation component of the transparent film; Based on the historical confidence data, the historical high-frequency change components are weighted, and the weighted historical high-frequency change components are used to predict the predicted high-frequency change components of the scan point. The sum of the predicted high-frequency variation component and the substrate component of the scan point is calculated and used as the predicted thickness value of the scan point.

[0009] In one embodiment, the step of weighting the historical high-frequency change components based on the historical confidence data and using the weighted historical high-frequency change components to predict the predicted high-frequency change components of the scan point includes: The Kalman filter algorithm is used to obtain the prior high-frequency change components of the scanning point based on the historical high-frequency change components. The observation noise covariance of the Kalman filter algorithm is determined based on the historical confidence data, and the process noise covariance of the Kalman filter algorithm is determined based on the degree of change of the historical high-frequency variation components. Based on the prior high-frequency variation component, the observation noise covariance, and the process noise covariance, the predicted high-frequency variation component of the scan point is determined by the Kalman filtering algorithm. The observation noise covariance is used to adjust the trust ratio of the historical high-frequency variation component in the Kalman filter algorithm, so as to achieve weighting of the historical high-frequency variation component; the process noise covariance is used to adjust the response speed of the Kalman filter algorithm to the sudden change of the historical high-frequency variation component. When a sudden change of the historical high-frequency variation component is detected, the process noise covariance is increased to enable the Kalman filter algorithm to quickly track the sudden change of the historical high-frequency variation component.

[0010] In one embodiment, the step of determining the classification of the scan points based on the original thickness value, the predicted thickness value, and the spatial distribution characteristics of the scan points compared to other scan points includes: The absolute value of the difference between the original thickness value and the predicted thickness value of the scan point is calculated as the fusion residual of the scan point; When the fusion residual of the scan point is less than a preset residual threshold, the scan point is classified as a normal point. When the fusion residual of the scan point is greater than or equal to a preset residual threshold, the scan point is determined to be a candidate anomaly point; The classification of the candidate anomalies is determined based on the spatial distribution characteristics between the scan points and other scan points.

[0011] In one embodiment, the step of determining the classification of the candidate anomaly points based on the spatial distribution characteristics between the scan points and other scan points includes: Based on the spatial distribution characteristics between the scan points and other scan points, the candidate anomaly points are spatially connected component marked to determine the abnormal connected region to which the candidate anomaly points belong. When the area of ​​the abnormal connected region is less than a preset area threshold, the candidate abnormal points within the abnormal connected region are classified as random noise points. When the area of ​​the abnormal connected region is greater than or equal to a preset area threshold, the candidate abnormal points within the abnormal connected region are classified as real defect points.

[0012] In one embodiment, the step of spatially labeling the candidate anomaly points based on the spatial distribution characteristics between the scan points and other scan points, and determining the abnormal connected region to which the candidate anomaly points belong, includes: Check whether there are any marked candidate anomalies within the preset neighborhood of the candidate anomalies; If it exists, the candidate anomaly point is marked with the same abnormal connected region label as the already marked candidate anomaly point; If it does not exist, the candidate anomaly point is marked as a new anomaly connected region label.

[0013] In one embodiment, the step of correcting the fusion thickness value and confidence level of the scan points based on their classification, and determining the final confidence level and final thickness value of the scan points, includes: When the scan point is classified as a normal point, the final confidence level of the scan point is determined as the confidence level of the scan point, and the final thickness value of the scan point is determined as the fusion thickness value of the scan point; When the scan point is classified as a real defect point, the final confidence level of the scan point is determined to be a preset first confidence level, and the final thickness value of the scan point is determined based on the original thickness value, the predicted thickness value, and the final confidence level. When the scan point is classified as a random noise point, the final confidence level of the scan point is determined to be a preset second confidence level, and the final thickness value of the scan point is determined based on the original thickness value, the predicted thickness value, and the final confidence level. Wherein, the first confidence level is greater than the second confidence level.

[0014] In one embodiment, the step of determining the original thickness value of the scanning point and the confidence level of the scanning point based on the interference spectral data of the scanning point includes: Based on the interferometric spectral data of the scanning points, the wavenumber domain data of the scanning points are determined; The wavenumber domain data of the scanning points are processed using a film thickness analysis algorithm to determine the original thickness value of the scanning points. The film thickness analysis algorithm includes fast Fourier transform or phase unenvelope algorithm. Based on the interference spectral data of the scanning points, the interference modulation index of the scanning points is calculated; The interference modulation of the scan points is processed using a piecewise linear mapping strategy to determine the confidence level of the scan points. The piecewise linear mapping strategy includes threshold calibration and / or weight calculation.

[0015] Furthermore, to achieve the above objectives, this application also proposes an online thickness detection device for transparent films, the online thickness detection device for transparent films comprising: The spectral acquisition module is used to acquire interference spectral data of the scanning points formed during the online scanning of transparent films; The confidence level determination module is used to determine the original thickness value of the scanning point and the confidence level of the scanning point based on the interference spectral data of the scanning point; The thickness prediction module determines the predicted thickness value of the scan point based on the collected historical raw thickness value data and historical confidence data. The fusion thickness determination module generates the fusion thickness value of the scan point based on the original thickness value, the predicted thickness value, and the confidence level of the scan point. The classification determination module determines the classification of the scan points based on the original thickness value, the predicted thickness value, and the spatial distribution characteristics between the scan points and other scan points. The classification includes real defect points, random noise points, and normal points. The data correction module corrects the fusion thickness value and confidence level of the scan points based on their classification, and determines the final confidence level and final thickness value of the scan points.

[0016] In addition, to achieve the above objectives, this application also proposes an online thickness detection device for transparent films, the device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the online thickness detection method for transparent films as described above.

[0017] In addition, to achieve the above objectives, this application also proposes a storage medium, which is a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the online thickness detection method for transparent films as described above.

[0018] In addition, to achieve the above objectives, this application also provides a computer program product, which includes a computer program that, when executed by a processor, implements the steps of the online thickness detection method for transparent films as described above.

[0019] This application provides an online thickness detection method for transparent films. First, it synchronously extracts the original thickness value and its data quality label—confidence level—from interference spectral data, providing a dynamic judgment basis for subsequent processing. Next, it generates a predicted thickness value based on the collected historical original thickness value data and historical confidence level data, dynamically weighting the historical data using the historical confidence level to ensure the predicted value remains stable and reliable even in vibration-affected areas. Then, it dynamically adjusts the fusion weight according to the confidence level, adaptively fusing the original thickness value and the predicted thickness value. In areas where vibration interference leads to low confidence levels, the method automatically... The system relies on predicted values ​​to smooth out noise, while preserving original measurement details in areas with normal signal and high confidence. Based on this, by analyzing the difference between the original thickness value and the predicted thickness value, as well as the spatial distribution characteristics between the scanning points and other scanning points, each scanning point is accurately classified into normal points, real defect points, or random noise points, thereby distinguishing random noise caused by vibration from real defects such as scratches and pinholes. Finally, based on the classification results, targeted thickness values ​​and confidence levels are adjusted for different categories of scanning points, making noise points more reliant on predicted values ​​to eliminate interference, and defect points more reliant on original measurement values ​​to preserve defect characteristics.

[0020] Therefore, this application achieves dynamic evaluation of the reliability of measurement data through confidence perception, thereby overcoming the shortcomings of fixed parameter filtering in the prior art which cannot distinguish signal quality differences. It dynamically achieves the optimal balance between noise reduction and fidelity preservation, and finally outputs high-precision thickness values, confidence distribution, and reliable defect classification results simultaneously. This provides the production line with high-quality online detection data that has both anti-interference capabilities and defect discrimination capabilities, reduces the missed detection rate of minor defects, and solves the technical problem that it is impossible to simultaneously effectively suppress random noise and accurately retain real defects in the online detection of transparent film thickness under production line vibration interference. Attached Figure Description

[0021] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0022] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a schematic flowchart of an embodiment of the online thickness detection method for transparent films in this application. Figure 2 This is a schematic flowchart of Embodiment 2 of the online thickness detection method for transparent films in this application; Figure 3 This is a partial schematic diagram of the abnormal connected regions used to determine each candidate abnormal point in Embodiment 2 of this application; Figure 4 This is a schematic diagram of the module structure of the online thickness detection device for transparent films according to an embodiment of this application; Figure 5 This is a schematic diagram of the equipment structure of the hardware operating environment involved in the online thickness detection method for transparent films in this application embodiment.

[0024] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0025] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of this application and are not intended to limit this application.

[0026] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0027] It should be noted that the executing entity in this embodiment can be a computing service device with data processing, network communication, and program execution functions, such as a tablet computer, personal computer, or mobile phone, or an electronic device capable of performing the above functions, such as an online thickness detection device for transparent films. The following description uses an online thickness detection device for transparent films as an example to illustrate this embodiment and the subsequent embodiments.

[0028] Example 1 Currently, in the online production of transparent films, the thickness measurement method based on the principle of white light spectral interferometry is mainly used. By calculating the frequency components in the interference spectrum that are related to the thickness of the transparent film, the thickness value at each scanning point can be obtained. However, in actual production line operation, mechanical vibration can cause instantaneous fluctuations in the distance between the detector and the transparent film, resulting in a sharp decline in the quality of the interference spectral signal. This leads to obvious instantaneous jumps in the calculated thickness value, forming random noise.

[0029] Existing processing solutions mainly fall into two categories: one is to use a low-pass filter with fixed parameters (such as Gaussian smoothing) to process the entire thickness image. However, this method cannot distinguish between isolated noise caused by vibration and real defects such as scratches and pinholes that are continuously distributed. If the filter parameters are not set properly, either noise residue will cause large data fluctuations, or excessive smoothing will cause tiny defects to be erased, resulting in missed detections. The other is to use multiple independent single-point measurement probes for sampling inspection. However, the data from each probe is isolated and has a detection blind zone, which cannot reflect the lateral uniformity of the transparent film thickness. Tiny defects with extremely small lateral areas are easily missed.

[0030] Furthermore, the aforementioned solutions only output a single thickness value, lacking a quantitative assessment of the reliability of data at each measurement point, making it impossible for process engineers to determine the authenticity and credibility of the test results. Therefore, how to achieve a balance between "noise reduction" and "fidelity preservation" under production line vibration interference is a technical problem that urgently needs to be solved in online film thickness detection.

[0031] Based on this, embodiments of this application provide an online thickness detection method for transparent films, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the online thickness detection method for transparent films according to this application.

[0032] In this embodiment, the online thickness detection method for transparent films includes steps S10-S60: Step S10: Acquire interference spectral data of the scanning points formed during the online scanning of the transparent thin film; It should be noted that the online scanning process refers to the process by which a line-scanning white light spectral interferometry system scans and acquires data line by line along the direction perpendicular to the movement of the transparent film (i.e., the scanning width direction) while the transparent film is moving continuously. During this process, the system is fixedly installed above the transparent film, which moves continuously at a certain speed, and the system acquires data line by line, forming a scanning point array covering the entire surface of the transparent film.

[0033] It should be noted that a scan point refers to a spatial location on the transparent thin film surface corresponding to each detection unit of the linear array detector in the online scanning white light spectral interferometry system along the scan width direction. Interferometric spectral data for one line of scan points are acquired in a single exposure.

[0034] It should be noted that the line-scan white light spectral interferometry system includes a broadband white light source, a beam splitting module (such as a Michelson or Linnick interferometer structure), a spectrometer, and an imaging optical system. The beam emitted from the white light source illuminates the surface of a transparent thin film after passing through the beam splitting module. The light reflected from the upper and lower surfaces of the transparent film interferes due to the optical path difference. After analysis by the spectrometer, the wavelength-varying interference spectral signal is obtained. The spectrometer has multiple detector units arranged along the scan width direction, allowing simultaneous acquisition of interference spectral data from multiple spatial points in a single exposure. As the transparent film moves, the system acquires data line by line, forming a three-dimensional interference spectral data cube D(x,y,λ), which represents the interference spectral data of the scanned points. Here, x is the spatial coordinate (pixel index along the scan width direction), y is the time coordinate (scan row index), and λ is the spectral dimension (wavelength).

[0035] The interference spectral data serves as the raw input for all subsequent processing. It contains both interference fringe frequency information related to the thickness of the transparent film and interference fringe contrast information reflecting signal quality, providing a data foundation for simultaneously calculating the original thickness value and confidence level in subsequent steps.

[0036] Step S20: Based on the interference spectral data of the scanning points, determine the original thickness value of the scanning points and the confidence level of the scanning points; It should be noted that the original thickness value refers to the physical thickness of the transparent film directly calculated from the interference spectrum data using a film thickness analysis algorithm. Since the white light interference spectrum contains the modulation signal formed by the interference of reflected light from the upper and lower surfaces of the transparent film, and the frequency of this modulation signal is proportional to the optical thickness of the transparent film, the physical thickness of the transparent film can be calculated by performing frequency domain analysis or phase analysis on the interference spectrum.

[0037] It should be noted that confidence level is a quantitative assessment index of the reliability of the original thickness value, with a value ranging from 0 to 1, where 0 indicates that the measurement value is completely unreliable and 1 indicates that the measurement value is completely reliable. Confidence level is calculated based on the quality characteristics of the interference spectral signal itself, rather than on the statistical distribution of the thickness value. In the actual production line environment, due to factors such as mechanical vibration, defocusing of the transparent film, and dust obstruction, the quality of the interference spectral signal at some scanning points will significantly decrease, causing the calculated original thickness value to deviate from the true value. Therefore, it is necessary to generate a confidence level synchronously for each scanning point to measure how reliable the original thickness value at that point is, thus providing a basis for judgment in the dynamic weighted fusion step S40.

[0038] Understandably, the original thickness value and the confidence level are two independent yet complementary pieces of information extracted simultaneously from the same interferometric spectral data—the former answers "what is the thickness," and the latter answers "how accurate is this thickness value."

[0039] Step S30: Based on the collected historical raw thickness data and historical confidence data, determine the predicted thickness value of the scanning point; It should be noted that historical raw thickness data refers to the raw thickness values ​​of scan points acquired before the current scan point. These scan points include spatial neighbors that are on the same scan line as the current scan point and were scanned earlier in the sequence, as well as temporal neighbors that are on different scan lines than the current scan point and have already been scanned. Historical confidence data refers to the confidence level corresponding to the above-mentioned acquired scan points, reflecting the spectral signal quality of each historical scan point.

[0040] It should be noted that the predicted thickness value is based on historical data, utilizing the strong spatial and temporal correlation of line scan data to extrapolate the theoretical thickness value of the current scan point through a predictive model. In the online production process of transparent films, because the surface of the transparent film is continuous and moves smoothly, the thickness of the current scan point is highly correlated with the height of the previously scanned scan point, and also with the height of the corresponding scan point in the previous scan line. Therefore, this collected historical data can be used to reasonably predict the thickness of the current scan point.

[0041] Step S40: Generate the fused thickness value of the scan point based on the original thickness value, the predicted thickness value, and the confidence level of the scan point; It should be noted that the fused thickness value is the result of dynamically weighting and fusing the original thickness value and the predicted thickness value based on the confidence level. For example, the formula for calculating the fused confidence level is: ; in, The fusion thickness value of the scan points. The original thickness value of the scanned point. The predicted thickness value for the scan point. This represents the confidence level of the scan point, ranging from 0 to 1. The confidence level reflects the reliability of the original thickness value of the current scan point—a high confidence level indicates good spectral signal quality and reliable original thickness value, while a low confidence level indicates that the scan point may be affected by interference such as vibration defocusing or dust obstruction, making the original thickness value unreliable.

[0042] As can be seen from the above formula, during the fusion process, when the confidence of the scan point is high (tending to 1), the fused thickness value is closer to the original thickness value, preserving the true measurement details to the greatest extent. When the confidence of the scan point is low (tending to 0), the fused thickness value is closer to the predicted thickness value. The current point is filled by inferring the trend of historical data in the spatiotemporal neighborhood, effectively suppressing random interference such as vibration.

[0043] Understandably, vibration interference often causes instantaneous defocusing, resulting in a sharp drop in the quality of the interference spectral signal and a corresponding decrease in confidence. Through this confidence-driven dynamic weighted fusion, this step can automatically switch to a predicted value-dependent mode when confidence decreases and automatically switch back to a mode dependent on the original measurement value when confidence returns to normal. This achieves adaptive suppression of random noise without compromising the true defect morphology, providing a stable and reliable thickness data foundation for defect identification in step S50.

[0044] Step S50: Based on the original thickness value, predicted thickness value, and spatial distribution characteristics between the scan point and other scan points, determine the classification of the scan point. The classification includes real defect points, random noise points, and normal points. It should be noted that after generating the fusion thickness value in step S40, this step performs defect identification and classification for each scan point. The classification is based on two aspects: first, the magnitude of the difference between the original thickness value and the predicted thickness value of the scan point, that is, whether the measured value of the point deviates significantly from the theoretical value calculated based on historical trends; second, the spatial distribution relationship between the scan point and other scan points, that is, whether the abnormal points are isolated or continuously distributed within the two-dimensional scanning area.

[0045] While the dynamic weighted fusion in step S40 effectively suppresses random noise, the original thickness values ​​for genuine minor defects (such as scratches and pinholes) represent actual thickness abrupt changes and should not be smoothed. Similarly, for individual low-confidence measurement jumps caused by vibration interference, the fusion step has already filled them with predicted values. The purpose of this step is to further distinguish between "random noise suppressed by the fusion step" and "genuine defects that need to be retained and identified," providing a classification basis for the targeted correction in step S60.

[0046] It should be noted that normal points refer to scan points where the difference between the original thickness value and the predicted thickness value is small and falls within the normal thickness fluctuation range. These points correspond to normal areas on the surface of the transparent film, and their fused thickness value is the final result, requiring no additional correction.

[0047] It should be noted that true defects refer to scan points where the original thickness value differs significantly from the predicted thickness value and are spatially continuous. These points correspond to real physical defects on the transparent film, such as scratches, pinholes, bubbles, and foreign particles. Because defects have continuous physical extension in space, they form a continuous adjacent abnormal region on the scanned image, rather than an isolated pixel.

[0048] It should be noted that random noise points refer to scan points where the original thickness value differs significantly from the predicted thickness value, but are spatially isolated. These points are usually caused by random factors such as electronic noise and instantaneous vibration jumps, and have no spatial continuity, appearing as one or a few isolated abnormal pixels.

[0049] Step S60: Based on the classification of scan points, the fusion thickness value and confidence level of the scan points are corrected to determine the final confidence level and final thickness value of the scan points; It should be noted that the final confidence score is obtained by correcting the original confidence score based on the classification results of the scan points.

[0050] It should be noted that the final thickness value is obtained by correcting the fused thickness value after combining the classification results of the scan points and the final confidence level.

[0051] It should be noted that in step S50, each scan point has been classified into normal points, real defect points, or random noise points based on the residual size and spatial distribution characteristics. Different categories of scan points have different data reliability and processing requirements, so targeted correction strategies are needed to generate the final confidence level and final thickness value.

[0052] For scan points classified as normal, the difference between their original thickness value and predicted thickness value is small, falling within the normal thickness fluctuation range, and the fused thickness value is already a reliable result. Therefore, the fused thickness value and confidence level of this scan point are not modified and are directly output as the final thickness value and final confidence level.

[0053] For scan points classified as real defects, the original thickness value differs significantly from the predicted thickness value, and this difference is spatially continuous, indicating that the point corresponds to a real physical defect on the transparent film. Since the fusion process in step S40 may smooth the defect morphology to some extent, in order to ensure that the complete features of the defect are not lost, the fused thickness value and confidence level of the scan point need to be corrected accordingly, so that the final result is closer to the original measurement value, thus preserving the true morphology of the defect.

[0054] For scan points classified as random noise points, although the original thickness value differs significantly from the predicted thickness value, this difference is spatially isolated. This is due to measurement jumps caused by random factors such as electronic noise or transient vibrations, rather than anomalies in the transparent film itself. To eliminate the interference of such noise on the final thickness map, the fused thickness value and confidence level of the scan point need to be corrected accordingly, making the final result closer to the predicted value, thus effectively suppressing random noise.

[0055] Understandably, this step achieves differentiated processing of different types of scan points through targeted correction based on classification results. Simultaneously, the corrected final confidence level provides process engineers with an intuitive data quality label—which can be used to distinguish reliable measurement areas, confirmed defect areas, and algorithm-filled areas, thereby providing highly reliable detection data with quality labels for process control in transparent film production lines.

[0056] In this embodiment, steps S10-S60 achieve dynamic evaluation of the reliability of measurement data through confidence perception, thereby overcoming the defect of fixed parameter filtering in the prior art that cannot distinguish signal quality differences. It dynamically achieves the optimal balance between noise reduction and fidelity preservation, and finally outputs high-precision thickness values, confidence distribution, and reliable defect classification results simultaneously. This provides the production line with high-quality online detection data that has both anti-interference capability and defect discrimination capability, reduces the missed detection rate of minor defects, and solves the technical problem that it is impossible to simultaneously effectively suppress random noise and accurately retain real defects in the online detection of transparent film thickness under production line vibration interference.

[0057] In one possible implementation, step S20 may include steps S21-S24: Step S21: Determine the wavenumber domain data of the scanning points based on the interference spectral data of the scanning points; It should be noted that while interference spectral data is typically acquired in the wavelength domain, the frequencies of the interference signals from transparent thin films are uniformly distributed in the wavenumber domain, which is more advantageous for subsequent frequency domain analysis. Therefore, the wavelength domain data is mathematically transformed to the wavenumber domain, where the wavenumber is the reciprocal of the wavelength. This transformation ensures that subsequent film thickness analysis algorithms can accurately extract the frequency components related to the thickness of the transparent thin film.

[0058] Step S22: The wavenumber domain data of the scanning point is processed using a film thickness analysis algorithm to determine the original thickness value of the scanning point. The film thickness analysis algorithm includes fast Fourier transform or phase unenvelope algorithm. It should be noted that the Fast Fourier Transform (FFT) obtains the spectrum by performing a Fourier transform on the wavenumber domain data, locates the peak frequency corresponding to the thickness of the transparent film in the spectrum, and then converts this frequency into the physical thickness according to the principle of transparent film interference. The phase unenvelope algorithm extracts the phase information of the interference signal, combines it with the optical model of the transparent film, and uses an iterative optimization method to calculate the thickness value. In practical implementation, a suitable analytical algorithm can be selected based on factors such as the thickness range of the transparent film to be measured, the measurement accuracy requirements, and the signal-to-noise ratio.

[0059] Step S23: Calculate the interference modulation index of the scanning points based on the interference spectral data of the scanning points; It should be noted that the interference modulation level is the core indicator for measuring the strength of the interference signal and directly reflects the quality of the spectral signal.

[0060] Interference modulation can be calculated using a time-domain sliding window method. Within the effective spectral range of the wavenumber domain data, a window is slid across the spectrum, and the maximum and minimum values ​​are found within each window. The difference between these two values ​​is then divided by their sum to obtain the local modulation intensity. The specific formula is as follows: ; in, and These represent the maximum and minimum values ​​of the interference signal within the window, respectively. The value ranges from 0 to 1, and the larger the value, the stronger the contrast of the interference fringes in the window.

[0061] Alternatively, during the Fast Fourier Transform (FFT), the zero-frequency component amplitude and the peak amplitude within the effective frequency band can be extracted, and the ratio of the peak amplitude to the zero-frequency amplitude can be used as the interference modulation index. The specific formula is as follows: ; in, The DC background component represents the interference spectrum. This represents the amplitude of the interference frequency component corresponding to the thickness of the transparent film. This ratio... The larger the value, the more significant the modulation of the interference signal against the DC background, and the better the signal quality.

[0062] Step S24: The interference modulation of the scan points is processed using a piecewise linear mapping strategy to determine the confidence level of the scan points. The piecewise linear mapping strategy includes threshold calibration and / or weight calculation.

[0063] It should be noted that piecewise linear mapping is a specific method for converting the interferometric modulation intensity into a built-in confidence value within the range of 0 to 1. During the threshold calibration stage, the system pre-determines the lower threshold for effective interference and the upper threshold for ideal interference. The lower threshold corresponds to the system noise floor level; when the modulation intensity is below this value, it is determined that no effective interference has occurred at that scan point, such as in a shadowed, severely out-of-focus, or sample-free area. The upper threshold corresponds to the modulation level of the standard sample at its optimal focal plane; when the modulation intensity is above this value, the interference signal quality at that scan point is considered excellent.

[0064] During the weight calculation phase, confidence level Determined according to the following rules: ; in, For the interference modulation determined in step S23, and In the threshold calibration stage, the system predetermines the lower threshold for effective interference and the upper threshold for ideal interference. When the interference modulation degree is lower than or equal to the lower threshold, the confidence level is 0, indicating that the interference modulation degree of that scan point is lower than the noise floor, and the measurement value is completely unreliable. When the interference modulation degree is higher than or equal to the upper threshold, the confidence level is 1, indicating that the interference modulation degree of that scan point reaches or exceeds the ideal level, and the measurement value is completely reliable. When the interference modulation degree is between the two, the confidence level increases linearly with the increase of the modulation degree. The closer the modulation degree is to the upper threshold, the closer the confidence level is to 1. This mapping strategy allows the confidence level to intuitively reflect the gradual process from "completely unreliable" to "completely reliable," providing a continuously adjustable weight basis for the dynamic weighted fusion in step S40.

[0065] In this embodiment, steps S21 to S24 enable the simultaneous extraction of the original thickness value and its confidence level from the interferometric spectral data. The original thickness value is calculated through wavenumber domain transformation and analytical solution of the film thickness, while the confidence level is generated through interferometric modulation calculation and piecewise linear mapping. This ensures that each scanning point not only obtains a measurement result indicating "how much thickness" but also simultaneously obtains a quality label indicating "how accurate the measurement is." This provides a crucial basis for the dynamic weighted fusion in step S40, overcoming the deficiency in existing technologies where fixed-parameter filtering cannot distinguish signal quality differences.

[0066] In one possible implementation, step S30 may include steps S31-S34: Step S31: Based on the collected historical raw thickness data, determine the substrate component of the transparent film. The substrate component is used to characterize the amount of macroscopic curling change of the transparent film substrate. It should be noted that transparent films are typically attached to flexible substrates, which themselves exhibit macroscopic wavy deformations or curling curvatures of several micrometers, while the thickness of the transparent film coating is usually in the range of several to tens of micrometers. If the original thickness data, which includes substrate deformation, is directly used for prediction, the prediction model will be dominated by the low-frequency general trend of the substrate and will be insensitive to the minute defects of the transparent film itself. Therefore, by performing low-pass filtering or polynomial fitting on the historical original thickness data within the spatiotemporal neighborhood window, the substrate component representing the slow undulations of the substrate can be extracted, thus separating macroscopic deformation from microscopic thickness changes.

[0067] Step S32: Calculate the difference between the historical original thickness value data and the substrate component, as the historical high-frequency change component of the transparent film; It should be noted that subtracting the substrate component from the historical raw thickness data results in a historical high-frequency variation component that removes the macroscopic undulation information of the substrate while retaining the microscopic thickness variation of the transparent film and high-frequency measurement noise. This calculation can be expressed as: ; in, This represents the original thickness value of each historical scan point within the spatiotemporal neighborhood window. The basis components at the corresponding positions determined in step S31. This represents the historical high-frequency variation component obtained after substrate stripping. Subsequent predictive modeling is performed only on this high-frequency component, allowing the predictor to focus on the thickness variation of the transparent film itself.

[0068] Step S33: Based on historical confidence data, the historical high-frequency change components are weighted, and the weighted historical high-frequency change components are used to predict the predicted high-frequency change components of the scan points. It should be noted that weighting refers to using historical confidence levels as weights to assign different importance to the historical high-frequency variation components of each scan point within the spatiotemporal neighborhood. Scan points with high confidence levels have a larger contribution from their historical high-frequency variation components during prediction; the contribution of scan points with low confidence levels is automatically suppressed. Through this dynamic weighting, even in areas where some scan points are affected by vibration, the predicted value can be mainly inferred from the data of high-confidence points in the neighborhood, achieving adaptive denoising.

[0069] Step S34: Calculate the sum of the predicted high-frequency variation component and the substrate component of the scanning point, and use it as the predicted thickness value of the scanning point.

[0070] It should be noted that after obtaining the predicted high-frequency variation component, it is superimposed with the substrate component separated in step S31 to obtain the predicted thickness value of the scanning point: ; in, The base component at the scan point determined in step S31. The predicted high-frequency change components obtained in step S33 The predicted thickness value for a given scan point.

[0071] This "stripping-prediction-overlay" process ensures that the predicted values ​​closely follow the macroscopic contours of the substrate while also keenly reflecting the microscopic thickness changes of the transparent film itself.

[0072] In this embodiment, thickness prediction based on historical data is achieved through steps S31 to S34. By decoupling the substrate and transparent film, the interference of substrate macroscopic deformation on the prediction model is eliminated, allowing the predictor to focus on the thickness change of the transparent film itself. Through dynamic weighting based on confidence level, the predicted value remains stable and reliable even in vibration-affected areas. Thus, a high-quality theoretical reference value is provided for the dynamic weighted fusion in step S40—when the confidence level of the scanning point decreases due to vibration interference, the predicted value can serve as a reliable substitute value, avoiding jumps in the thickness map.

[0073] In one feasible implementation, step S33 may include steps S331-S334: Step S331: Using the Kalman filter algorithm, the prior high-frequency change components of the scanning point are obtained based on the historical high-frequency change components. It should be noted that the prior high-frequency variation components are the output of the Kalman filter prediction step. The Kalman filter defines the state vector as the true state to be estimated at the scan point, i.e., the true high-frequency variation components: ; in, This represents the true high-frequency variation component at the k-th scan point. This represents the rate of change of the true high-frequency variation component along the scanning direction. Kalman filtering, based on historical high-frequency variation components, uses the state equation to extrapolate the prior state of the current scanning point from the optimal state estimate of the previous scanning point, and extracts the prior high-frequency variation component from it. This prior high-frequency variation component is a preliminary prediction of the true high-frequency variation component, representing an estimate extrapolated from historical high-frequency variation components before obtaining the current measurement value.

[0074] Step S332: Determine the observation noise covariance of the Kalman filter algorithm based on historical confidence data, and determine the process noise covariance of the Kalman filter algorithm based on the degree of change of historical high-frequency variation components; It should be noted that the Kalman filter algorithm requires the use of state equations and observation equations. The state equations are: ; in, Let k be the state vector of the k-th scan point. Let F be the state vector of the (k-1)th scan point, and let F be the state transition matrix, which is set according to the scan speed (e.g., uniform speed model). This is process noise, and its statistical characteristics are described by the process noise covariance.

[0075] The observation equation is: ; in, Here, C represents the current measured value of the high-frequency changing component, and C is the observation matrix used to extract the high-frequency changing component from the state vector. The statistical properties of the observation noise are described by the observation noise covariance.

[0076] The observation noise covariance is determined by historical confidence data. Historical confidence data reflects the spectral signal quality at each historical scan point. When historical confidence data indicates poor spectral quality, the observation noise covariance is large, indicating that the current measurement value is less reliable.

[0077] The process noise covariance is determined by the drastic changes in historical high-frequency components. When the drastic changes in historical high-frequency components are small, the process noise covariance is small, indicating that the prediction model is more reliable; when a sudden change is detected in historical high-frequency components, the process noise covariance is increased to allow for rapid state transitions.

[0078] Step S333: Based on the prior high-frequency variation component, observation noise covariance, and process noise covariance, the predicted high-frequency variation component of the scanning point is determined by the Kalman filter algorithm. It should be noted that the Kalman filter update requires integrating the prior high-frequency variation components and the current high-frequency variation components, calculating the Kalman gain based on the observation noise covariance and the process noise covariance, and then determining the posterior state estimate. The thickness component in the posterior state is the predicted high-frequency variation component, which is the optimal estimate given by the filter after integrating model predictions and actual measurements.

[0079] Step S334, wherein the observation noise covariance is used to adjust the trust ratio of the Kalman filter algorithm for the historical high-frequency change components, so as to achieve weighting of the historical high-frequency change components; the process noise covariance is used to adjust the response speed of the Kalman filter algorithm to the sudden change of the historical high-frequency change components. When a sudden change of the historical high-frequency change components is detected, the process noise covariance is increased to enable the Kalman filter algorithm to quickly track the sudden change of the historical high-frequency change components.

[0080] It should be noted that the magnitude of the observation noise covariance determines the confidence allocation of the Kalman filter algorithm for historical high-frequency variation components and the current measurement value in the update step. When the observation noise covariance is large, the Kalman filter algorithm has a higher confidence ratio for historical high-frequency variation components, and the final estimate is closer to the prior prediction value, thus achieving weighting of historical high-frequency variation components—scan points with poor quality indicated by historical confidence data are automatically marginalized during the prediction process.

[0081] When the process noise covariance is large, the Kalman filter algorithm is more sensitive to the current measurement value, thus enabling the Kalman filter algorithm to respond quickly to sudden changes in historical high-frequency components, achieve rapid tracking of sudden changes, and ensure that the thickness changes of real defects are not smoothed out as noise.

[0082] In another feasible implementation, step S330 can also be achieved using local polynomial fitting. Specifically, the historical high-frequency variation components within the spatiotemporal neighborhood window are used as the data to be fitted. The confidence level of each historical scan point is used as a weighting coefficient, and weighted least squares fitting is performed on the data within the window to obtain a low-order polynomial surface. This fitted surface is then used to interpolate or extrapolate the current scan point position to obtain the predicted high-frequency variation components. In this method, historical scan points with high confidence levels contribute significantly to the fitting process, while the contributions of historical scan points with low confidence levels are automatically suppressed, thus achieving a dynamic weighting effect based on confidence levels.

[0083] Alternatively, the prediction model can be replaced by a lightweight convolutional neural network or a long short-term memory network, or other deep learning models. By pre-collecting good-quality data from historical production processes, using historical high-frequency variation components and historical confidence data as input features, and the corresponding actual high-frequency variation components as labels, the network model is trained offline to learn the complex nonlinear spatiotemporal mapping relationship of transparent film thickness variation. During online detection, the historical high-frequency variation components and historical confidence data within the current spatiotemporal neighborhood window are input into the trained network model to obtain the predicted high-frequency variation components for the current scanning point.

[0084] It is understandable that, regardless of whether Kalman filtering, local multinomial fitting, or deep learning models are used, the core of this implementation method lies in using historical confidence data to dynamically weight historical high-frequency change components, so that the predicted value can remain stable in the vibration interference area based on high-confidence historical data, providing a reliable theoretical reference value for the dynamic weighted fusion in step S40.

[0085] In this embodiment, the Kalman filter algorithm is applied to the prediction of high-frequency variation components through steps S331 to S334. On the one hand, the observation noise covariance is dynamically adjusted using historical confidence data, so that the confidence ratio of the Kalman filter algorithm for historical high-frequency variation components changes with the confidence level—scanning points with low confidence are automatically marginalized during the prediction process, thereby achieving dynamic weighting of historical high-frequency variation components. This ensures that the predicted value remains stable in the vibration interference area based on high-confidence historical data. On the other hand, the process noise covariance is dynamically adjusted based on the drastic changes in historical high-frequency variation components, enabling a rapid response when a sudden change in thickness is detected, ensuring that the real defect is not smoothed out. Thus, an adaptive balance between "dynamic denoising" and "defect preservation" is achieved.

[0086] Example 2 Based on the first embodiment of this application, in the second embodiment of this application, the content that is the same as or similar to that in the first embodiment described above can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 2The online thickness detection method for transparent films also includes steps S51-S54 in step S50: Step S51: Calculate the absolute value of the difference between the original thickness value and the predicted thickness value of the scan point, and use it as the fusion residual of the scan point; It should be noted that the fusion residual reflects the degree to which the actual measured value at a scan point deviates from the theoretical value of the prediction model. The formula for calculating the fusion residual is: ; in, For the fusion residual of the scan points, The original thickness value of the scanned point. The difference between the two values ​​is the predicted thickness value of the scan point. The larger the absolute value of the difference, the more likely there is an anomaly at the scan point. This could be a sudden change in thickness caused by a real physical defect, or a measurement jump caused by vibration interference.

[0087] Step S52: When the fusion residual of the scan point is less than the preset residual threshold, the scan point is classified as a normal point. It should be noted that when the fusion residual is less than the preset residual threshold, it indicates that the deviation between the original thickness value and the predicted thickness value is within the normal fluctuation range, and this scanning point is directly classified as a normal point. The specific value of the preset residual threshold can be set according to the measurement accuracy requirements of the production line and the tolerance range of the transparent film thickness. For example, the preset residual threshold... A dynamic threshold can be used, determined based on the statistical characteristics of the fused residuals within a local window: ; in, This represents the mean of the fused residuals within the local window. The standard deviation of the fused residuals within the local window. The threshold is a preset confidence factor (e.g., n=3). This dynamic thresholding method allows the judgment criteria to adapt to the local statistical characteristics of different regions, automatically relaxing the threshold in areas with high noise levels and automatically tightening the threshold in areas with low noise levels, thus exhibiting better robustness compared to a fixed threshold.

[0088] Step S53: When the fusion residual of the scan point is greater than or equal to the preset residual threshold, the scan point is determined as a candidate anomaly point; It should be noted that candidate outliers refer to scan points whose fusion residuals exceed a preset residual threshold, but whose specific anomaly type has not yet been determined. This step only completes the initial screening based on the size of the single-point residuals, including all points that "appear abnormal" in the candidate range. The final classification of candidate outliers requires further discrimination based on their spatial distribution characteristics.

[0089] Step S54: Based on the spatial distribution characteristics between the scan points and other scan points, determine the classification of candidate anomalies.

[0090] It should be noted that the spatial distribution characteristics between scan points refer to the spatial clustering of scan points within the two-dimensional scanning area. Candidate anomalies can be classified into anomalies caused by real physical defects and anomalies caused by random noise. Real physical defects form continuous, adjacent anomaly regions during scanning, exhibiting spatial continuity; while random noise is spatially isolated and scattered. Therefore, analyzing the spatial connectivity of candidate anomalies can effectively distinguish between these two different types of anomalies.

[0091] In this embodiment, steps S51 to S54 achieve a two-level classification of scan points from "single-point residual screening" to "spatial feature discrimination". First, by comparing the fused residual with a preset threshold, candidate anomaly points are quickly screened out; then, based on spatial distribution characteristics, the candidate anomaly points are further discriminated, and finally, each scan point is classified as a normal point, a real defect point, or a random noise point. This effectively distinguishes between "random noise caused by vibration" and "real physical defects such as scratches and pinholes", providing a reliable classification basis for the targeted correction in step S60.

[0092] In one feasible implementation, step S54 may include steps S541-S543: Step S541: Based on the spatial distribution characteristics between the scan points and other scan points, spatial connectivity domains are marked for candidate anomaly points to determine the abnormal connected regions to which the candidate anomaly points belong. It should be noted that spatial distribution characteristics refer to the spatial clustering or dispersion of candidate anomalies within a two-dimensional scanning area. Since line scanning acquires data line by line from a continuously moving transparent film, a real physical defect on the film will span multiple adjacent scan lines and spatial locations during the scanning process, appearing as a continuous set of adjacent anomalies in the two-dimensional scan image. In contrast, measurement jumps caused by electronic noise or instantaneous vibrations appear spatially as isolated, scattered single or very few anomalies. Spatial connectivity labeling utilizes this difference in spatial distribution to divide candidate anomalies into regions.

[0093] Specifically, spatial connectivity labeling involves examining the spatial adjacency of each candidate outlier with other scanned points, grouping spatially adjacent candidate outliers into the same connected region. Adjacent candidate outliers are assigned the same anomalous connected region label, while non-adjacent candidate outliers are assigned different labels. Through this process, all candidate outliers are divided into several unconnected anomalous connected regions. Each anomalous connected region consists of a set of spatially consecutive candidate outliers, representing a spatially independent cluster of outliers.

[0094] Step S542: When the area of ​​the abnormal connected region is less than the preset area threshold, the candidate abnormal points in the abnormal connected region are classified as random noise points. It should be noted that the area of ​​an abnormal connected region represents the number of candidate anomalies contained within that region, reflecting the spatial distribution scale of the anomaly. For example, the number of candidate anomalies within a candidate anomaly region and a preset threshold number can also be used to distinguish between different abnormal connected regions.

[0095] It should be noted that the preset area threshold is a pre-defined critical value that distinguishes random noise from real defects. Its value can be set according to the required detection resolution and the minimum acceptable defect size. When the area of ​​the abnormal connected region is smaller than the preset area threshold, it indicates that the abnormal points within that region are spatially isolated and scattered, lacking the continuous expansion characteristic of a real physical defect, and are therefore classified as random noise points. The anomalies corresponding to random noise points are measurement jumps caused by random factors such as electronic noise or instantaneous vibrations, and are not due to defects in the transparent film itself.

[0096] Step S543: When the area of ​​the abnormal connected region is greater than or equal to the preset area threshold, the candidate abnormal points in the abnormal connected region are classified as real defect points.

[0097] It should be noted that when the area of ​​the connected abnormal region is greater than or equal to the preset area threshold, that is, when there are no fewer than a certain number of candidate abnormal points in the connected region, it indicates that the abnormal points in the region have continuous spatial expansion and their distribution scale conforms to the morphological characteristics of real physical defects (such as scratches, pinholes, and bubbles). Therefore, they are determined to be real defect points. The abnormality corresponding to the real defect point is a real physical defect that exists on the transparent film. It needs to be retained and marked in subsequent steps to ensure that it is not smoothed out as noise.

[0098] In this embodiment, through steps S541 to S543, candidate anomalies are first grouped into different abnormal connected regions by spatial connectivity marking. Then, by comparing the area of ​​each abnormal connected region with a preset area threshold, the candidate anomalies are finally classified—isolated small regions with an area less than the threshold are determined as random noise points, and continuous large regions with an area greater than or equal to the threshold are determined as real defect points. This discrimination method based on spatial distribution characteristics achieves accurate differentiation between "random noise caused by vibration" and "real defects of the transparent film itself," providing a reliable classification basis for the targeted correction in step S60.

[0099] In one feasible implementation, step S541 may include steps S5411-S5413: Step S5411: Check whether there are any marked candidate anomalies within the preset neighborhood of the candidate anomalies; It should be noted that the preset neighborhood can refer to either the eight directly adjacent positions around the current scan point, including adjacent points in the horizontal, vertical, and diagonal directions (i.e., an 8-neighborhood), or it can refer to the four directly adjacent positions around the current scan point (i.e., a 4-neighborhood). The specific adjacent points of the 4-neighborhood can be any four adjacent points selected from the eight adjacent points in the 8-neighborhood. Checking whether there are marked candidate anomalies within the preset neighborhood is to determine whether the candidate anomalies are spatially connected to a certain anomaly connectivity region that has already been processed.

[0100] Step S5412: If it exists, mark the candidate outlier with the same outlier connected region label as the already marked candidate outlier. It should be noted that the abnormal connected region label is used to identify the connected region to which each candidate anomaly belongs. When there is a labeled neighboring candidate anomaly within the preset neighborhood of a candidate anomaly, it means that the candidate anomaly and the neighboring point belong to the same abnormal connected region. Therefore, the current point is labeled with the same abnormal connected region label as the neighboring point, thereby classifying the point into an existing connected region, realizing the growth and expansion of connected regions.

[0101] Step S5413: If it does not exist, mark the candidate outlier as a new outlier connected region label.

[0102] It should be noted that when there are no labeled candidate anomalies within the preset neighborhood of a candidate anomaly, it means that the point is not connected to any existing connected region and belongs to the starting point of a new connected region. In this case, a new, unused anomaly connected region label is assigned to the point. This new label may be reused by neighboring candidate anomalies in subsequent processing, thereby gradually constructing a complete connected region.

[0103] In this embodiment, steps S5411 to S5413 are used to label the spatial connected regions of candidate anomalies. By checking whether there are already labeled candidate anomalies in the preset neighborhood, candidate anomalies with adjacent relationships are labeled with the same abnormal connected region label, and candidate anomalies without adjacent relationships are labeled with a new abnormal connected region label. This accurately determines the connected region to which each candidate anomaly belongs, providing a reliable spatial partitioning result for subsequent classification based on the area of ​​the connected region.

[0104] For example, please refer to Figure 3 , Figure 3 A partial schematic diagram of the abnormal connected regions for determining each candidate anomaly point in Embodiment 2 of this application is provided. Figure 3 There are three abnormal connected regions within the range. The first abnormal connected region 101 is a special abnormal connected region with only one candidate abnormal point. The second abnormal connected region 104 is the abnormal connected region to which candidate abnormal point 102 belongs. Figure 3 From this, we can see that the second abnormal connected region 104 contains 7 candidate abnormal points. The third abnormal connected region 105 is the abnormal connected region to which candidate abnormal point 103 belongs. Figure 3 From this, we can see that the third abnormal connected region 105 contains two candidate abnormal points.

[0105] For example, the preset area threshold can be the unit area occupied by 5 scan points. Figure 3 In the second abnormal connected region 104, the area is greater than the preset area threshold. Therefore, the candidate abnormal points in the second abnormal connected region 104 are classified as real defect points. The areas of the first abnormal connected region 101 and the third abnormal connected region 105 are less than the preset area threshold. Therefore, the candidate abnormal points in the first abnormal connected region 101 and the third abnormal connected region 105 are classified as random noise points.

[0106] It should be noted that the above examples are only for understanding this application and do not constitute a limitation on the online thickness detection method for transparent films of this application. Any simple modifications based on this technical concept are within the protection scope of this application.

[0107] In one feasible implementation, step S60 may include steps S61-S64: Step S61: When the scan point is classified as a normal point, the final confidence level of the scan point is determined as the confidence level of the scan point, and the final thickness value of the scan point is determined as the fusion thickness value of the scan point. It should be noted that normal points refer to scan points determined in step S50 whose fusion residual is less than a preset residual threshold, and whose original thickness value deviates from the predicted thickness value within the normal fluctuation range. For these scan points, the fusion thickness value generated in step S40 is already a reliable result, and its original confidence level truly reflects the spectral signal quality of that point. Therefore, no additional correction is needed for normal points; their confidence level can be directly used as the final confidence level, and their fusion thickness value can be output as the final thickness value.

[0108] Step S62: When the scan point is classified as a real defect point, the final confidence level of the scan point is determined to be the preset first confidence level, and the final thickness value of the scan point is determined based on the original thickness value, the predicted thickness value and the final confidence level. It should be noted that the preset first confidence level is a pre-set high confidence value used to identify that the point has been confirmed as a real physical defect, possessing high data reliability. When determining the final thickness value, the original thickness value and the predicted thickness value are re-fused using this preset first confidence level as a weight. The underlying calculation logic is consistent with the method used in step S40 to generate the fused thickness value. The weighted fusion formula used in step S40 is a preferred implementation. Since the first confidence level is close to or equal to the highest confidence value, the final thickness value is highly biased towards the original thickness value, thereby completely preserving the original measurement information of the defect and ensuring that the morphology and depth features of the defect are not smoothed by the fusion process in step S40.

[0109] Step S63: When the scan point is classified as a random noise point, the final confidence level of the scan point is determined to be the preset second confidence level, and the final thickness value of the scan point is determined based on the original thickness value, the predicted thickness value and the final confidence level.

[0110] It should be noted that the preset second confidence level is a pre-set low confidence value used to identify the thickness data at that point as an algorithmic estimate based on the predicted value, with low reliability. When determining the final thickness value, the original thickness value and the predicted thickness value are re-fused using this preset second confidence level as a weight. The underlying calculation logic is consistent with the method used in step S40 to generate the fused thickness value. The weighted fusion formula used in step S40 is a preferred implementation. Since the second confidence level is close to or equal to the lowest confidence value, the final thickness value is highly biased towards the predicted thickness value, thereby effectively suppressing the interference of random noise on the final thickness map.

[0111] Step S64, wherein the first confidence level is greater than the second confidence level.

[0112] It should be noted that the first confidence level is a preset higher confidence level value, and the second confidence level is a preset lower confidence level value, with the first confidence level being greater than the second confidence level. This ensures that for real defect points, the final thickness value is closer to the original thickness value; for random noise points, the final thickness value is closer to the predicted thickness value, thus achieving differentiated processing for noise reduction and fidelity preservation.

[0113] In this embodiment, through steps S61 to S64, based on the classification results of step S50, targeted correction strategies are adopted for different categories of scan points. Normal points are not corrected, and the fusion result is directly used; for real defect points, the confidence level is increased to make the final thickness value closer to the original measurement value, ensuring that defect features are not missed; for random noise points, the confidence level is decreased to make the final thickness value closer to the predicted value, eliminating noise interference. Thus, while smoothing random noise, the true physical defects are accurately preserved, achieving an adaptive balance between noise reduction and fidelity.

[0114] This application also provides an online thickness detection device for transparent films; please refer to [reference needed]. Figure 4 The online thickness detection device for transparent films includes: The spectral acquisition module 10 is used to acquire interference spectral data of the scanning points formed during the online scanning of the transparent film; The confidence level determination module 20 is used to determine the original thickness value and the confidence level of the scanning point based on the interference spectral data of the scanning point; The thickness prediction module 30 is used to determine the predicted thickness value of the scanning point based on the collected historical raw thickness value data and historical confidence data. The fusion thickness determination module 40 is used to generate the fusion thickness value of the scan point based on the original thickness value, the predicted thickness value, and the confidence level of the scan point. The classification determination module 50 is used to determine the classification of the scanning points based on the original thickness value, the predicted thickness value, and the spatial distribution characteristics between the scanning points and other scanning points. The classification includes real defect points, random noise points, and normal points. The data correction module 60 is used to correct the fusion thickness value and confidence level of the scan points based on the classification of the scan points, and to determine the final confidence level and final thickness value of the scan points.

[0115] In one feasible implementation, the thickness prediction module 30 is further configured to: Based on the collected historical raw thickness data, the substrate component of the transparent film is determined. The substrate component is used to characterize the amount of macroscopic curling change of the transparent film substrate. The difference between the historical original thickness data and the substrate component is calculated as the historical high-frequency variation component of the transparent film; Based on historical confidence data, the historical high-frequency change components are weighted, and the weighted historical high-frequency change components are used to predict the predicted high-frequency change components of the scan points. The sum of the predicted high-frequency variation component and the substrate component at the scan point is calculated and used as the predicted thickness value of the scan point.

[0116] In one feasible implementation, the thickness prediction module 30 is further configured to: The Kalman filter algorithm is used to obtain the prior high-frequency variation components of the scanning point based on historical high-frequency variation components. The observation noise covariance of the Kalman filter algorithm is determined based on historical confidence data, and the process noise covariance of the Kalman filter algorithm is determined based on the degree of change of historical high-frequency components. Based on the prior high-frequency variation components, observation noise covariance, and process noise covariance, the predicted high-frequency variation components of the scanning points are determined using the Kalman filtering algorithm. Among them, the observation noise covariance is used to adjust the trust ratio of the Kalman filter algorithm for the historical high-frequency change components, so as to achieve weighting of the historical high-frequency change components; the process noise covariance is used to adjust the response speed of the Kalman filter algorithm to the sudden change of the historical high-frequency change components. When a sudden change of the historical high-frequency change components is detected, the process noise covariance is increased to enable the Kalman filter algorithm to quickly track the sudden change of the historical high-frequency change components.

[0117] In one feasible implementation, the classification determination module 50 is further configured to: The absolute value of the difference between the original thickness value and the predicted thickness value of the scan point is calculated as the fusion residual of the scan point; When the fusion residual of the scan point is less than the preset residual threshold, the scan point is classified as a normal point. When the fusion residual of a scan point is greater than or equal to a preset residual threshold, the scan point is determined as a candidate anomaly point. Based on the spatial distribution characteristics between scan points and other scan points, the classification of candidate anomalies is determined.

[0118] In one feasible implementation, the classification determination module 50 is further configured to: Based on the spatial distribution characteristics between scan points and other scan points, spatial connectivity domains are marked for candidate anomalies to determine the abnormal connected regions to which the candidate anomalies belong. When the area of ​​the abnormal connected region is less than the preset area threshold, the candidate abnormal points within the abnormal connected region are classified as random noise points. When the area of ​​the abnormal connected region is greater than or equal to a preset area threshold, the candidate abnormal points within the abnormal connected region are classified as real defect points.

[0119] In one feasible implementation, the classification determination module 50 is further configured to: Check if there are any labeled candidate anomalies within the preset neighborhood of the candidate anomalies; If it exists, the candidate outlier will be marked with the same outlier connected region label as the already marked candidate outlier; If it does not exist, the candidate outlier is marked as a new outlier connected region label.

[0120] In one feasible implementation, the data correction module 60 is further configured to: When a scan point is classified as a normal point, the final confidence level of the scan point is determined as the confidence level of the scan point, and the final thickness value of the scan point is determined as the fusion thickness value of the scan point. When the scan point is classified as a real defect point, the final confidence level of the scan point is determined to be the preset first confidence level, and the final thickness value of the scan point is determined based on the original thickness value, the predicted thickness value, and the final confidence level. When the scan point is classified as a random noise point, the final confidence level of the scan point is determined to be the preset second confidence level, and the final thickness value of the scan point is determined based on the original thickness value, the predicted thickness value, and the final confidence level. The first confidence level is greater than the second confidence level.

[0121] In one feasible implementation, the confidence determination module 20 is further configured to: Based on the interferometric spectral data of the scanning points, determine the wavenumber domain data of the scanning points; The wavenumber domain data of the scanning points is processed using a film thickness analysis algorithm to determine the original thickness value of the scanning points. The film thickness analysis algorithm includes fast Fourier transform or phase unenvelope algorithm. Calculate the interference modulation index of the scanning points based on the interference spectral data of the scanning points; A piecewise linear mapping strategy is used to process the interference modulation of the scan points and determine the confidence level of the scan points. The piecewise linear mapping strategy includes threshold calibration and / or weight calculation.

[0122] The online thickness detection device for transparent films provided in this application, employing the online thickness detection method for transparent films in the above embodiments, can solve the technical problem of effectively suppressing random noise while accurately preserving true defect characteristics when detecting the thickness of transparent films online under the interference of production line vibration. Compared with the prior art, the beneficial effects of the online thickness detection device for transparent films provided in this application are the same as those of the online thickness detection method for transparent films provided in the above embodiments, and other technical features in the online thickness detection device for transparent films are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0123] This application provides an online thickness detection device for transparent films, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, which are executed by the at least one processor to enable the at least one processor to perform the online thickness detection method for transparent films in Embodiment 1 described above.

[0124] The following is for reference. Figure 5 The diagram illustrates a structural schematic of a transparent film online thickness detection device suitable for implementing embodiments of this application. The transparent film online thickness detection device in the embodiments of this application may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 5 The online thickness detection device for transparent films shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments in this application.

[0125] like Figure 5As shown, the online thickness detection device for transparent films may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage device 1003 into a random access memory (RAM) 1004. The RAM 1004 also stores various programs and data required for the operation of the online thickness detection device for transparent films. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to the I / O interface 1006: input devices 1007 including, for example, a touch screen, touchpad, keyboard, mouse, image sensor, microphone, accelerometer, gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), speaker, vibrator, etc.; storage devices 1003 including, for example, magnetic tape, hard disk, etc.; and communication devices 1009. The communication device 1009 allows the online thickness detection device for transparent films to communicate wirelessly or wiredly with other devices to exchange data. Although the figures show online thickness detection devices for transparent films with various systems, it should be understood that implementation or possession of all the systems shown is not required. More or fewer systems may be implemented alternatively.

[0126] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from ROM 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0127] The online thickness detection device for transparent films provided in this application, employing the online thickness detection method for transparent films in the above embodiments, can solve the technical problem of effectively suppressing random noise while accurately preserving true defect characteristics when detecting the thickness of transparent films online under the interference of production line vibration. Compared with the prior art, the beneficial effects of the online thickness detection device for transparent films provided in this application are the same as those of the online thickness detection method for transparent films provided in the above embodiments, and other technical features in this online thickness detection device are the same as those disclosed in the method of the previous embodiment, and will not be repeated here.

[0128] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0129] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0130] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, which are used to execute the online thickness detection method for transparent films in the above embodiments.

[0131] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems or devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0132] The aforementioned computer-readable storage medium may be included in the online thickness detection device for transparent films; or it may exist independently and not be assembled into the online thickness detection device for transparent films.

[0133] The aforementioned computer-readable storage medium carries one or more programs, which, when executed by the transparent film online thickness detection device, cause the transparent film online thickness detection device to: acquire interference spectral data of scanning points formed during the online scanning of the transparent film; Based on the interference spectral data of the scanning points, determine the original thickness value and the confidence level of the scanning points; Based on the collected historical raw thickness data and historical confidence data, the predicted thickness value of the scanning point is determined; The fused thickness value of the scan points is generated based on the original thickness value, the predicted thickness value, and the confidence level of the scan points. Based on the original thickness value, predicted thickness value, and spatial distribution characteristics of the scan points, the classification of the scan points is determined, including real defect points, random noise points, and normal points. Based on the classification of scan points, the fusion thickness value and confidence level of the scan points are corrected to determine the final confidence level and final thickness value of the scan points.

[0134] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a Local Area Network (LAN) or a Wide Area Network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0135] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0136] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0137] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for executing the above-described online thickness detection method for transparent films. This solves the technical problem of effectively suppressing random noise while accurately preserving true defect characteristics during online thickness detection of transparent films under the interference of production line vibration. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the online thickness detection method for transparent films provided in the above embodiments, and will not be repeated here.

[0138] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the online thickness detection method for transparent films as described above.

[0139] The computer program product provided in this application solves the technical problem of effectively suppressing random noise while accurately preserving true defect characteristics when detecting the thickness of transparent films online under the interference of production line vibration. Compared with the prior art, the beneficial effects of the computer program product provided in this application are the same as those of the online thickness detection method for transparent films provided in the above embodiments, and will not be repeated here.

[0140] The above are only some embodiments of this application and do not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.

Claims

1. A method for online thickness detection of a transparent film, characterized in that, The method includes: Acquire interference spectral data of scanning points formed during online scanning of transparent thin films; Based on the interference spectral data of the scanning points, the original thickness value of the scanning points and the confidence level of the scanning points are determined; Based on the collected historical raw thickness data and historical confidence data, the predicted thickness value of the scan point is determined; The fused thickness value of the scan point is generated based on the original thickness value, the predicted thickness value, and the confidence level of the scan point. Based on the original thickness value, the predicted thickness value, and the spatial distribution characteristics of the scan point and other scan points, the classification of the scan point is determined, including real defect points, random noise points, and normal points. Based on the classification of the scan points, the fusion thickness value and confidence level of the scan points are corrected to determine the final confidence level and final thickness value of the scan points.

2. The method as described in claim 1, characterized in that, The step of determining the predicted thickness value of the scan point based on the collected historical raw thickness value data and historical confidence data includes: Based on the collected historical raw thickness data, the substrate component of the transparent film is determined, and the substrate component is used to characterize the amount of macroscopic curl change of the transparent film substrate; The difference between the historical original thickness value data and the substrate component is calculated as the historical high-frequency variation component of the transparent film; Based on the historical confidence data, the historical high-frequency change components are weighted, and the weighted historical high-frequency change components are used to predict the predicted high-frequency change components of the scan point. The sum of the predicted high-frequency variation component and the substrate component of the scan point is calculated and used as the predicted thickness value of the scan point.

3. The method as described in claim 2, characterized in that, The step of weighting the historical high-frequency change components based on the historical confidence data and using the weighted historical high-frequency change components to predict the predicted high-frequency change components of the scan point includes: The Kalman filter algorithm is used to obtain the prior high-frequency change components of the scanning point based on the historical high-frequency change components. The observation noise covariance of the Kalman filter algorithm is determined based on the historical confidence data, and the process noise covariance of the Kalman filter algorithm is determined based on the degree of change of the historical high-frequency variation components. Based on the prior high-frequency variation component, the observation noise covariance, and the process noise covariance, the predicted high-frequency variation component of the scan point is determined by the Kalman filtering algorithm. The observation noise covariance is used to adjust the trust ratio of the historical high-frequency variation component in the Kalman filter algorithm, so as to achieve weighting of the historical high-frequency variation component; the process noise covariance is used to adjust the response speed of the Kalman filter algorithm to the sudden change of the historical high-frequency variation component. When a sudden change of the historical high-frequency variation component is detected, the process noise covariance is increased to enable the Kalman filter algorithm to quickly track the sudden change of the historical high-frequency variation component.

4. The method as described in claim 1, characterized in that, The step of determining the classification of the scan points based on the original thickness value, predicted thickness value, and spatial distribution characteristics between the scan points and other scan points includes: The absolute value of the difference between the original thickness value and the predicted thickness value of the scan point is calculated as the fusion residual of the scan point; When the fusion residual of the scan point is less than a preset residual threshold, the scan point is classified as a normal point. When the fusion residual of the scan point is greater than or equal to a preset residual threshold, the scan point is determined to be a candidate anomaly point; The classification of the candidate anomalies is determined based on the spatial distribution characteristics between the scan points and other scan points.

5. The method as described in claim 4, characterized in that, The step of determining the classification of the candidate anomaly points based on the spatial distribution characteristics between the scan points and other scan points includes: Based on the spatial distribution characteristics between the scan points and other scan points, the candidate anomaly points are spatially connected component marked to determine the abnormal connected region to which the candidate anomaly points belong. When the area of ​​the abnormal connected region is less than a preset area threshold, the candidate abnormal points within the abnormal connected region are classified as random noise points. When the area of ​​the abnormal connected region is greater than or equal to a preset area threshold, the candidate abnormal points within the abnormal connected region are classified as real defect points.

6. The method as described in claim 5, characterized in that, The step of spatially labeling the candidate anomaly points based on the spatial distribution characteristics between the scan points and other scan points, and determining the abnormal connected regions to which the candidate anomaly points belong, includes: Check whether there are any marked candidate anomalies within the preset neighborhood of the candidate anomalies; If it exists, the candidate anomaly point is marked with the same abnormal connected region label as the already marked candidate anomaly point; If it does not exist, the candidate anomaly point is marked as a new anomaly connected region label.

7. The method as described in claim 1, characterized in that, The step of correcting the fusion thickness value and confidence level of the scan points based on the classification of the scan points, and determining the final confidence level and final thickness value of the scan points includes: When the scan point is classified as a normal point, the final confidence level of the scan point is determined as the confidence level of the scan point, and the final thickness value of the scan point is determined as the fusion thickness value of the scan point; When the scan point is classified as a real defect point, the final confidence level of the scan point is determined to be a preset first confidence level, and the final thickness value of the scan point is determined based on the original thickness value, the predicted thickness value, and the final confidence level. When the scan point is classified as a random noise point, the final confidence level of the scan point is determined to be a preset second confidence level, and the final thickness value of the scan point is determined based on the original thickness value, the predicted thickness value, and the final confidence level. Wherein, the first confidence level is greater than the second confidence level.

8. The method as described in claim 1, characterized in that, The step of determining the original thickness value and the confidence level of the scanning point based on the interferometric spectral data of the scanning point includes: Based on the interferometric spectral data of the scanning points, the wavenumber domain data of the scanning points are determined; The wavenumber domain data of the scanning points are processed using a film thickness analysis algorithm to determine the original thickness value of the scanning points. The film thickness analysis algorithm includes fast Fourier transform or phase unenvelope algorithm. Based on the interference spectral data of the scanning points, the interference modulation index of the scanning points is calculated; The interference modulation of the scan points is processed using a piecewise linear mapping strategy to determine the confidence level of the scan points. The piecewise linear mapping strategy includes threshold calibration and / or weight calculation.

9. A transparent film online thickness detection device, characterized in that, The online thickness detection device for the transparent film includes: The spectral acquisition module is used to acquire interference spectral data of the scanning points formed during the online scanning of transparent films; The confidence level determination module is used to determine the original thickness value of the scanning point and the confidence level of the scanning point based on the interference spectral data of the scanning point; The thickness prediction module is used to determine the predicted thickness value of the scanning point based on the collected historical raw thickness value data and historical confidence data. The fusion thickness determination module is used to generate the fusion thickness value of the scan point based on the original thickness value, the predicted thickness value, and the confidence level of the scan point. The classification determination module is used to determine the classification of the scan point based on the original thickness value, the predicted thickness value, and the spatial distribution characteristics between the scan point and other scan points. The classification includes real defect points, random noise points, and normal points. The data correction module is used to correct the fusion thickness value and confidence level of the scan points based on the classification of the scan points, and to determine the final confidence level and final thickness value of the scan points.

10. A storage medium, characterized in that, The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, it implements the steps of the online thickness detection method for transparent films as described in any one of claims 1 to 8.

Images