Mobile phone glass cover plate flatness detection method and system

Abstract

The invention relates to the technical field of optical detection, and particularly discloses a mobile phone glass cover plate flatness detection method and system, and the method comprises the steps: projecting a multiband composite light beam to to-be-detected glass through a confocal optical system, and synchronously receiving a mixed spectrum signal comprising multi-layer interface reflection; generating a key characteristic signal through compressed sampling and spectrum channel optimization; analyzing and separating reflection components of each layer in real time through matching with a pre-established template library; and reconstructing a surface three-dimensional shape based on a surface layer reflection signal, further extracting macroscopic warping degree and microdefect features, constructing a normalized and weighted comprehensive feature vector, and inputting the comprehensive feature vector into a deep neural network model to realize flatness intelligent evaluation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of optical detection, and in particular to a method and system for detecting flatness of a mobile phone glass cover plate. BACKGROUND

[0002] With the rapid development of consumer electronics such as smart phones, tablet computers and the like towards thinness, high screen ratio and multi-functional integration, the appearance quality and structural precision of mobile phone glass cover plates, as the core components of human-computer interaction, are increasingly demanding. Flatness, as a key indicator for measuring the quality of glass cover plates, directly affects the screen display effect, touch sensitivity, and the sealing and aesthetics of the entire machine assembly. Small macro warping or microscopic surface defects (such as scratches and pits) can all lead to display distortion, touch failure or assembly stress concentration, thereby affecting product yield and user experience. Therefore, it has become an important technical requirement in the modern optoelectronic manufacturing field to achieve high-precision, high-efficiency and non-contact detection of the flatness of mobile phone glass cover plates.

[0003] The prior art has the following disadvantages:

[0004] In the detection of the surface topography of transparent multi-layer composite materials based on the principle of spectral confocal, the sample to be measured (mobile phone glass cover plate) has a multi-layer transparent medium structure, and the difference in refractive index between the layers causes multiple reflections and transmissions of the incident measurement light beam at the interface. When the sensor receiving end collects the reflected spectrum, the reflected light signal from the lower interface will be mixed with the reflected light signal from the target surface in the spectral domain, resulting in aliasing, which makes the system unable to accurately identify the characteristic wavelength peak corresponding to the real surface, and thus false height information is generated. SUMMARY

[0005] The present application aims to provide a method and system for detecting the flatness of a mobile phone glass cover plate to solve the problems in the background.

[0006] The object of the present application can be achieved by the following technical solutions:

[0007] A method for detecting the flatness of a mobile phone glass cover plate, comprising the following steps:

[0008] S1. Projecting a multi-band composite light beam onto the glass cover plate to be measured by a confocal optical detection system, and simultaneously receiving a mixed spectrum signal containing multiple interface reflections;

[0009] S2. Compressively sampling the mixed spectrum signal, selecting a spectrum channel according to the refractive index characteristics of the glass material, and generating an optimized sampling signal;

[0010] S3. Real-time analysis of the optimized sampling signal to separate the reflection components of each layer of interface;

[0011] S4, calculate the three-dimensional topography data of the glass cover plate surface for the separated surface reflection component;

[0012] S5, extract the macro warping degree feature and the micro surface defect feature of the glass cover plate based on the three-dimensional topography data;

[0013] S6, construct the macro warping degree feature and the micro surface defect feature into a comprehensive feature vector, and use the comprehensive feature vector as the input of the mobile phone glass cover plate detection model, and output the flatness evaluation result of the mobile phone glass cover plate.

[0014] As a further scheme of the present application, the received mixed spectrum signal containing multi-layer interface reflection specifically includes:

[0015] An ultrashort pulse laser source is used to generate a light beam covering the visible light to near-infrared wave band, and three characteristic wavelength sub-bands are selected by an acousto-optic tunable filter, wherein the first sub-band wavelength corresponds to the glass surface reflectivity, the second sub-band wavelength is sensitive to the interface pollution, and the third sub-band wavelength has the maximum material penetration depth;

[0016] The light beams of the three characteristic wavelength sub-bands are combined by a dispersive prism to form a coaxial composite detection light beam, and a tomographic detection spot is formed after focusing by an aperture objective lens;

[0017] The phase of the reflected light is modulated by a high-speed rotating spiral phase plate, so that the interface reflection light at different depths produces a characteristic vortex phase distribution;

[0018] The modulated reflected light signal is synchronously captured, and the reflection intensity distribution of each layer interface is demodulated to obtain a mixed spectrum signal containing multi-layer interface reflection characteristics.

[0019] As a further scheme of the present application, the generated optimized sampling signal specifically includes:

[0020] A glass material spectrum feature library is established, and corresponding three characteristic wavelength intervals are extracted from the glass material spectrum feature library, wherein the first interval corresponds to the surface reflection peak value, the second interval is sensitive to the internal interface change, and the third interval has the maximum penetration ability;

[0021] The mixed spectrum signal is filtered, and only the light signals of the three characteristic wavelength intervals are transmitted;

[0022] The filtered light signal is converted into an electrical signal, and environmental noise interference is eliminated by using a digital signal processing technology, and finally an optimized sampling signal containing key optical characteristics is generated.

[0023] As a further scheme of the present application, the real-time analysis of the optimized sampling signal, and the separation of the reflection component of each layer interface, specifically includes:

[0024] Spectrum form characteristics and time domain response characteristics of interface reflection signals of different material combinations of the mobile phone glass cover plate are stored in advance as a multi-layer interface reflection feature template library;

[0025] The optimized sampling signal is matched with the interface reflection features in the feature template library, and each interface reflection component contained in the optimized sampling signal is matched;

[0026] The surface layer reflection component, the intermediate layer reflection component and the bottom layer reflection component are separated from the optimized sampling signal in turn, wherein the remaining signal is updated and matched again after each decomposition;

[0027] The separation effect of each layer reflection component is analyzed and verified, the residual energy of each layer reflection component is calculated, and when the residual energy is lower than the set threshold, the separation is confirmed to be completed, and the reflection components of each layer interface are output.

[0028] As a further scheme of the present application, the acquisition process of the three-dimensional topographic data is:

[0029] The reflection signal is analyzed in time and frequency, and the arrival time difference and frequency domain feature parameters of the reflection signal are extracted;

[0030] The optical path propagation path is reconstructed, and the actual height value of each measurement point is calculated in combination with the focusing characteristics of the objective lens;

[0031] The height data of adjacent measurement points is fitted to generate continuous three-dimensional topographic data, and random noise is eliminated through moving average filtering.

[0032] As a further scheme of the present application, the acquisition process of the macroscopic warping degree feature is:

[0033] The three-dimensional topographic data is fitted to calculate the reference datum, and the reference datum matched with the curved surface shape is established;

[0034] The height deviation value of each measurement point from the reference datum is calculated, a height deviation distribution map is generated, and the root mean square value of the overall height deviation is extracted as a warping quantization index;

[0035] The principal curvature and average curvature of each point on the surface are calculated through the curvature analysis algorithm, the curvature abnormal area is identified and marked as a potential warping area;

[0036] The height deviation distribution and curvature distribution characteristics are combined to generate the macroscopic warping degree feature.

[0037] As a further scheme of the present application, the acquisition process of the micro-surface defect feature is:

[0038] The three-dimensional topographic data is subjected to multi-scale wavelet decomposition to separate the surface feature components of different frequency bands;

[0039] Adaptive threshold segmentation algorithm is used in high frequency component to identify micro defect area, and the micro defect is classified according to the depth, area and shape characteristics of the defect;

[0040] The defect area is accurately profiled to calculate the depth distribution, volume and edge steepness of each defect;

[0041] The extracted micro defects are classified into three types of scratches, pits and protrusions, and the number and distribution density of each type of defect are counted.

[0042] As a further scheme of the application, the construction process of the comprehensive feature vector is:

[0043] The macroscopic warping degree feature is normalized to convert the root mean square value of height deviation, maximum warping amount and curvature unevenness into standard dimensions;

[0044] The micro surface defect feature is weighted and coded, and different weight coefficients are assigned according to the defect type, wherein the scratch type defect is given a higher weight, and the pit and protrusion type defects are given a medium weight;

[0045] The normalized macro feature and the weighted micro feature are combined into a multi-dimensional feature vector, wherein the macro feature occupies the first three dimensions, and the micro feature occupies the subsequent dimensions;

[0046] The multi-dimensional feature vector is reduced in dimension to retain the top five principal component features with the highest contribution rate to form the final comprehensive feature vector.

[0047] As a further scheme of the application, the specific implementation mode of the mobile phone glass cover plate detection model includes:

[0048] A deep neural network is used to construct a mobile phone glass cover plate detection model, and the network structure includes a feature extraction layer, an attention weighting layer and a classification output layer;

[0049] In the feature extraction layer, a one-dimensional convolution kernel is used to process the comprehensive feature vector to extract feature representations at different scales;

[0050] The attention weighting layer dynamically adjusts the weight of each feature dimension to enhance the attention to key defect features;

[0051] The classification output layer calculates the probability distribution of each flatness level to output the final evaluation result.

[0052] A mobile phone glass cover plate flatness detection system, comprising:

[0053] A multi-band optical detection module, the multi-band optical detection module projects a multi-band composite light beam to the glass cover plate to be detected through a confocal optical detection system, and synchronously receives a mixed spectrum signal containing multi-layer interface reflection;

[0054] The intelligent compressive sampling module compressively samples the mixed spectrum signal, selects a spectrum channel according to the refractive index characteristics of the glass material, and generates an optimized sampling signal;

[0055] The multi-layer signal analysis module analyzes the optimized sampling signal in real time, and separates the reflection components of each layer interface;

[0056] The three-dimensional topography reconstruction module calculates the three-dimensional topography data of the glass cover plate surface for the separated surface reflection component;

[0057] The multi-scale feature extraction module extracts the macro warping degree feature and the micro surface defect feature of the glass cover plate based on the three-dimensional topography data;

[0058] The intelligent evaluation and decision module constructs the macro warping degree feature and the micro surface defect feature into a comprehensive feature vector as the input of the mobile phone glass cover plate detection model, and outputs the mobile phone glass cover plate flatness evaluation result.

[0059] The beneficial effects of the present application are:

[0060] (1) The present application separates the multi-layer interface reflection signal, reconstructs the three-dimensional topography data of the glass cover plate surface, and simultaneously extracts the macro warping degree feature and the micro surface defect feature on this basis. Further, by weighting coding and multi-dimensional feature fusion, a comprehensive feature vector containing macro and micro information is constructed, which is input into a deep neural network model for intelligent evaluation.

[0061] (2) The present application uses an ultrashort pulse laser source combined with an acousto-optic tunable filter and a dispersive prism to generate a three-waveband composite probe beam optimized for the characteristics of glass material, and realizes the phase modulation of multi-layer reflection signal through a spiral phase plate, which enhances the recognition ability of different interfaces. In the signal processing link, based on the glass material spectrum feature library, channel selection and filtering are carried out, intelligent compressive sampling is implemented, only the key waveband information is retained, and the data redundancy and environmental noise interference are effectively reduced. While ensuring the detection accuracy, the data processing amount is reduced, the system response speed is improved, and the stability and practicality in complex industrial environment are enhanced. BRIEF DESCRIPTION OF DRAWINGS

[0062] The present application will be further described below with reference to the accompanying drawings.

[0063] Figure 1 is a flowchart of the method of the present application;

[0064] Figure 2 is a flowchart of the system of the present application. DETAILED DESCRIPTION

[0065] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative effort belong to the scope of protection of the present application.

[0066] Please refer to Figure 1 As shown in the figure, the present application is a mobile phone glass cover plate flatness detection method, comprising the following steps:

[0067] S1. Projecting a multi-band composite light beam on the glass cover plate to be measured through a confocal optical detection system, and synchronously receiving a mixed spectrum signal containing multi-layer interface reflection;

[0068] S2. Compressively sampling the mixed spectrum signal, selecting a spectrum channel according to the refractive index characteristics of the glass material, and generating an optimized sampling signal;

[0069] S3. Real-time analysis of the optimized sampling signal, separating the reflection components of each layer interface;

[0070] S4. Calculating the three-dimensional topographic data of the glass cover plate surface for the separated surface reflection component;

[0071] S5. Extracting the macro warping degree feature and the micro surface defect feature of the glass cover plate based on the three-dimensional topographic data;

[0072] S6. Constructing the macro warping degree feature and the micro surface defect feature into a comprehensive feature vector as the input of the mobile phone glass cover plate detection model, and outputting the mobile phone glass cover plate flatness evaluation result.

[0073] In S1, a multi-band composite light beam is projected on the glass cover plate to be measured through a confocal optical detection system, and a mixed spectrum signal containing multi-layer interface reflection is synchronously received, which specifically includes:

[0074] The confocal optical detection system contains multiple special sensors, which jointly complete the functions of light beam generation, signal acquisition and data processing. The system uses a super-continuous spectrum laser source as an illumination sensor. The sensor uses a photonic crystal fiber structure pumped by an ytterbium-doped fiber laser, which can generate a broadband light beam with a wavelength range of 420 nanometers to 1,100 nanometers. The light beam processing includes an acousto-optic tunable filter. The sensor uses a tellurium dioxide crystal as an acousto-optic medium, and can accurately extract a specific wavelength sub-band from the broadband light beam by controlling the wavelength selection through a radio frequency signal.

[0075] The signal acquisition uses an avalanche photodiode array as a photoelectric conversion sensor. The sensor array includes one hundred and forty-four by one hundred and forty-four detection units, each unit having a size of five micrometers by five micrometers, a working bandwidth of one hundred megahertz, and a detection sensitivity of a single photon level. The phase modulation uses a high-speed rotating spiral phase plate. The sensor is driven by a precision stepper motor, and the rotational speed stability reaches one ten-thousandth. The sensor can accurately modulate the phase of the reflected light.

[0076] The supercontinuum laser source sensor first generates a broadband light beam, which enters an acousto-optic tunable filter sensor. The acousto-optic tunable filter selects three characteristic wavelength subbands from the broadband light beam according to pre-set control parameters. The first wavelength subband has a center wavelength of four hundred and fifty nanometers and a bandwidth of ten nanometers. The wavelength has a high reflectivity on the glass surface. The second wavelength subband has a center wavelength of five hundred and fifty nanometers and a bandwidth of fifteen nanometers. The wavelength is sensitive to interface contamination and defects. The third wavelength subband has a center wavelength of eight hundred nanometers and a bandwidth of twenty nanometers. The wavelength has a large penetration depth in the glass material.

[0077] The three subband light beams after wavelength selection are combined by a dispersive prism. The dispersive prism is made of fused quartz material and coated with an antireflection film on the surface. The dispersive prism can adjust the light beams of different wavelengths to the same propagation direction. The combined light beam is focused by a high numerical aperture objective lens. The numerical aperture of the objective lens is zero point nine five, and the focal length is four millimeters. The objective lens can form a detection spot with a diameter of about one micrometer on the surface of the glass cover plate.

[0078] In the signal acquisition path, the high-speed rotating spiral phase plate sensor modulates the phase of the reflected light beam. The optical thickness of the spiral phase plate changes in a spiral shape with the rotation angle, so that the interface reflection light of different depths produces a specific vortex phase distribution. The phase modulation depth of the surface reflection light is about two-sevenths of pi, and the phase modulation depth of the deep reflection light is about four-sevenths of pi. The difference provides a characteristic basis for signal separation.

[0079] The avalanche photodiode array sensor captures the modulated reflected light signal. The working temperature of the sensor array is controlled at negative twenty degrees Celsius, and the temperature is kept stable by a thermoelectric cooler. The output signal of each detection unit is amplified by a transimpedance amplifier and converted into a digital signal by a high-speed analog-to-digital converter. The sampling rate is two hundred megahertz, and the resolution is sixteen bits.

[0080] The digital signal processor demodulates the collected signal. The demodulation algorithm uses a synchronous detection technique, and the rotation signal of the spiral phase plate is used as a reference signal to extract the characteristic phase components of each interface reflection light. For each wavelength subband, the corresponding reflection intensity distribution of each layer interface is demodulated. The digital lock-in amplification technique is used in the demodulation process.

[0081] After demodulation, the system obtains the intensity data of each layer interface reflection corresponding to each wavelength sub-band. These data are arranged in order of wavelength and interface depth to form a two-dimensional data set containing spectral dimension and depth dimension. The data set represents the multi-layer interface reflection characteristics of the glass cover plate, that is, the mixed spectral signal. Each data point in the mixed spectral signal contains three dimensions of wavelength information, depth information and intensity information.

[0082] The system uses standard reference samples for regular calibration. During calibration, surface roughness standard blocks and step height standard blocks are used to establish system response functions and error correction models. Temperature sensors monitor real-time changes in ambient temperature, and pressure sensors monitor changes in atmospheric pressure. These data are used for real-time environmental compensation of measurement results. Humidity sensors monitor the ambient humidity to ensure that the optical elements are not affected by water vapor.

[0083] In S2, the mixed spectral signal is compressed and sampled, spectral channels are selected according to the refractive index characteristics of the glass material, and an optimized sampling signal is generated, which specifically includes:

[0084] The establishment of the glass material spectral feature library is a systematic process. First, standard samples of various common glass materials need to be collected, including soda-lime glass, alumino-silicate glass, lithium-alumino-silicate glass, etc. The number of samples for each material is not less than fifty, from different production batches, to ensure the representativeness of the data. The measurement process is carried out in a standard experimental environment, with an ambient temperature of 20 degrees Celsius and a relative humidity of 50%, to avoid the influence of environmental factors on the measurement results.

[0085] Each sample is scanned using a spectral measurement system. The measurement system uses a halogen tungsten lamp as the light source and a grating spectrometer as the detector, with a spectral range of 300 nanometers to 1,100 nanometers and a spectral resolution of 0.5 nanometers. Each sample is measured ten times at different positions, and the average value is taken as the spectral data of the sample. When measuring, pay attention to keep the incident angle at zero degrees to avoid the influence of angle on reflectance measurement.

[0086] The collected raw spectral data is preprocessed. First, the dark current correction is performed to eliminate the noise signal of the detector itself. Then, the light source intensity correction is performed to eliminate the influence of non-uniformity of the light source spectrum. Finally, the wavelength calibration is performed to verify and adjust the wavelength accuracy of the spectrometer using the standard spectral lines of the mercury lamp.

[0087] The preprocessed spectral data is stored in the feature library. The feature library uses a relational database management system, and each record contains material type, chemical composition, thickness, surface treatment process and corresponding spectral reflectance data. The spectral data is stored in array form, with a wavelength interval of 0.5 nanometers and a reflectance value of four significant digits.

[0088] The feature library is regularly updated and maintained. New material types are added every quarter, and existing data is re-measured and verified annually. Data quality monitoring is established, and re-measurement is performed promptly when data anomalies are discovered. Data backups are also established to ensure the security and reliability of the feature library.

[0089] The selection of the characteristic wavelength range is based on the optical property analysis of glass materials. The first range is chosen primarily considering the surface reflection properties. By analyzing the reflection spectra of different materials in the visible light band, it is found that the 420-480 nm band generally has a high reflectivity. The photon energy in this band is relatively high, and the interaction with the glass surface electrons is strong, which can generate a strong reflection signal. The center wavelength of this range is chosen at 450 nm, with a bandwidth of 60 nm, which can cover the surface reflection peak region of most glass materials.

[0090] The second range is chosen for its sensitivity to internal interface changes. Through experiments, it is found that the 550-600 nm band of light signals is sensitive to interface defects and contamination. The photon energy in this band is moderate, allowing it to penetrate the surface layer to the internal interface while maintaining sufficient energy to generate detectable reflection signals. The center wavelength of this range is chosen at 575 nm, with a bandwidth of 50 nm. In this band, changes in the refractive index at the interface will cause significant changes in reflectivity, providing effective information for interface quality assessment.

[0091] The third range is chosen based on the consideration of penetration ability. The photon energy in the near-infrared band is relatively low, and the interaction with the material is weak, thus having good penetration performance. The 700-800 nm band has been proven to have good penetration in most glass materials. The center wavelength of this range is chosen at 750 nm, with a bandwidth of 100 nm. The light signal in this band can penetrate thicker glass layers and reach deep interfaces, providing necessary information support for detecting multi-layer structures.

[0092] The determination of the wavelength range also needs to consider the hardware limitations of the detection system. The selection of the three ranges avoids the intensity valleys of common light sources and the sensitivity blind areas of detectors, ensuring that the signals have sufficient intensity and quality. At the same time, appropriate spacing is maintained between the ranges to avoid crosstalk problems caused by spectral overlap.

[0093] The spectral filtering process is implemented using a tunable Fabry-Perot interference filter. The filter is composed of two highly parallel mirrors, with high-reflectivity dielectric films coated on the mirror surfaces, achieving a reflectivity of 99.5%. The mirror spacing is controlled by a piezoelectric ceramic driver, with a nanometer-level adjustment accuracy. By changing the voltage applied to the piezoelectric ceramic, the transmission wavelength of the filter can be accurately adjusted.

[0094] The filtering process is performed in three steps. First, the filter is adjusted to the first interval, transmitting light signals in the 420- to 480-nanometer band. This step lasts five milliseconds, during which the output signal of the detector is recorded. Then the filter is adjusted to the second interval, transmitting light signals in the 550- to 600-nanometer band, also for five milliseconds. Finally, the filter is adjusted to the third interval, transmitting light signals in the 700- to 800-nanometer band. After each interval is collected, the filter returns to the initial state, ready for the next measurement cycle.

[0095] To ensure filtering accuracy, the operating temperature of the filter is controlled at 25 degrees Celsius, with a temperature fluctuation of no more than 0.1 degrees Celsius. A thermoelectric cooler and a temperature sensor form a closed-loop control system to adjust the temperature of the filter in real time. At the same time, changes in ambient air pressure are monitored, as changes in air pressure can affect the transmission characteristics of the filter. When the air pressure changes by more than 100 Pa, the compensation algorithm is automatically started to adjust the control voltage to maintain the stability of the transmission wavelength.

[0096] During the filtering process, attention is paid to avoid spectral distortion. The passband shape of the filter is close to a Gaussian distribution, with a full width at half maximum of 80% of the set bandwidth. The suppression ratio outside the passband reaches -60 decibels, ensuring that light signals in other bands are effectively suppressed. Regular calibration is performed to maintain the accuracy and consistency of the transmission characteristics of the filter. Calibration uses a standard mercury lamp and a helium-neon laser to verify the accuracy of the transmission wavelength.

[0097] The photoelectric conversion uses a silicon photodiode detector, with a spectral response range covering 350 to 1,100 nanometers, and a responsivity of 0.5 ampere per watt in the target band. The detector is installed on a thermoelectric cooling platform, with an operating temperature of -10 degrees Celsius, effectively reducing dark current noise. The load resistance of the detector is selected to be 1,000 ohms, ensuring response speed while obtaining sufficient output voltage.

[0098] The converted electrical signal is first subjected to preliminary amplification by a preamplifier. The amplifier uses a low-noise design with a noise figure of less than 2 decibels and a gain of 20. The amplified signal is filtered by a bandpass filter to remove DC components and high-frequency noise. The passband frequency range of the filter is from 100 Hz to 100 kHz, covering the main components of the light signal modulation frequency.

[0099] Signal conditioning also includes impedance matching and level adjustment. Since the subsequent digital processing system uses single-ended input, the differential output of the amplifier needs to be converted to single-ended signal. A differential amplifier is used to complete this conversion, with a common-mode rejection ratio of 80 decibels. The signal level is adjusted to the range of +2.5 volts, matching the input range of the analog-to-digital converter.

[0100] In order to eliminate the interference of ambient light, optical modulation and synchronous detection technology are adopted. A square wave modulation of 1 kHz is applied at the light source end, and a lock-in amplifier is used at the detection end to extract the signal component at the modulation frequency. This method can effectively suppress the DC component and power interference of ambient light, and improve the signal-to-noise ratio.

[0101] The analog-to-digital conversion adopts a 16-bit precision converter with a sampling rate of 500 kHz. This sampling rate can meet the signal bandwidth requirements while not excessively increasing the data processing burden. The converted digital signal is first subjected to digital filtering, using a finite-length unit impulse response filter with a passband of 950 Hz to 1050 Hz and a stopband attenuation of -80 dB.

[0102] The signal processing algorithm includes the following steps: first, baseline correction is performed to eliminate slow-changing components caused by temperature drift and light source fluctuations. A moving average algorithm is used with a window width of 1 second to update the baseline value in real time. Then, noise suppression is performed using a wavelet threshold denoising method, selecting a sym8 wavelet basis for five-layer decomposition. The soft threshold function can effectively preserve the signal characteristics while suppressing noise.

[0103] Signal optimization also includes feature extraction and data compression. Five characteristic parameters are extracted from the signal in each wavelength interval: peak amplitude, peak position, full width at half maximum, asymmetry, and area integral. These parameters can fully describe the characteristics of the signal while compressing the data to 5% of the original data. The characteristic parameters are stored in a 32-bit floating-point format to ensure calculation accuracy.

[0104] In S3, the optimized sampling signal is analyzed in real time to separate the reflection components of each layer interface, including:

[0105] The establishment of the multi-layer interface reflection characteristic template library is based on the accumulation and analysis of a large amount of experimental data. First, standard samples of various typical material combinations need to be prepared, including single-layer glass, double-layer bonded structure, and three-layer composite structure, etc. Each type of structure requires at least thirty samples, covering different thickness combinations and interface treatment processes. The measurement environment is controlled under standard laboratory conditions, with a temperature of 20°C and a relative humidity of 50%, to avoid the influence of environmental fluctuations on the measurement results.

[0106] High-precision optical measurement systems are used to collect the reflection signals of each layer interface. The measurement system uses a broadband light source and a high-speed spectrometer, with a spectral range of 400 nm to 800 nm and a spectral resolution of 0.2 nm. Multiple measurements are taken at different positions for each sample to ensure the statistical reliability of the data. Fixed incident and detection angles are used during measurement to maintain consistency in measurement conditions.

[0107] The original data collected are subjected to feature extraction and processing. From the time domain feature aspect, the parameters such as the rise time, fall time, pulse width and oscillation characteristics of the signal are extracted. From the frequency domain feature aspect, the spectral distribution, peak frequency and bandwidth characteristics of the signal are analyzed. Meanwhile, the intensity distribution and noise characteristics of the signal are recorded. All the feature parameters are subjected to normalization processing to eliminate the influence of absolute intensity and highlight the shape characteristics.

[0108] A feature parameter database is established, which adopts a hierarchical storage structure. The first layer is classified according to the material type, the second layer is classified according to the thickness combination, and the third layer stores specific feature parameters. The database adopts time stamp version management to record the time and content of each data update. The database is periodically optimized and arranged, and the outdated data is deleted and the new measurement results are added.

[0109] The similarity matching adopts a multi-feature weighted matching algorithm. First, the optimized sampling signal is preprocessed, including baseline correction, noise filtering and amplitude normalization. The preprocessed signal is compared with the reference signal in the feature template library. The comparison process is divided into three levels: overall waveform similarity comparison, local feature point matching and statistical characteristic analysis.

[0110] The overall waveform similarity is calculated by using the dynamic time warping algorithm, which can handle the slight differences in the time scale of the signal. The minimum warping path distance between the two signals is calculated, and the smaller the distance value, the higher the similarity. The local feature point matching selects the extreme points, inflection points and characteristic frequency points of the signal for comparison, and calculates the matching degree of the feature point position and amplitude. The statistical characteristic analysis compares the parameters such as mean, variance, skewness and kurtosis of the signal.

[0111] The matching result adopts a percentage scoring system, and the overall similarity score is generated by combining the comparison results of the three levels. The matching threshold is set, and when the similarity score is higher than eighty percent, it is considered to be a successful match. In order to improve the matching efficiency, a hierarchical search strategy is adopted, which first performs coarse matching to select the candidate templates, and then performs fine matching to determine the final result.

[0112] The signal decomposition adopts a layer-by-layer stripping method. First, the surface reflection component is separated, because the surface signal is usually the strongest and the most obvious in feature. The best template signal matched is used to estimate the amplitude and phase parameters of the surface signal by using the least squares method. The surface reflection signal is reconstructed according to the estimated parameters, and the reconstructed signal is subtracted from the original signal to obtain the remaining signal.

[0113] The remaining signal is matched again to separate the intermediate layer reflection component. Since the intermediate layer signal may be disturbed by the surface signal, an iterative optimization algorithm needs to be used. In each iteration, the parameters of the template signal are adjusted so that the matching degree of the remaining signal with the template is the highest. This process is repeated until all the signals of the layers are separated.

[0114] An adaptive step size control algorithm is used in the decomposition process. A larger step size is used in the initial stage to quickly approach the optimal solution, and a smaller step size is used in the later stage for fine adjustment. At the same time, the maximum number of iterations is set to prevent infinite loops, and the decomposition process is terminated when the number of iterations exceeds the set value.

[0115] The separation effect is verified by residual energy analysis. The energy value of the remaining signal after separation is calculated and compared with the total energy of the original signal. The residual energy ratio formula is the ratio of the remaining signal energy to the original signal energy. When the ratio is less than 1%, it is considered that the separation effect is good.

[0116] At the same time, waveform consistency check is carried out. The separated layer signals are superimposed to synthesize the reconstructed signal, and the waveform is compared with the original signal. The correlation coefficient and root mean square error are calculated to evaluate the reconstruction quality. The correlation coefficient should be higher than 0.95, and the root mean square error should be lower than 3%.

[0117] Physical reasonableness verification is also carried out. Check whether the intensity ratio of each layer signal conforms to the prediction of optical theory, and whether the time sequence matches the optical path difference. Any result that violates the physical law needs to be rechecked for separation process.

[0118] An evaluation system for separation quality is established, including multiple evaluation indexes and corresponding threshold standards. After each separation is completed, a quality report is generated, recording the values of each index and the evaluation results. For the case of substandard quality, the re-separation process is automatically started or manual intervention is prompted.

[0119] In order to realize real-time processing requirements, digital signal processors and optimized algorithms are used. The algorithm complexity is carefully designed to ensure that the processing task is completed within limited computing resources. Parallel computing technology is used to process data from multiple channels simultaneously.

[0120] Cache management is designed to balance data processing speed and memory usage efficiency. Ring buffer is used to store input data, and double buffer is used to store intermediate results. Memory mapping technology is used to improve data access speed.

[0121] Real-time monitoring and exception handling are implemented. Parameters such as processing delay, memory usage and computing load are continuously monitored. When abnormal conditions are detected, processing parameters are automatically adjusted or backup solutions are started to ensure stable operation of the system.

[0122] In S4, the three-dimensional topography data of the glass cover plate surface is calculated for the surface reflection component obtained by separation, including:

[0123] The time-frequency joint analysis uses the short-time Fourier transform method to process the reflection signal. First, the reflection signal is divided into multiple time segments, and the duration of each segment is set to ten microseconds according to the signal characteristics. The Fourier transform is performed on each time segment to obtain the frequency component distribution in that time period. By analyzing the frequency variation characteristics in different time segments, the time-frequency distribution characteristics of the signal are extracted. The energy concentration of the signal in a specific frequency range and the law of frequency component change over time are observed. Three key parameters are extracted from the time-frequency analysis results: the variation curve of the main frequency component over time, the distribution characteristics of the frequency energy, and the phase relationship between different frequency components. These parameters collectively reflect the time-frequency characteristics of the reflection signal, providing a basis for subsequent height calculation. In the analysis process, the Hanning window function is used to reduce spectral leakage, and the window function overlap rate is set to 50% to ensure the continuity and accuracy of the time-frequency analysis.

[0124] The optical path propagation path reconstruction is based on the principles of geometric optics. First, a mathematical model of the optical system is established, including the positions of the light source, the optical center of the objective lens, and the detector, and other key parameters. According to the time difference of the reflection signal, the optical path length change is calculated. Considering the focusing characteristics of the objective lens, there is a certain correspondence between the reflection point position of the light on the sample surface and the receiving position of the detector. Through an iterative optimization algorithm, the optimal optical path configuration is found, which makes the calculated optical path best match the actual measurement data. In the reconstruction process, the influence of the aberration characteristics of the objective lens, including spherical aberration, coma, and field curvature, is considered, and the pre-calibrated aberration correction coefficients are used for compensation. At the same time, the influence of the refractive index of the glass material is considered, and the optical path is corrected according to the known material optical parameters. Finally, the accurate spatial coordinates of each measurement point are obtained, including the lateral position and vertical height information.

[0125] The curved surface fitting uses the non-uniform rational B-spline method. First, the discrete height data points are grid processed to establish a regular grid structure. According to the density and distribution characteristics of the measurement points, the appropriate grid resolution is selected, usually set to one-half of the measurement point spacing. The least squares method is used to fit the surface model, making the deviation between the fitted surface and the original data points minimal. In the fitting process, an adaptive weight distribution strategy is used, giving higher weights to higher quality data points and lower weights to data points that may have errors. After the surface fitting is completed, a moving average filtering method is used to eliminate random noise. The filter window size is adaptively adjusted according to the surface characteristics, using a larger window in flat areas to improve the smoothing effect, and a smaller window in areas with rich features to preserve the details. Finally, a continuous and smooth three-dimensional topography data is generated, accurately reflecting the geometric characteristics of the glass cover plate surface. During data processing, the quality evaluation indicators of each point are recorded, providing a reliability basis for subsequent analysis.

[0126] In S5, based on the three-dimensional topography data, the macroscopic warping degree characteristics and microscopic surface defect characteristics of the glass cover plate are extracted, specifically including:

[0127] The reference surface fitting is performed using the moving least squares method. This method establishes a smooth reference surface through local weighted regression, which can adapt to different shapes of surface features. First, the three-dimensional topography data is divided into multiple local regions, each containing no less than one hundred data points. Quadratic surface fitting is performed on each local region, and the fitting parameters are calculated using the weighted least squares method. The weight function uses a Gaussian kernel function, and the kernel width is adaptively adjusted according to the data point density. In the calculation process, each data point is assigned a different weight, with data points closer to the fitting point having a larger weight and data points farther away having a smaller weight. Through an iterative optimization algorithm, the weighted deviation square sum of the fitting surface and the original data points is minimized. After fitting, the local fitting surfaces are smoothly spliced to generate a continuous reference surface. The resolution of the reference surface is set to one-half of the original data point density, ensuring that both the overall shape characteristics and the excessive detail noise are reflected. The entire fitting process uses parallel computing technology to improve processing efficiency and ensure that the reference surface calculation is completed within a limited time.

[0128] The height deviation calculation uses the point-by-point comparison method. The actual height value of each measurement point is subtracted from the height value of the corresponding point on the reference surface to obtain the height deviation value. The deviation value is stored in numerical form, and a visual height deviation distribution map is generated. The distribution map uses pseudo-color display, with different colors representing different degrees of deviation, making it easy to visually observe the surface flatness. Several statistical characteristic quantities are extracted from the height deviation data, including the root mean square value of the overall height deviation, the maximum positive deviation, the maximum negative deviation, and the standard deviation of the deviation distribution. The root mean square value is calculated as the square root of the average of the square sum of the point deviations, which can comprehensively reflect the overall warping degree of the surface. The probability density function of the deviation distribution is also calculated to analyze the distribution characteristics of the deviation values. For areas with large deviations, their position coordinates and deviation values are recorded as areas of focus. All deviation data are normalized to eliminate the influence of absolute values, facilitating comparison and analysis between different samples.

[0129] The curvature analysis is based on the principle of differential geometry. First, the three-dimensional topography data is Gaussian smoothed to eliminate the influence of high-frequency noise on curvature calculation. The standard deviation of the smoothing kernel function is adaptively selected according to the surface characteristics, usually between two to five times the data point spacing. Then the principal curvatures of each point on the surface, i.e. the maximum curvature and the minimum curvature, are calculated. The curvature calculation uses a local quadratic surface fitting method, which fits a quadratic surface in the neighborhood of each point, and then calculates the curvature value through the coefficients of the surface. The neighborhood radius is selected to be three to five times the data point spacing, ensuring that there are enough data points for reliable fitting. The average curvature and the Gaussian curvature are calculated according to the principal curvature values, the average curvature is the arithmetic mean of the two principal curvatures, and the Gaussian curvature is the product of the two principal curvatures. By setting appropriate curvature threshold values, curvature anomaly regions are identified. The threshold value is dynamically adjusted according to the surface type, for general glass surfaces, the curvature threshold value is set to zero point zero one micrometer negative one power. The identified curvature anomaly regions are marked, recording their position, area and curvature characteristics, as candidates for potential warping regions.

[0130] Wavelet decomposition uses the Mallat algorithm for discrete wavelet transform. The Daubechies wavelet basis function is selected, and the decomposition level is four. The first layer of decomposition obtains the highest frequency component, which contains the surface roughness information; the second layer of decomposition obtains the second highest frequency component, which contains the medium scale surface features; the third and fourth layers of decomposition obtain the low frequency component, which contains the overall shape information of the surface. After each layer of decomposition, the approximation coefficients and detail coefficients are stored and processed respectively. The pyramid algorithm is used in the decomposition process, and each layer of decomposition reduces the signal length by half while maintaining the frequency characteristics of the signal. The wavelet coefficients are thresholded to eliminate noise effects, using a soft threshold function. The threshold value is adaptively selected according to the noise level, calculated by the Donoho-Johnstone method. After decomposition, the component signals of each frequency band are reconstructed to obtain the surface feature components at different scales. High frequency components are used for microscopic defect analysis, and low frequency components are used for overall shape analysis of the surface.

[0131] The defect recognition adopts an adaptive threshold segmentation algorithm. First, the local statistical features of the high-frequency component are calculated, including the mean and standard deviation. The segmentation threshold is dynamically calculated according to the local statistical features, and the threshold calculation formula is the local mean plus three times the local standard deviation. The high-frequency component image is binarized, and the area greater than the threshold is marked as a potential defect area. Morphological operations are performed on the binary image, including opening and closing operations, to eliminate noise points and connect adjacent defect areas. Then the feature parameters of each defect area are calculated, including area, perimeter, compactness, maximum depth, average depth, and depth standard deviation. The area is obtained by counting the pixels, and the perimeter is obtained by counting the boundary pixels. The compactness calculation formula is four times the circumference divided by the square of the perimeter. According to these feature parameters, a decision tree algorithm is used to classify the defects. Scratch defects usually have a large aspect ratio and small area; pit defects have negative depth values and steep edges; and protrusion defects have positive depth values and gentle edges. After classification, the number and distribution density of each type of defect are counted, and a defect statistical report is generated.

[0132] Detailed feature parameter calculation is performed on each identified defect area. Depth distribution is obtained by height value statistics of each point in the defect area, and maximum depth, average depth, and depth standard deviation are calculated. Volume calculation uses the integral method to accumulate the volume elements in the defect area to obtain the total volume. Edge steepness is calculated by calculating the height gradient of the defect edge points, and the Sobel operator is used to calculate the gradient value, and then the distribution characteristics of the gradient value are calculated. For scratch defects, additional length, width, and direction angle parameters are calculated; for pit and protrusion defects, additional curvature radius and symmetry parameters are calculated. All feature parameters are normalized to eliminate the effects of size. Interpolation methods are used to improve the calculation accuracy during parameter calculation, and sub-pixel level interpolation calculation is performed in the defect edge area. Finally, a defect feature vector containing all feature parameters is generated, providing data support for subsequent quality evaluation.

[0133] In S6, the macroscopic warping degree feature and the microscopic surface defect feature are constructed into a comprehensive feature vector, which is used as the input of the mobile phone glass cover plate detection model, and the output is the mobile phone glass cover plate flatness evaluation result, which specifically includes:

[0134] The feature normalization process adopts the minimum-maximum scaling method. First, a large amount of sample data is collected to statistically analyze the distribution range of each feature parameter. For the height deviation root mean square value in the macro warping feature, the normal range is zero to five microns, and values exceeding this range are considered abnormal. The normal range of the maximum warping amount is zero to ten microns, and the normal range of the curvature unevenness is zero to zero point five microns. In the normalization process, each feature parameter is linearly converted to the range of zero to one. The conversion formula is the parameter value minus the minimum value divided by the difference between the maximum value and the minimum value. The normalized feature parameters have the same dimension and value range, which facilitates subsequent feature fusion and processing. The normalized parameters are updated regularly based on historical data to adapt to the feature changes of different batches of products. At the same time, an abnormal value processing is established, and when the feature parameter exceeds the normal range, a special mark and processing are performed to avoid interference with subsequent analysis.

[0135] The defect feature weighted encoding is based on the importance of defect types for weight allocation. By analyzing a large amount of quality data, the influence degree of different types of defects on product quality is determined. Scratch defects may affect product appearance and strength, so they are given a weight coefficient of zero point five; pit defects may become stress concentration points, so they are given a weight coefficient of zero point three; and protrusion defects affect assembly flatness, so they are given a weight coefficient of zero point two. After weight allocation, the feature parameters of each defect are calculated with weights. The weighted feature parameters can better reflect the actual influence degree of the defects. At the same time, the size factor of the defects is considered, and the weight of large area defects is appropriately increased, and the weight of small area defects is appropriately reduced. The weight coefficient is dynamically adjusted according to the update of product quality standards to ensure that the evaluation results are consistent with the latest quality standards. In the weighting process, a segmented weighting strategy is adopted, and different weight coefficients are used for defects in different size ranges to improve the rationality of weight allocation.

[0136] The feature vector construction adopts a direct splicing method. The three normalized macro feature parameters are placed in the first three dimensions of the feature vector, corresponding to the height deviation root mean square value, the maximum warping amount, and the curvature unevenness. The subsequent dimensions place the weighted micro defect feature parameters, including the number of defects, the total defect area, the maximum defect size, the average defect depth, and the defect distribution density. The total dimension of the feature vector is determined according to the actual detection requirements, usually between ten and twenty dimensions. Each dimension of the feature parameter is standardized to ensure the consistency of the numerical distribution. After the construction of the feature vector is completed, an integrity check is performed to ensure that all necessary feature parameters are included. At the same time, a consistency verification is performed to check whether the logical relationship between different feature parameters is reasonable.

[0137] The principal component analysis adopts the eigenvalue decomposition method of the covariance matrix. First, the covariance matrix of the eigenvector is calculated, which reflects the correlation between each feature dimension. Then the eigenvalue and eigenvector of the covariance matrix are calculated, and the size of the eigenvalue reflects the variance contribution rate of the corresponding principal component. The principal components are selected in the order of eigenvalues from large to small, and the first few principal components with a cumulative contribution rate of more than 85% are retained. Usually, five to eight principal components are retained, which can retain most of the information in the original eigenvector. After principal component analysis, the original eigenvector is projected into the principal component space to obtain the reduced comprehensive eigenvector. The load matrix of the principal component is recorded during the dimension reduction process, which is convenient for subsequent result interpretation and analysis.

[0138] The deep neural network adopts a fully connected feedforward structure. The number of input layer nodes is the same as the dimension of the comprehensive eigenvector, the hidden layer is set to two layers, the first hidden layer contains thirty-two nodes, the second hidden layer contains sixteen nodes, and the output layer contains three nodes, corresponding to the qualified, unqualified and pending three quality grades. The hidden layer activation function uses the ReLU function, and the output layer activation function uses the Softmax function. The network weight is initialized using the Xavier method, and the bias term is initialized to zero. The learning rate is set to zero.001, and the Adam optimization algorithm is used for parameter update. In order to prevent overfitting, a Dropout layer is added after the hidden layer, and the dropout rate is set to 0.2. The network structure is optimized and adjusted according to the actual application effect, and the optimal network parameter configuration is determined through cross-validation.

[0139] The attention weighting layer is implemented using the self-attention mechanism. First, the attention score of each feature dimension is calculated, and the attention score is obtained through a fully connected layer and a Softmax function. Then, the attention score is multiplied by the original eigenvector to obtain the weighted feature representation. The attention mechanism can make the model automatically pay attention to the feature dimensions that are more important for quality evaluation, improving the accuracy of evaluation. The calculation process of attention weight is interpretable, and the model decision basis can be understood by analyzing the attention weight. The attention layer parameters are trained together with other parts of the network, and are optimized through the back propagation algorithm. The distribution of attention weights is also used as auxiliary information for model output to help users understand the evaluation results.

[0140] The classification output layer uses a Softmax function to calculate the probability of each quality level. The output result is three probability values, representing the likelihood of belonging to the qualified, unqualified, and pending three levels, respectively. The level with the maximum probability is taken as the final evaluation result. The confidence of the evaluation result is also calculated, which is the difference between the maximum probability value and the second largest probability. When the confidence is lower than the threshold, the result is marked as a low-confidence evaluation, and manual review is recommended. The evaluation result is accompanied by detailed explanation information, including the main influencing factors and key defect characteristics. The distribution of attention weights is displayed through visualization to help users understand the model's decision-making process. All evaluation results are stored in a database to establish a complete quality evaluation record, supporting historical data query and statistical analysis.

[0141] Referring to Figure 2 As shown in the figure, a mobile phone glass cover plate flatness detection system comprises:

[0142] A multi-band optical detection module projects a multi-band composite light beam onto the glass cover plate to be detected through a confocal optical detection system, and synchronously receives a mixed spectrum signal containing multi-layer interface reflection;

[0143] An intelligent compressed sampling module compressively samples the mixed spectrum signal, selects a spectrum channel according to the refractive index characteristics of the glass material, and generates an optimized sampling signal;

[0144] A multi-layer signal analysis module analyzes the optimized sampling signal in real time, and separates the reflection components of each layer interface;

[0145] A three-dimensional topography reconstruction module calculates the three-dimensional topography data of the glass cover plate surface for the separated surface reflection component;

[0146] A multi-scale feature extraction module extracts the macro warping degree feature and the micro surface defect feature of the glass cover plate based on the three-dimensional topography data;

[0147] An intelligent evaluation and decision module constructs the macro warping degree feature and the micro surface defect feature into a comprehensive feature vector as the input of the mobile phone glass cover plate detection model, and outputs the mobile phone glass cover plate flatness evaluation result.

[0148] Working principle of the present application: the present application projects a multi-band composite light beam to the glass cover plate to be measured through a confocal optical detection system and receives a mixed spectrum signal containing multi-layer interface reflection; the mixed spectrum signal is compressed and sampled, and an optimized sampling signal is generated according to the spectral channel selected according to the refractive index characteristics of the glass material; after separating the interface reflection components of each layer by using a real-time analytical algorithm, the surface layer reflection component is subjected to time-frequency joint analysis and optical path reconstruction, and the accurate three-dimensional topographic data is calculated; based on the three-dimensional data, the macroscopic warping degree characteristics and the microscopic surface defect characteristics are extracted, and the comprehensive feature vector is constructed through feature normalization, weighted coding and principal component analysis; finally, a deep neural network model with attention mechanism is used for processing, and the grade evaluation result and the confidence index of the flatness of the glass cover plate are output, so that accurate quality detection is realized.

[0149] The above describes one embodiment of the present application in detail, but the content described is only the preferred embodiment of the present application, and cannot be considered as limiting the scope of the present application. Any equivalent changes and improvements made within the scope of the present application should still belong to the patent coverage of the present application.

Claims

1. A method for detecting flatness of a mobile phone glass cover plate, characterized in that, The method comprises the following steps: S1, projecting a multi-band composite light beam onto the glass cover plate to be detected by a confocal optical detection system, and synchronously receiving a mixed spectrum signal containing multi-layer interface reflection; S2, compressively sampling the mixed spectrum signal, selecting a spectrum channel according to the refractive index characteristics of the glass material, and generating an optimized sampling signal; S3, real-time analysis of the optimized sampling signal, separating the reflection components of each layer interface; S4, calculating the three-dimensional topography data of the glass cover plate surface for the separated surface layer reflection component; S5, based on the three-dimensional topography data, extracting the macro warping degree characteristics and micro surface defect characteristics of the glass cover plate; S6, constructing the macro warping degree characteristics and micro surface defect characteristics into a comprehensive feature vector as the input of the mobile phone glass cover plate detection model, and outputting the mobile phone glass cover plate flatness evaluation result. 2.The mobile phone glass cover plate flatness detection method of claim 1, wherein, The received mixed spectrum signal containing multi-layer interface reflection specifically includes: An ultrashort pulse laser source is used to generate a light beam covering the visible light to near-infrared band, and an acousto-optic tunable filter is used to select three characteristic wavelength subbands, wherein the first subband wavelength corresponds to the glass surface reflectivity, the second subband wavelength is sensitive to the interface pollution, and the third subband wavelength has the maximum material penetration depth; The light beams of the three characteristic wavelength subbands are combined by a dispersive prism to form a coaxial composite detection light beam, which is focused by an aperture objective lens to form a tomographic detection spot; The phase of the reflected light is modulated by a high-speed rotating spiral phase plate to generate a characteristic vortex phase distribution of the interface reflection light at different depths; The modulated reflected light signal is synchronously captured, and the reflection intensity distribution of each layer interface is demodulated to obtain a mixed spectrum signal containing multi-layer interface reflection characteristics. 3.The mobile phone glass cover plate flatness detection method of claim 1, wherein, The generated optimized sampling signal specifically includes: A glass material spectrum feature library is established, and the corresponding three characteristic wavelength intervals are extracted from the glass material spectrum feature library, wherein the first interval corresponds to the surface reflection peak value, the second interval is sensitive to the internal interface change, and the third interval has the maximum penetration ability; The mixed spectrum signal is filtered to transmit only the optical signals in the three characteristic wavelength intervals; The filtered optical signal is converted into an electrical signal, and environmental noise interference is eliminated by using digital signal processing technology, and finally an optimized sampling signal containing key optical characteristics is generated.

4. The method for detecting flatness of a mobile phone glass cover plate according to claim 1, characterized in that, The real-time analysis of the optimized sampling signal and the separation of the reflection components of each layer interface specifically includes: The spectral form characteristics and time domain response characteristics of the interface reflection signals of mobile phone glass cover plates with different material combinations are pre-stored as a multi-layer interface reflection feature template library; The optimized sampling signal is matched with the reflection characteristics of each layer interface in the feature template library, and each interface reflection component contained in the optimized sampling signal is matched; The surface layer reflection component, the intermediate layer reflection component and the bottom layer reflection component are sequentially separated from the optimized sampling signal, and the remaining signal is updated and matched again after each decomposition; The separation effect of each layer reflection component is analyzed and verified, the residual energy of each layer reflection component is calculated, and when the residual energy is lower than a set threshold, the separation is completed, and the reflection components of each layer interface are output. 5.The mobile phone glass cover plate flatness detection method of claim 1, wherein, The acquisition process of the three-dimensional topography data is: The separated reflection part is subjected to time-frequency joint analysis to extract the time difference of arrival and frequency domain characteristic parameters of the reflection signal; The propagation path of the reconstructed light path is combined with the focusing characteristics of the objective lens to calculate the actual height value of each measurement point; The height data of adjacent measurement points are subjected to surface fitting to generate continuous three-dimensional topography data, and random noise is eliminated through moving average filtering. 6.The mobile phone glass cover plate flatness detection method according to claim 1, characterized in that, The macro warping degree feature acquisition process is as follows: A reference datum plane is calculated by fitting the three-dimensional topography data, and a reference datum plane matching the surface shape is established; The height deviation values of each measurement point from the datum plane are calculated to generate a height deviation distribution map, and the root mean square value of the overall height deviation is extracted as a warping quantification index; The principal curvature and average curvature of each point on the surface are calculated through a curvature analysis algorithm, abnormal curvature regions are identified, and are marked as potential warping regions; The macro warping degree feature is generated by combining the height deviation distribution and curvature distribution characteristics.

7. The method for detecting flatness of a mobile phone glass cover plate according to claim 1, characterized in that, The micro surface defect feature acquisition process is as follows: The three-dimensional topography data are subjected to multi-scale wavelet decomposition to separate the surface feature components of different frequency bands; In the high-frequency component, an adaptive threshold segmentation algorithm is used to identify micro defect regions, which are classified according to the depth, area and shape characteristics of the defects; The defect regions are subjected to accurate contour extraction, and the depth distribution, volume and edge steepness parameters of each defect are calculated; The extracted micro defects are classified into three types: scratches, pits and protrusions, and the number and distribution density of each type of defect are counted. 8.The mobile phone glass cover plate flatness detection method of claim 1, wherein, The construction process of the comprehensive feature vector is as follows: The macro warping degree feature is normalized to convert the root mean square value of the height deviation, the maximum warping amount and the curvature non-uniformity into standard dimensions; The micro surface defect feature is weighted and encoded, and different weight coefficients are assigned according to the defect type, with scratches being assigned a higher weight and pits and protrusions being assigned a medium weight; The normalized macro feature and the weighted micro feature are combined into a multi-dimensional feature vector, with the macro feature occupying the first three dimensions and the micro feature occupying the subsequent dimensions; The multi-dimensional feature vector is subjected to dimension reduction processing, and the top five principal component features with the highest contribution rate are retained to form the final comprehensive feature vector. 9.The mobile phone glass cover plate flatness detection method of claim 1, wherein, The specific implementation mode of the mobile phone glass cover plate detection model includes: A deep neural network is used to construct the mobile phone glass cover plate detection model, and the network structure includes a feature extraction layer, an attention weighting layer and a classification output layer; In the feature extraction layer, a one-dimensional convolution kernel is used to process the comprehensive feature vector to extract feature representations at different scales; The attention weighting layer dynamically adjusts the weights of each feature dimension to enhance the attention to key defect features; In the classification output layer, the probability distribution of each flatness level is calculated to output the final evaluation result. 10.A mobile phone glass cover plate flatness detection system, characterized in that, A mobile phone glass cover plate flatness detection method according to any one of claims 1-9, comprising: A multi-band optical detection module that projects a multi-band composite light beam onto the glass cover plate to be detected through a confocal optical detection system, and simultaneously receives mixed spectral signals containing multiple interface reflections; The intelligent compressive sampling module compressively samples the mixed spectrum signal, selects a spectrum channel according to the refractive index characteristic of the glass material, and generates an optimized sampling signal; The multi-layer signal analysis module analyzes the optimized sampling signal in real time, and separates the reflection components of each layer interface; The three-dimensional topography reconstruction module calculates the three-dimensional topography data of the glass cover plate surface for the separated surface reflection component; The multi-scale feature extraction module extracts the macro warping degree feature and the micro surface defect feature of the glass cover plate based on the three-dimensional topography data; The intelligent evaluation and decision module constructs the macro warping degree feature and the micro surface defect feature into a comprehensive feature vector as the input of the mobile phone glass cover plate detection model, and outputs the mobile phone glass cover plate flatness evaluation result.

Images