A method for real-time detection of laser glass defects

Abstract

This application provides a real-time laser glass defect detection method, comprising: acquiring ambient light intensity and vibration frequency fluctuation data, capturing initial image data of the laser glass crack area and locating crack intersections, analyzing the gray value attenuation distribution pattern of crack intersections, and obtaining quantitative indicators of spectral interference; acquiring the light and shadow micro-tremor suppression effect of branch textures, and quantitatively evaluating the remaining masking degree of background light artifacts on the crack edge structure by combining the optimized gray value uniformity index of crack intersections; executing a determined pulse frequency configuration scheme, acquiring laser glass defect detection image sequences in real time, evaluating the improvement degree of the real-time detection image sequence relative to the initial image data, quantifying the multi-dimensional imaging optimization effect of the mesh crack texture, and dynamically generating a light source parameter adjustment sequence.

Description

Technical Field

[0001] This invention relates to the field of information technology, and in particular to a method for real-time detection of defects in laser glass. Background Technology

[0002] Laser glass defect detection is a crucial step in modern optical manufacturing and quality control, directly impacting the performance and reliability of high-precision optical components. Especially in high-end applications such as aerospace and laser manufacturing, defects like network cracks can lead to catastrophic consequences. Traditional detection methods rely on fixed light sources and single imaging modes, with a major drawback being their inability to adapt to dynamic interference in complex environments. For example, synchronous fluctuations in ambient light intensity and vibration frequency can degrade image quality, manifesting as uneven grayscale at crack intersections due to spectral interference, blurred branch textures due to light and shadow tremors, and background artifacts further obscuring crack edges. These methods struggle to capture subtle crack features in dynamic environments, limiting the accuracy and stability of real-time detection. The core challenge lies in effectively handling the complex interaction between the spatial frequency distribution of crack textures and the instantaneous spectral characteristics of ambient light. The spatial frequency distribution reflects the geometric characteristics of the crack, but dynamic changes in ambient light interfere with signal acquisition by the imaging system, making crack edge identification difficult. For instance, in high-frequency vibration environments, the grayscale values ​​at crack intersections decay non-uniformly due to spectral interference, making it impossible for traditional fixed light sources to accurately distinguish between cracks and background artifacts. This interference further exacerbates the light and shadow tremor effect, causing blurring of branch textures in the image. Especially in scenes with strong background light artifacts, the system has difficulty in stably extracting the true edge structure of the crack. Summary of the Invention

[0003] This invention provides a method for real-time detection of defects in laser glass, mainly comprising:

[0004] Ambient light intensity and vibration frequency fluctuation data are collected to capture initial image data of the laser glass crack area, locate the crack intersection point, analyze the gray value attenuation distribution pattern of the crack intersection point, and determine the spectral interference quantification index, which is the dynamic statistical mean of the spectral interference intensity in each direction of the crack intersection point.

[0005] Based on the spectral interference quantification index, branch texture features are extracted from the non-uniform attenuation distribution of gray values ​​at the crack intersection, and the range of the area covered by the background light artifacts on the crack edge structure is identified.

[0006] Based on the range of the masked area and the spatial frequency distribution characteristics of the crack texture, the level of interference of the instantaneous spectral characteristics of ambient light on the imaging system is determined.

[0007] Obtain the combination parameters of the light source polarization angle and wavelength, combine them with the interference level, determine the pulse frequency adjustment vector, optimize the multi-dimensional imaging parameters of crack intersections and branch textures, and determine the gray-scale uniformity index of crack intersections.

[0008] The branch texture light and shadow tremor suppression rate is calculated by comparing the inter-frame position offset of the branch texture before and after imaging parameter optimization. Based on the branch texture light and shadow tremor suppression effect, combined with the gray-scale uniformity index of the crack intersection point, the remaining degree of the background light artifacts covering the crack edge structure is evaluated.

[0009] Based on the remaining coverage level, a pulse frequency configuration scheme is determined, and laser glass defect imaging detection results are obtained.

[0010] Furthermore, the process involves collecting ambient light intensity and vibration frequency fluctuation data, capturing initial image data of the laser-cut glass crack area, locating crack intersections, analyzing the grayscale attenuation distribution pattern at the crack intersections, and determining a spectral interference quantification index. This spectral interference quantification index is the dynamic statistical mean of the spectral interference intensity in each direction at the crack intersection, including:

[0011] Ambient light intensity fluctuation data and vibration frequency change sequence are collected, and the time sequence data of light intensity and vibration acceleration data on the surface of the laser glass are recorded through a photoelectric sensor array.

[0012] The vibration frequency distribution is determined based on the vibration acceleration data;

[0013] Based on the vibration frequency distribution, the initial image data of the laser glass crack region is obtained, the crack outline is identified, the pixel coordinates of the crack intersection point are located, and the gray values ​​of the crack intersection point and its neighborhood are obtained.

[0014] Based on the gray values ​​of the crack intersection and its neighborhood, the gray value difference is calculated, the gray value attenuation amplitude distribution is statistically analyzed, and the directional characteristics of the gray value attenuation distribution pattern are determined.

[0015] Based on the aforementioned directional characteristics, the correlation coefficient between the grayscale attenuation amplitude and the light intensity time series data is calculated to determine the spectral interference quantification index.

[0016] Furthermore, based on the spectral interference quantization index, extracting branch texture features from the non-uniform attenuation distribution of gray values ​​at the crack intersection, and identifying the range of the area obscured by background light artifacts on the crack edge structure, includes:

[0017] Based on the spectral interference quantification index, the branch texture extension direction in the non-uniform attenuation distribution of gray values ​​at the crack intersection is identified, the angle of the dominant direction of the branch texture is statistically analyzed, the gray value sequence is extracted, and a branch texture feature vector containing texture width, curvature change rate and gray value contrast parameters is constructed.

[0018] Based on the texture width parameter in the branch texture feature vector, the offset of the branch texture position in adjacent frame images is calculated, the frequency and amplitude distribution of micro-vibration are statistically analyzed, and the light and shadow micro-vibration effect enhancement coefficient is determined.

[0019] The initial image data of the laser glass crack area is processed by a filtering algorithm to determine the distribution of background light artifacts, identify the overlapping area between the background light artifacts and the crack edge, and determine the range of the masked area. The standard deviation of the Gaussian kernel function is dynamically adjusted according to the light and shadow tremor effect enhancement coefficient.

[0020] Based on the range of the masked area, a spatial frequency distribution analysis of the crack texture is performed to extract the amplitude and phase of the main frequency component of the spectrum. Combined with ambient light intensity fluctuation data, the interference level of the instantaneous spectral characteristics of ambient light on the imaging system is determined.

[0021] Furthermore, the acquisition of the light source polarization angle and wavelength combination parameters, combined with the interference level, determines the pulse frequency adjustment vector, optimizes the multi-dimensional imaging parameters of the crack intersection point and branch texture, and determines the gray-scale uniformity index of the crack intersection point, including:

[0022] Obtain the combined parameters of the light source polarization angle and wavelength, including the dominant wavelength value and spectral distribution curve data;

[0023] Based on the interference level, the interference frequency distribution is determined, and the pulse frequency adjustment vector is calculated, including the frequency change amplitude and phase offset.

[0024] Based on the pulse frequency adjustment vector, different exposure times and gain parameters are set for the crack intersection region and the branch texture region, and multiple frames of images are acquired and the grayscale information of the multiple frames of images is fused.

[0025] Based on the grayscale distribution of the fused image, equalization processing is performed to compensate for the non-uniform attenuation of grayscale values. The standard deviation of grayscale values ​​in the neighborhood of the crack intersection is calculated to determine the grayscale uniformity index of the crack intersection.

[0026] Furthermore, the step of calculating the branch texture light and shadow jitter suppression rate by comparing the inter-frame position offset of the branch texture before and after imaging parameter optimization, and evaluating the remaining degree of background light artifacts covering the crack edge structure based on the branch texture light and shadow jitter suppression effect and the gray-level uniformity index of the crack intersection, includes:

[0027] Calculate the difference in the branch texture position offset before and after optimization to determine the suppression rate;

[0028] Based on the suppression rate and the gray-scale uniformity index of the crack intersection, a comprehensive quality score is calculated;

[0029] Based on the comprehensive quality score, the sensitivity parameters for detecting background light artifacts are adjusted, the distribution area of ​​background light artifacts is identified, the set of pixel points of the crack edge contour is extracted, the overlap ratio between the background light artifacts and the crack edge contour is calculated, and the remaining camouflage degree is determined.

[0030] Furthermore, the step of updating the light source polarization angle and wavelength combination parameters based on the remaining masking degree to determine the pulse frequency configuration scheme includes:

[0031] Calculate the difference between the remaining masking degree and the threshold, determine the adjustment range of the light source polarization angle and wavelength combination parameters, update the light source polarization angle and wavelength combination parameters, and determine the pulse frequency configuration scheme;

[0032] The updated crack image is acquired, the edge sharpness value of the crack image is calculated, and the clarity index is determined.

[0033] Furthermore, after determining the pulse frequency configuration scheme, the process includes acquiring a laser glass defect detection image sequence, evaluating the degree of improvement of the laser glass defect detection image sequence relative to the initial image data, wherein the optimization effect of multi-dimensional imaging of mesh crack texture is quantified; using a cyclic feedback method to process ambient light intensity and vibration frequency fluctuation data to generate a light source parameter adjustment sequence; executing the light source parameter adjustment sequence to obtain high-precision imaging detection results of laser glass defects.

[0034] Furthermore, the acquisition of laser glass defect imaging detection results includes:

[0035] The pulse frequency configuration scheme is executed to drive the light source to emit a pulse beam, acquire a laser glass defect detection image sequence, calculate the difference in grayscale distribution between the image sequence and the initial image data, and determine the signal-to-noise ratio improvement and contrast enhancement rate.

[0036] Based on the signal-to-noise ratio improvement and contrast enhancement rate, the geometric feature parameters and photometric feature parameters of the mesh crack texture are extracted, the change in feature parameters before and after optimization is calculated, and the multidimensional imaging optimization effect score is determined.

[0037] Based on the multidimensional imaging optimization effect score, monitor the changes in ambient light intensity and vibration frequency, calculate the light source polarization angle and wavelength adjustment amount, and generate a light source parameter adjustment sequence.

[0038] The light source parameter adjustment sequence is executed to calculate the grayscale standard deviation of crack intersections, the inter-frame displacement of branch textures, and the pixel ratio of background light artifacts, thereby determining the laser glass defect imaging detection results.

[0039] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects:

[0040] This invention discloses a real-time laser glass defect detection method, solving operational problems such as gray-scale attenuation at crack intersections, enhanced light and shadow vibration effects in branched textures, and background light artifact masking in complex environments. By collecting ambient light intensity and vibration frequency fluctuation data, an initial crack image is captured and intersections are located. The gray-scale attenuation distribution pattern is analyzed, branched texture features are extracted, and spectral interference and light and shadow vibration effects are quantified. A Gaussian filtering algorithm is used to eliminate background light artifacts and identify masked areas. When masking exceeds a threshold, the ambient light interference level is assessed based on the spatial frequency distribution of the crack texture. The pulse frequency adjustment vector is optimized by combining the light source polarization angle and wavelength parameters to compensate for non-uniform gray-scale attenuation and improve the gray-scale uniformity at crack intersections. If the masking level is insufficient, the frequency vector is iteratively adjusted, ultimately generating a dynamic light source parameter adjustment sequence. This invention optimizes imaging parameters in real time through a cyclic feedback mechanism, ensuring continuous suppression of gray-scale attenuation, light and shadow vibration, and artifact masking in complex environments, achieving high-precision imaging detection of laser glass defects and significantly improving image quality and detection accuracy. Attached Figure Description

[0041] Figure 1 This is a flowchart of a laser glass defect real-time detection method according to the present invention. Detailed Implementation

[0042] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0043] like Figure 1 This embodiment of a laser glass defect real-time detection method specifically includes:

[0044] Step S101: Collect fluctuation data of ambient light intensity and vibration frequency, capture initial image data of the laser glass crack area and locate the crack intersection point, analyze the gray value attenuation distribution pattern of the crack intersection point, and obtain the quantitative index of spectral interference.

[0045] Ambient light intensity fluctuation data is collected to obtain a real-time sequence of vibration frequency changes. A photoelectric sensor array records the time-series data of light intensity and vibration acceleration on the laser glass surface. The time-series data of light intensity includes a curve showing the change of light intensity amplitude over time, and the vibration acceleration data is transformed using Fourier transform to obtain the vibration frequency distribution. At each frequency band of the vibration frequency distribution, initial image data of the laser glass crack area is captured. The Canny edge detection operator is used to identify the crack outline, and morphological closing operations are used to connect the fracture edges. The pixel coordinates of the intersection nodes formed by the intersection of crack branches are located, and the gray values ​​of the intersection nodes and their eight neighborhoods are obtained. For the gray values ​​of the intersection nodes and their eight neighborhoods, the gray value difference between the center pixel and surrounding pixels is calculated, and the amplitude distribution of gray value attenuation in different directions is statistically analyzed. This amplitude distribution reflects the gray value decreasing pattern extending outward from the center of the intersection node, identifying the directional characteristics of the gray value attenuation distribution pattern. Based on the directional characteristics of the gray value attenuation distribution pattern, the correlation coefficient between the gray value attenuation amplitude and the light intensity time series data in each direction is calculated. When the correlation coefficient exceeds a preset threshold, it is determined that there is spectral interference in that direction. The average value of the interference intensity in all directions is used as a quantitative indicator of spectral interference.

[0046] In one embodiment, ambient light intensity fluctuation data is acquired using a photoelectric sensor array arranged around the laser glass inspection platform. The photoelectric sensor array comprises at least eight silicon photodiodes, located at different positions on the glass surface, with each sensor recording the ambient light intensity value at a sampling frequency of 1000 Hz. The light intensity time-series data is represented as a continuous voltage signal sequence, where the voltage amplitude is proportional to the light intensity. The real-time variation sequence of vibration frequency is acquired using a triaxial accelerometer attached to the support base of the inspection platform, which collects vibration acceleration data in the x, y, and z directions.

[0047] Specifically, a Fast Fourier Transform (FFT) is performed on the vibration acceleration data to convert the time-domain signal into a frequency-domain representation. During the FFT, the acquired acceleration time-series data is first processed using a Hamming window to reduce spectral leakage. The resulting vibration frequency distribution covers a range of 0 to 500 Hz, including the dominant frequency components and harmonic components generated by the operation of mechanical equipment in the environment. Power spectral density analysis identifies the frequency bands where vibration energy is concentrated, as these bands have the most significant impact on image quality.

[0048] It should be noted that under the influence of different vibration frequencies, the surface of the laser glass will undergo minute displacements and deformations, causing dynamic changes in the focal plane of the imaging system. When the vibration frequency has a specific relationship with the exposure time of the imaging system, a motion blur effect will occur. Therefore, image data of the laser glass crack area are captured at each of the identified main vibration frequencies. During image acquisition, the exposure time of the CCD camera is dynamically adjusted according to the vibration frequency to ensure imaging at a stable phase point of the vibration cycle.

[0049] For example, the application of the Canny edge detection operator includes multiple processing stages. First, the original image is Gaussian filtered to smooth noise while preserving edge information. Then, the gradient magnitude and direction of the image are calculated, and the horizontal and vertical gradient components are calculated separately using the Sobel operator. In the non-maximum suppression stage, the gradient magnitude of pixels is compared along the gradient direction, and local maxima are retained as potential edges. Double thresholding divides edge points into strong and weak edges, and the weak edges are connected to the strong edges through an edge connection algorithm to form a complete crack contour.

[0050] In one possible implementation, morphological closing operations are used to connect the edges of fractured cracks. The closing operation consists of dilation and erosion operations. The dilation operation processes the binarized edge image using a 3×3 structuring element, filling small gaps in the crack contour. The subsequent erosion operation restores the original crack width while preserving connectivity. Intersections of crack branches are identified through connected component analysis; these locations are called intersection nodes. Intersection node determination is based on pixel connectivity; a pixel is marked as an intersection node when its eight-neighborhood contains three or more crack branch pixels.

[0051] Preferably, the analysis of grayscale value attenuation distribution patterns focuses on the local region of the intersection nodes. For each identified intersection node, a 9×9 pixel window centered on that node is extracted. Within this window, the central pixel represents the grayscale value of the intersection node, and the eight surrounding directions correspond to extension directions of 0 degrees, 45 degrees, 90 degrees, 135 degrees, 180 degrees, 225 degrees, 270 degrees, and 315 degrees, respectively. A decreasing sequence of grayscale values ​​is calculated along each direction, forming eight grayscale attenuation curves. The grayscale difference is calculated using the absolute difference between adjacent pixels, reflecting the drastic degree of grayscale change. When there is interference from ambient light, the attenuation curves in different directions will exhibit an asymmetrical distribution, with the attenuation rate in some directions being significantly faster than in others.

[0052] Understandably, the quantification process of spectral interference involves joint analysis in the time and frequency domains. The grayscale attenuation amplitude sequences in each direction are cross-correlated with the light intensity time-series data acquired simultaneously. The cross-correlation function reflects the similarity between two signals at different time delays. When there is a causal relationship between grayscale attenuation and light intensity fluctuations, the cross-correlation function will show a peak at a specific time delay. The correlation coefficient is calculated using the Pearson correlation coefficient formula, with values ​​ranging from -1 to 1. When the absolute value of the correlation coefficient in a certain direction exceeds a preset threshold of 0.7, it indicates that that direction is subject to significant spectral interference.

[0053] For example, in a real-world inspection scenario, when fluorescent lights in a workshop flicker at a frequency of 50Hz, periodic bright and dark stripes will appear in the crack image. This interference is most pronounced in the direction perpendicular to the light tube, and the grayscale attenuation curve in the corresponding direction will exhibit periodic fluctuations. By statistically analyzing the correlation coefficients exceeding a threshold in eight directions, the arithmetic mean of these coefficients is calculated as the interference intensity of that intersection node. A weighted average of the interference intensities of all intersection nodes is then performed, with the weights determined based on the importance of the node in the crack network, resulting in a quantitative index of spectral interference for the entire crack region. This index typically ranges from 0 to 1; a higher value indicates more severe spectral interference, providing a quantitative basis for subsequent optimization of imaging parameters.

[0054] Step S102: Based on the quantitative index of spectral interference, extract the branch texture features, analyze the enhancement degree of the branch texture light and shadow tremor effect caused by grayscale attenuation, identify and mark the range of the area covered by the background light artifacts on the crack edge structure, and when the range of the covered area exceeds the preset threshold, evaluate the interference level of the instantaneous spectral characteristics of ambient light on the imaging system based on the spatial frequency distribution characteristics of the crack texture.

[0055] Based on spectral interference quantization, the extension direction of branch textures is identified from the non-uniform attenuation distribution of gray values ​​at crack intersections. The dominant direction angle of each branch is statistically analyzed using a gradient direction histogram. Gray-level sequences of consecutive pixels are extracted along these dominant direction angles. Based on the fluctuation amplitude and periodic variation characteristics of the gray-level sequences, a branch texture feature vector containing parameters such as texture width, curvature change rate, and gray-level contrast is constructed. For the texture width parameter in the branch texture feature vector, the positional offset of the corresponding branch texture in adjacent frames is calculated. When the offset exceeds a preset threshold, a light and shadow tremor phenomenon is determined. The frequency and amplitude distribution of tremors per unit time are statistically analyzed, and the light and shadow tremor effect enhancement coefficient is calculated by the ratio of the mean amplitude to the degree of gray-level attenuation. A Gaussian filtering algorithm is used to perform convolution operations on the original image. The standard deviation of the Gaussian kernel function is dynamically adjusted according to the enhancement coefficient of the light and shadow micro-tremor effect. The difference between the filtered image and the original image is the background light artifact distribution map. The pixel regions covered by artifacts are identified by binarization and connected component labeling. When the overlap area between the artifact region and the crack edge exceeds a preset threshold, it is marked as a masked region. If the range of the masked region exceeds a preset proportional threshold of the total crack area, a two-dimensional discrete Fourier transform is performed on the crack texture to obtain the spatial frequency distribution. The amplitude and phase of the dominant frequency component are extracted from the spectrum. The instantaneous spectral characteristics are calculated according to the time-varying light intensity curve of the ambient light. The instantaneous spectral characteristics are then correlated with the spatial frequency distribution in the frequency domain to obtain the numerical value of the interference level of the ambient light on the imaging system.

[0056] In one embodiment, the extraction of branch texture features is guided by a spectral interference quantization index. When the spectral interference quantization index exceeds 0.6, it indicates severe non-uniform grayscale attenuation at the crack intersection. In this case, it is necessary to accurately identify the extension direction of the branch texture from the attenuation distribution. The construction process of the gradient direction histogram includes: selecting a 15×15 pixel analysis window around the crack intersection, calculating the gradient magnitude and direction of each pixel within the window. The gradient direction is divided into 36 intervals from 0 to 360 degrees, each interval representing a 10-degree angle range. The gradient magnitudes falling into each interval are summed to form a direction distribution histogram. The peak value in the histogram corresponds to the dominant direction angle of the branch texture.

[0057] Specifically, pixel grayscale sequences are extracted by extending outwards from the intersection point along the identified dominant directional angle. The extraction process employs bilinear interpolation to ensure accurate grayscale values ​​even at non-integer pixel locations. The grayscale sequence length is set to 50 pixels, covering the main part of the branched texture. By analyzing the first and second differences of the grayscale sequence, the variation patterns of fluctuation amplitude are identified. Periodic variation characteristics are calculated using an autocorrelation function; when the autocorrelation function shows a peak at a specific delay position, it indicates the existence of a periodic texture structure.

[0058] It should be noted that the construction of the branch texture feature vector integrates multiple geometric and photometric parameters. The texture width is determined by a grayscale profile perpendicular to the dominant direction, taking the full width when the grayscale value drops to half of the peak value. The rate of curvature change reflects the degree of curvature of the branch texture, obtained by calculating the angle change formed by three consecutive points along the dominant direction. The grayscale contrast parameter adopts the Weber contrast formula, which is the difference between the average grayscale of the texture region and the average grayscale of the background region divided by the average grayscale of the background. These three parameters together constitute the branch texture feature vector, providing basic data for subsequent light and shadow micro-vibration analysis.

[0059] For example, the detection of light and shadow tremor effects is achieved through inter-frame comparison. During laser glass defect detection, multiple frames of images are continuously acquired at a frame rate of 30 frames per second. For each branch texture, a feature matching algorithm is used to find the corresponding position in adjacent frames. The matching process is based on normalized cross-correlation, searching for the position with the highest correlation coefficient within a search window. The position offset is calculated with sub-pixel precision, using a parabolic fitting method to fit three points near the correlation peak into a quadratic function; the peak position is the precise offset.

[0060] In one possible implementation, the enhancement coefficient for the light and shadow micro-vibration effect considers the statistical characteristics of the amplitude distribution. The unit time is typically set to 1 second, and all detected micro-vibration events within this time period are counted. The micro-vibration frequency is defined as the number of times the offset exceeds 0.5 pixels. The amplitude distribution is represented by a histogram, with the horizontal axis representing the offset magnitude and the vertical axis representing the frequency of occurrence. The mean amplitude is calculated using a weighted average, with the weights proportional to the offset magnitude. The degree of grayscale attenuation is obtained from the aforementioned grayscale sequence analysis and is represented as the slope of the grayscale decrease. The enhancement coefficient is equal to the ratio of the mean amplitude to the degree of grayscale attenuation, reflecting the extent of the impact of micro-vibration on image quality.

[0061] Preferably, the parameters of the Gaussian filtering algorithm are dynamically adjusted according to the enhancement coefficient of the light and shadow micro-tremor effect. The initial standard deviation of the Gaussian kernel function is set to 1.5 pixels. When the enhancement coefficient exceeds 1.0, the standard deviation increases linearly, with the standard deviation increasing by 0.2 pixels for every 0.1 increase in the enhancement coefficient. The size of the Gaussian kernel is determined based on the standard deviation, taking 6 times the standard deviation and rounding up to an odd number. The convolution operation is accelerated using a Fast Fourier Transform, converting spatial convolution into frequency domain multiplication.

[0062] Understandably, background light artifact identification is achieved through difference image analysis. The filtered image is subtracted pixel-by-pixel from the original image; the difference reflects the filtered high-frequency components, primarily originating from ambient light scattering and imaging system noise. The difference image automatically determines the binarization threshold using Otsu's method, separating artifact regions from normal regions. Connected component labeling employs the 8-connectivity criterion, marking adjacent artifact pixels as the same region. When the number of overlapping pixels between a connected component and the crack edge exceeds 30% of the total pixels in that connected component, the region is marked as a masked region.

[0063] For example, in actual testing, when the flicker of the fluorescent tube resonates with the natural frequency of the vibration platform, a noticeable halo effect is formed at the crack edge. This halo appears as a bright band surrounding the crack in the difference image, typically 3 to 5 pixels wide. By statistically analyzing the total area of ​​all masked areas and comparing it with the total area of ​​the crack area, an interference assessment process based on spatial frequency analysis is initiated when the ratio exceeds a preset threshold of 0.4. Further, a two-dimensional discrete Fourier transform converts the crack texture from the spatial domain to the frequency domain. Before the transform, the image is processed using a Hanning window to reduce boundary effects. The spectrum is displayed on a logarithmic scale, with the center corresponding to zero frequency and the frequency gradually increasing outwards. The dominant frequency component is determined by finding the local maxima of the spectral amplitude, which typically corresponds to the main texture period of the crack. Phase information reflects the spatial positional relationship of the texture. The instantaneous spectral characteristics of ambient light are extracted from the light intensity time-series data, and the time-frequency distribution is obtained through a short-time Fourier transform. Frequency domain correlation calculations compute the conjugate product of two spectra, and the magnitude of the result represents the degree of correlation. Finally, an interference level value in the range of 0 to 10 is obtained. The larger the value, the more severe the interference of ambient light on imaging.

[0064] Step S103: Obtain the polarization angle and wavelength combination parameters of the current light source, combine the interference level evaluation results of the instantaneous spectral characteristics of ambient light, determine the pulse frequency adjustment vector required to suppress the light and shadow micro-tremor effect, optimize the multi-dimensional imaging parameters of crack intersections and branch textures, and obtain the optimized grayscale uniformity index of crack intersections.

[0065] The polarization angle and wavelength combination parameters of the current light source are obtained. The polarization angle is read by the encoder of the polarizer rotation mechanism, and the wavelength combination parameters are extracted from the register of the light source controller, including the dominant wavelength value, bandwidth range, and spectral distribution curve data. Based on the interference level assessment results of the instantaneous spectral characteristics of ambient light, when the interference level exceeds a preset threshold, a correspondence table between interference frequency and light source parameters is established. Based on the interference frequency distribution in the correspondence table, a pulse frequency adjustment vector is calculated. The adjustment vector includes three components: frequency change amplitude, phase offset, and duty cycle adjustment parameter. The suppression coefficient corresponding to each frequency component is obtained by looking up a preset frequency suppression coefficient table. The difference between the suppression coefficient and the current pulse frequency is used as the adjustment amount to determine the pulse modulation parameter sequence. For the pulse modulation parameter sequence, a short exposure time with a high gain parameter is set for the crack intersection area, and a long exposure time with a low gain parameter is set for the branch texture area. Multiple frames of images are acquired according to the exposure time and gain parameter combination. The grayscale information of the multiple frames of images is fused by a weighted average method so that the grayscale values ​​of the fused image are concentrated within a preset target range. Histogram equalization is performed based on the grayscale distribution within the target range to compensate for grayscale differences caused by non-uniform attenuation. The grayscale standard deviation of each crack intersection point in the processed image is calculated within its eight neighborhoods. The ratio of the number of intersection points with standard deviations below a preset threshold to the total number of intersection points is calculated to obtain the optimized grayscale uniformity index of crack intersection points.

[0066] In one embodiment, the light source parameters are acquired through multi-channel synchronous acquisition. The polarizer rotation mechanism is driven by a stepper motor and equipped with a 16-bit absolute encoder, capable of reading the current polarization angle with a resolution of 0.01 degrees. The light source controller integrates a dedicated parameter storage register group containing 32 16-bit registers, which respectively store the dominant wavelength value, the upper and lower limits of the bandwidth range, and the sampling point data of the spectral distribution curve. The dominant wavelength value is typically set in the range of 532nm to 650nm, the bandwidth range is controlled between 10nm and 50nm, and the spectral distribution curve is sampled at 2nm intervals to form a complete spectral feature description.

[0067] Specifically, the interference level assessment results of the instantaneous spectral characteristics of ambient light are expressed in numerical form from 0 to 10. When this value exceeds a preset threshold of 6.0, it indicates that ambient light has a significant impact on image quality. At this point, it is necessary to establish a correspondence table between interference frequencies and light source parameters. This correspondence table adopts a two-dimensional matrix structure, where row indices represent the discretized intervals of interference frequencies, and column indices correspond to different combinations of light source parameters. Each element in the matrix records the recommended parameter adjustment value under specific frequency interference. The interference frequencies are discretized at 10Hz intervals, covering the range from 0Hz to 500Hz, forming 50 frequency intervals.

[0068] It should be noted that the calculation process of the pulse frequency adjustment vector involves the determination of three key components. The frequency change amplitude reflects the adjustment range from the current operating frequency to the target frequency, calculated by the difference between the suggested value and the current value in the corresponding relationship table. The phase offset is used to compensate for the time difference between the periodic changes in ambient light and the sampling time of the imaging system, ensuring image acquisition occurs when the ambient light intensity is stable. The duty cycle adjustment parameter controls the proportion of the pulse light source's on-time to the entire cycle; adjusting the duty cycle can change the average light intensity without affecting the peak power. The preset frequency suppression coefficient table contains empirical values ​​for the suppression effect corresponding to each frequency component. These coefficients were obtained through extensive experimental calibration, and their values ​​range from 0.1 to 0.9.

[0069] For example, differentiated imaging parameter configurations are used for crack intersections and branch texture regions. Crack intersection regions are prone to overexposure or underexposure due to drastic grayscale changes; therefore, a short exposure time combined with high gain is used. The short exposure time is set to 2ms to 5ms to capture instantaneous grayscale features and avoid motion blur. The high gain parameter is set to amplify the original signal by 3 to 5 times to compensate for insufficient signal strength caused by the short exposure. Branch texture regions have relatively gentle grayscale changes, so a long exposure time combined with low gain is used. The long exposure time is set to 10ms to 20ms to accumulate a sufficient number of photons and improve the signal-to-noise ratio. The low gain parameter is set to amplify the original signal by 1.2 to 1.8 times to avoid signal saturation.

[0070] In one possible implementation, the acquisition and fusion of multiple images is achieved through timing control. Eight to twelve images are acquired continuously based on a set combination of exposure time and gain parameters. The acquisition time of each image is synchronized with the trigger signal of the pulsed light source to ensure consistent imaging conditions. The weighting coefficients of the weighted averaging method are dynamically determined based on the quality assessment results of each image. Quality assessment metrics include contrast, sharpness, and noise level. Contrast is calculated by determining the standard deviation of the image's gray-level histogram; sharpness is measured using the response intensity of the Laplacian operator; and noise level is assessed by the energy proportion of high-frequency components. The weighting coefficients are proportional to the quality assessment metrics, with higher-quality frames receiving larger fusion weights.

[0071] Preferably, the grayscale values ​​of the fused image need to be mapped to a preset target range. The target range is typically set to a grayscale value range of 50 to 200. This range retains sufficient grayscale level information while avoiding the influence of extreme values. The mapping process uses a linear transformation to scale the original grayscale range proportionally to the target range. For grayscale values ​​exceeding the target range, truncation is applied, setting values ​​less than 50 to 50 and values ​​greater than 200 to 200.

[0072] Understandably, histogram equalization is used to further improve the uniformity of gray-level distribution. This process first statistically analyzes the gray-level histogram of the fused image and calculates the cumulative distribution function. Then, a gray-level mapping lookup table is constructed based on the cumulative distribution function, mapping the original gray-level values ​​to new gray-level values, making the processed histogram closer to a uniform distribution. This processing can enhance image contrast and make crack details clearer.

[0073] For example, in practical laser glass defect detection, a typical crack network may contain 20 to 30 intersections. For each intersection, the standard deviation of grayscale values ​​within its eight neighborhoods is calculated. The eight neighborhoods include the top, bottom, left, right, and four diagonally adjacent pixels. The standard deviation is calculated using an unbiased estimation formula, reflecting the dispersion of grayscale values. When the standard deviation is below a preset threshold of 15, the grayscale distribution in the intersection area is considered relatively uniform. The ratio of the number of all intersections meeting the condition to the total number of intersections is the grayscale uniformity index. This index ranges from 0 to 1; the closer the value is to 1, the better the overall grayscale uniformity.

[0074] Step S104: Obtain the light and shadow micro-tremor suppression effect of the branch texture, and combine it with the optimized gray-scale uniformity index of crack intersection to quantitatively evaluate the remaining masking effect of background light artifacts on the crack edge structure.

[0075] By comparing the positional offsets of the branch textures before and after optimization, the difference between the offsets before and after optimization is calculated. Dividing this difference by the offset before optimization yields the suppression rate. When the suppression rate exceeds a preset threshold, the grayscale uniformity index of the optimized crack intersections is extracted. Multiplying this grayscale uniformity index by 0.6 and adding the suppression rate multiplied by 0.4 yields a comprehensive quality score. The sensitivity parameter for background light artifact detection is adjusted based on the comprehensive quality score. When the score is higher than a preset threshold, the sensitivity is reduced. The distribution area of ​​background light artifacts under the current imaging conditions is identified. A set of crack edge contour pixels is extracted using an edge detection operator. The overlap between the artifact region and the edge contour pixels is calculated. Dividing the overlap by the total number of edge contour pixels determines the masking ratio. Based on this masking ratio, a difference calculation is performed between it and the comprehensive quality score. When the difference between the comprehensive quality score and the masking ratio is greater than a preset threshold, the impact of background light artifacts on the crack edge structure is considered acceptable; otherwise, the masking ratio is used as the remaining masking degree evaluation result.

[0076] In one embodiment, the effect of suppressing light and shadow tremors is obtained through comparative analysis of inter-frame position offsets. The branch texture position offset before optimization is obtained by tracking feature points in 10 consecutive frames of images. The Euclidean distance between feature points is calculated between each frame, and the average value is taken as the baseline offset. After optimization, the same number of image frames are acquired again, and the corresponding average position offset is calculated. The suppression rate is calculated using a relative rate of change formula, that is, the difference between the offsets before and after optimization is divided by the offset before optimization, and the result is expressed as a percentage.

[0077] Specifically, when the suppression rate exceeds a preset threshold of 60%, it indicates that the light and shadow tremors have been effectively suppressed. At this point, the optimized gray-scale uniformity index of the crack intersection is extracted from the aforementioned processing flow. The value of this index ranges from 0 to 1. The overall quality score is calculated using a weighted average method, with the weight of the gray-scale uniformity index set to 0.6 and the weight of the suppression rate set to 0.4. This weight allocation takes into account the dominant role of gray-scale uniformity in the overall imaging quality.

[0078] It should be noted that the sensitivity parameter for background light artifact detection is adaptively adjusted based on the overall quality score. When the score is higher than 0.7, the sensitivity parameter is reduced to 80% of its original value to reduce the false detection rate. The sensitivity parameter directly affects the recognition threshold of artifact regions; lower sensitivity means that only obvious artifacts will be marked. The artifact distribution region is analyzed using morphological operations to identify connected components, with each connected component representing an independent artifact block.

[0079] For example, the crack edge contour is extracted using the Sobel operator for edge detection, and the resulting set of edge pixels contains the complete contour information of the crack. When calculating the overlap between the artifact region and the edge contour, each pixel is checked to determine whether it belongs to both sets simultaneously. The number of overlapping pixels divided by the total number of pixels in the edge contour yields the masking ratio, which intuitively reflects the degree to which the artifact obscures the crack edge.

[0080] Preferably, the assessment of the remaining masking level is achieved by calculating the difference between the overall quality score and the masking ratio. When the difference between the overall quality score and the masking ratio is greater than a preset threshold of 0.3, it indicates that although there is some artifact masking, the overall imaging quality still meets the detection requirements. Otherwise, the masking ratio is directly used as the assessment result of the remaining masking level, and this result is used to guide whether further parameter optimization is needed.

[0081] In step S105, if the remaining occlusion level is lower than the preset threshold, the light source polarization angle and wavelength combination parameters are updated collaboratively to determine the final pulse frequency configuration scheme. If the remaining occlusion level is still not lower than the preset threshold, the pulse frequency adjustment vector is readjusted and iteratively optimized until the edge structure clarity requirements are met.

[0082] If the remaining masking level is lower than a preset threshold, the values ​​of the current light source polarization angle and wavelength combination parameters are read, the difference between the remaining masking level and the threshold is calculated and multiplied by a preset proportional coefficient to obtain the parameter adjustment range. The polarization angle is increased by the angle value corresponding to the adjustment range, and the wavelength combination parameter is decreased by the corresponding nanometer value. After the collaborative update, the values ​​are written to the light source control register to determine the pulse frequency configuration scheme. If the remaining masking level is still not lower than the preset threshold, the feedback control mechanism is activated. The frequency change amplitude, phase offset, and duty cycle adjustment parameters in the current pulse frequency adjustment vector are obtained. The proportion of the remaining masking level exceeding the threshold is calculated as a correction coefficient. The three parameters are multiplied by the correction coefficient to obtain a new adjustment vector. The light source control parameters are updated based on the new adjustment vector, and the updated crack image is acquired. The sharpness value of the image edge is calculated using a gradient operator as a sharpness index. When the sharpness index exceeds the preset sharpness threshold, it is determined that the edge structure sharpness requirement is met; otherwise, the parameter adjustment process of the feedback control mechanism is repeated.

[0083] In one embodiment, the threshold determination of the remaining occlusion level employs a two-branch processing mechanism. When the remaining occlusion level is below a preset threshold of 0.2, the system enters a parameter fine-tuning mode. At this time, the current polarization angle value and wavelength combination parameters are read from the register of the light source controller. The polarization angle is typically in the range of 0 to 180 degrees, and the wavelength combination parameters include two values: center wavelength and bandwidth. The difference between the remaining occlusion level and the threshold reflects the potential for image quality improvement; this difference is multiplied by a preset scaling factor of 0.5 to obtain the parameter adjustment range.

[0084] Specifically, the polarization angle is adjusted linearly, converting the calculated adjustment range into angle increments, typically 5 to 15 degrees each time. The wavelength combination parameters shift towards shorter wavelengths, with the reduction in the center wavelength in nanometers proportional to the adjustment range, generally within the range of 10 to 30 nanometers. This collaborative update mechanism considers the complementary effects of polarization and spectral characteristics. The updated parameters are written to the light source control register via a serial communication interface, forming a new pulse frequency configuration scheme.

[0085] It should be noted that when the remaining masking level is still not lower than the preset threshold, it indicates that simple parameter adjustments cannot meet the imaging requirements, and at this time, the feedback control mechanism is activated. The pulse frequency adjustment vector contains three key components: the frequency change amplitude controls the pulse time interval, the phase offset adjusts the relative timing relationship between the pulse and the ambient light, and the duty cycle adjustment parameter determines the proportion of the pulse's effective working time. The correction coefficient is calculated based on the excess ratio of the remaining masking level. If the remaining masking level is 0.3 and the threshold is 0.2, then the excess ratio is 0.5, which is used as the correction coefficient to adjust the three parameters proportionally.

[0086] For example, after updating the light source control parameters based on the new adjustment vector, a crack image is immediately acquired for effect evaluation. The sharpness index is calculated using the Laplacian gradient operator, which performs a convolution operation on the image and then calculates the root mean square value of the gradient magnitude. The larger this value, the sharper the edges and the clearer the image. The preset sharpness threshold is usually set as the 80th percentile of the root mean square value of the gradient. When the calculated sharpness index exceeds this threshold, the edge structure sharpness is considered to meet the requirements.

[0087] Preferably, if the sharpness index does not reach the threshold, the system continues to execute the parameter adjustment process of feedback control. In each iteration, the correction coefficient will be adaptively adjusted according to how close the sharpness index is to the threshold, gradually approaching the ideal parameter configuration until the imaging quality requirements for laser glass defect detection are met.

[0088] Step S106: Execute the determined pulse frequency configuration scheme, acquire the laser glass defect detection image sequence in real time, evaluate the improvement of the real-time detection image sequence relative to the initial image data, quantify the multi-dimensional imaging optimization effect of the mesh crack texture, and dynamically generate the light source parameter adjustment sequence.

[0089] A predetermined pulse frequency configuration scheme is executed, driving the light source to emit pulse beams according to the frequency parameters in the scheme. Real-time acquisition of laser glass defect detection image sequences is achieved, with a preset number of image frames acquired per second. The grayscale distribution difference between the image sequence and the initial image data is compared. The signal-to-noise ratio improvement is obtained by calculating the peak difference and distribution width change of the grayscale histogram. The contrast enhancement rate is obtained by calculating the change in the ratio of the maximum grayscale difference to the average grayscale, thus determining the quantitative value of the improvement degree. Based on the quantitative value of the improvement degree, weight coefficients for feature extraction are set. Geometric and photometric feature parameters of the mesh crack texture are extracted. The geometric feature parameters include crack width, branch angle, and radius of curvature. The photometric feature parameters include grayscale mean, standard deviation, and gradient magnitude. The changes in feature parameters before and after optimization are multiplied by the corresponding weight coefficients and then summed to obtain a comprehensive score for the multidimensional imaging optimization effect. Based on the comprehensive score, a cyclic feedback mechanism is initiated to monitor real-time changes in ambient light intensity and vibration frequency. When light intensity fluctuations exceed preset thresholds or vibration frequencies deviate from baseline values, the light source polarization angle adjustment is calculated based on the product of the fluctuation amplitude and the comprehensive score. Wavelength and pulse duty cycle adjustments are calculated based on the ratio of the deviation degree to the comprehensive score, forming a light source parameter adjustment sequence containing temporal information. The operating parameters of the light source controller are updated by executing this light source parameter adjustment sequence. Within each adjustment cycle, the grayscale standard deviation of crack intersections is calculated as an attenuation index, the inter-frame displacement of branch textures is statistically analyzed as a micro-vibration intensity index, and the ratio of background light artifact pixels to the total number of pixels is calculated as a masking range index. When all three indices are below their respective preset thresholds, high-precision imaging detection of laser glass defects is achieved.

[0090] In one embodiment, the pulse frequency configuration scheme is executed through the timing control module of the light source controller. The configuration scheme includes three core parameters: fundamental frequency, modulation depth, and phase offset. These parameters are encoded as 32-bit control words and written to the control register. The light source emits a pulsed beam according to the set frequency parameters, with the rise and fall times of the pulses controlled at the nanosecond level to ensure a steep change in light intensity. Real-time image acquisition uses a high-speed CCD camera with a frame rate set to 100 frames per second. The exposure time of each frame is synchronized with the pulse period, completing the exposure during the pulse peak to avoid interference from ambient light.

[0091] Specifically, the quantitative assessment of the degree of improvement involves comparative analysis across multiple dimensions. The peak difference of the grayscale histogram is obtained by calculating the shift in the position of the main peak of the histogram before and after optimization; this shift reflects the change in the overall brightness level. The change in distribution width is obtained by calculating the difference in the standard deviation of the histogram; an increase in the standard deviation indicates richer grayscale levels. The signal-to-noise ratio improvement is calculated using the ratio of the peak signal to the background noise, specifically the ratio of the average grayscale value of the crack area to the standard deviation of the grayscale value of the background area. The contrast enhancement rate is calculated using the Weber contrast formula, which is the grayscale difference between the crack and the background divided by the background grayscale value; the change in this ratio before and after optimization is the enhancement rate. These quantitative indicators together constitute a comprehensive evaluation system for the degree of improvement.

[0092] It should be noted that the weighting coefficients are adaptively adjusted based on the quantified value of the improvement level. When the improvement level exceeds a preset threshold, the weights of the geometric feature parameters increase, because at this point the image quality has met the basic requirements, and more attention needs to be paid to the morphological features of the crack. The crack width is obtained by measuring the distance between edges, the branch angle is determined by calculating the angle between the branch direction vectors, and the radius of curvature is calculated using a three-point circular arc fitting method. Among the photometric feature parameters, the gray-level mean reflects the overall brightness of the crack area, the standard deviation characterizes the dispersion of the gray-level distribution, and the gradient magnitude is calculated using the Sobel operator, reflecting the sharpness of the edges.

[0093] For example, the comprehensive score for the multidimensional imaging optimization effect adopts a weighted accumulation method. The change in each feature parameter is first normalized, mapping parameters of different dimensions to a unified range of 0 to 1. The normalized change is multiplied by the corresponding weight coefficient; the total weight for geometric features is 0.4, and the total weight for photometric features is 0.6. This allocation takes into account the importance of imaging quality for defect detection. The accumulated comprehensive score provides a quantitative assessment of the optimization effect, providing a basis for subsequent parameter adjustment.

[0094] In one possible implementation, the cyclic feedback mechanism is implemented through independent monitoring threads. Ambient light intensity is monitored using a photodiode array distributed across different locations on the detection platform, with each sensor sampling at a frequency of 1 kHz. Vibration frequency is acquired via an accelerometer, and the dominant frequency component is extracted using a Fast Fourier Transform. When the light intensity fluctuation exceeds 20% of the baseline value or the vibration frequency deviates by more than 5 Hz, a parameter adjustment process is triggered. The adjustment amount for the light source polarization angle is equal to the product of the fluctuation amplitude and the overall score, multiplied by a preset coefficient, typically within the range of 5 to 20 degrees. The wavelength adjustment takes into account the degree of frequency deviation; the greater the deviation, the more the wavelength is adjusted towards shorter wavelengths to improve the temporal resolution of the imaging.

[0095] Preferably, the light source parameter adjustment sequence is stored using a queue structure. Each adjustment instruction includes a timestamp, parameter type, and adjustment value. The timestamp ensures the timing accuracy of parameter adjustments, the parameter type identifies the specific parameter to be adjusted, and the adjustment value indicates the amount of parameter change. The queue operates on a first-in, first-out (FIFO) principle to ensure the continuity and stability of the adjustment.

[0096] Understandably, the monitoring of the three key indicators is performed in parallel within each adjustment cycle. The grayscale standard deviation at the crack intersection is calculated by selecting a 5×5 pixel window around the intersection; a smaller standard deviation indicates more uniform grayscale and lower attenuation. The micro-vibration intensity of the branch texture is evaluated by calculating the average offset of the texture position in five consecutive frames; a offset of less than 0.3 pixels is considered to indicate effective micro-vibration suppression.

[0097] For example, in practical applications of laser glass production lines, ambient light intensity fluctuates periodically due to workshop lighting and equipment operation, and vibration frequency is affected by surrounding machinery. By implementing the above technical solution, stable detection performance can be maintained in complex industrial environments. When all three indicators simultaneously meet the requirements, the high-precision imaging detection standard is achieved, at which point the minimum detectable crack width reaches the 10-micron level, providing a reliable guarantee for the quality control of laser glass.

[0098] The above description is merely an example and illustration of the structure of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the structure of the invention or exceed the scope defined in the claims, all of which should fall within the protection scope of the present invention.

Claims

1. A method for real-time detection of laser glass defects, characterized in that, The method includes: Ambient light intensity and vibration frequency fluctuation data are collected to capture initial image data of the laser glass crack area, locate the crack intersection point, analyze the gray value attenuation distribution pattern of the crack intersection point, and determine the spectral interference quantification index, which is the dynamic statistical mean of the spectral interference intensity in each direction of the crack intersection point. Based on the spectral interference quantification index, branch texture features are extracted from the non-uniform attenuation distribution of gray values ​​at the crack intersection, and the range of the area covered by the background light artifacts on the crack edge structure is identified. Based on the range of the masked area and the spatial frequency distribution characteristics of the crack texture, the interference level of the instantaneous spectral characteristics of ambient light on the imaging system is determined. Obtain the combination parameters of the light source polarization angle and wavelength, combine them with the interference level, determine the pulse frequency adjustment vector, optimize the multi-dimensional imaging parameters of crack intersections and branch textures, and determine the gray-scale uniformity index of crack intersections. The branch texture light and shadow tremor suppression rate is calculated by comparing the inter-frame position offset of the branch texture before and after imaging parameter optimization. Based on the branch texture light and shadow tremor suppression effect, combined with the gray-scale uniformity index of the crack intersection point, the remaining degree of the background light artifacts covering the crack edge structure is evaluated. Based on the remaining coverage level, a pulse frequency configuration scheme is determined, and laser glass defect imaging detection results are obtained.

2. The laser glass defect real-time detection method according to claim 1, characterized in that, The process involves collecting ambient light intensity and vibration frequency fluctuation data, capturing initial image data of the laser-cut glass crack area, locating crack intersections, analyzing the grayscale attenuation distribution pattern at the crack intersections, and determining a spectral interference quantification index. This spectral interference quantification index is the dynamic statistical mean of the spectral interference intensity in each direction at the crack intersection, including: Ambient light intensity fluctuation data and vibration frequency change sequence are collected, and the time sequence data of light intensity and vibration acceleration data on the surface of the laser glass are recorded through a photoelectric sensor array. The vibration frequency distribution is determined based on the vibration acceleration data; Based on the vibration frequency distribution, the initial image data of the laser glass crack region is obtained, the crack outline is identified, the pixel coordinates of the crack intersection point are located, and the gray values ​​of the crack intersection point and its neighborhood are obtained. Based on the gray values ​​of the crack intersection and its neighborhood, the gray value difference is calculated, the gray value attenuation amplitude distribution is statistically analyzed, and the directional characteristics of the gray value attenuation distribution pattern are determined. Based on the aforementioned directional characteristics, the correlation coefficient between the grayscale attenuation amplitude and the light intensity time series data is calculated to determine the spectral interference quantification index.

3. The real-time laser glass defect detection method according to claim 1, characterized in that, The step of extracting branch texture features from the non-uniform attenuation distribution of gray values ​​at the crack intersections based on the spectral interference quantization index, and identifying the range of the area obscured by background light artifacts on the crack edge structure, includes: Based on the spectral interference quantification index, the branch texture extension direction in the non-uniform attenuation distribution of gray values ​​at the crack intersection is identified, the angle of the dominant direction of the branch texture is statistically analyzed, the gray value sequence is extracted, and a branch texture feature vector containing texture width, curvature change rate and gray value contrast parameters is constructed. Based on the texture width parameter in the branch texture feature vector, the offset of the branch texture position in adjacent frame images is calculated, the frequency and amplitude distribution of micro-vibration are statistically analyzed, and the light and shadow micro-vibration effect enhancement coefficient is determined. The initial image data of the laser glass crack area is processed by a Gaussian filtering algorithm to determine the distribution of background light artifacts, identify the overlapping area between the background light artifacts and the crack edge, and determine the range of the masked area. The standard deviation of the Gaussian kernel function is dynamically adjusted according to the light and shadow tremor effect enhancement coefficient. Based on the range of the masked area, a spatial frequency distribution analysis of the crack texture is performed to extract the amplitude and phase of the main frequency component of the spectrum. Combined with ambient light intensity fluctuation data, the interference level of the instantaneous spectral characteristics of ambient light on the imaging system is determined.

4. The method for real-time detection of laser glass defects according to claim 1, characterized in that, The process involves acquiring the combined parameters of the light source polarization angle and wavelength, determining the pulse frequency adjustment vector based on the interference level, optimizing the multi-dimensional imaging parameters of the crack intersection point and branch texture, and determining the gray-scale uniformity index of the crack intersection point, including: Obtain the combined parameters of the light source polarization angle and wavelength, including the dominant wavelength value and spectral distribution curve data; Based on the interference level, the interference frequency distribution is determined, and the pulse frequency adjustment vector is calculated, including the frequency change amplitude and phase offset. Based on the pulse frequency adjustment vector, different exposure times and gain parameters are set for the crack intersection region and the branch texture region, and multiple frames of images are acquired and the grayscale information of the multiple frames of images is fused. Based on the grayscale distribution of the fused image, equalization processing is performed to compensate for the non-uniform attenuation of grayscale values. The standard deviation of grayscale values ​​in the neighborhood of the crack intersection is calculated to determine the grayscale uniformity index of the crack intersection.

5. The laser glass defect real-time detection method according to claim 1, characterized in that, The process involves calculating the branch texture light and shadow jitter suppression rate by comparing the inter-frame position offset of the branch texture before and after imaging parameter optimization. Based on the branch texture light and shadow jitter suppression effect, and combined with the gray-level uniformity index of the crack intersection point, the remaining degree of masking of the background light artifact on the crack edge structure is evaluated, including: Calculate the difference in the branch texture position offset before and after optimization to determine the suppression rate; Based on the suppression rate and the gray-scale uniformity index of the crack intersection, a comprehensive quality score is calculated; Based on the comprehensive quality score, the sensitivity parameter for detecting background light artifacts is adjusted, the distribution area of ​​background light artifacts is identified, the set of pixel points of the crack edge contour is extracted, the overlap ratio between the background light artifacts and the crack edge contour is calculated, and the remaining masking degree is determined.

6. The real-time laser glass defect detection method according to claim 5, characterized in that, The step of updating the light source polarization angle and wavelength combination parameters based on the remaining masking degree to determine the pulse frequency configuration scheme includes: Calculate the difference between the remaining masking degree and the threshold, determine the adjustment range of the light source polarization angle and wavelength combination parameters, update the light source polarization angle and wavelength combination parameters, and determine the pulse frequency configuration scheme; The updated crack image is acquired, the edge sharpness value of the crack image is calculated, and the clarity index is determined.

7. The laser glass defect real-time detection method according to claim 1, characterized in that, After determining the pulse frequency configuration scheme, the process includes acquiring a laser glass defect detection image sequence, evaluating the degree of improvement of the laser glass defect detection image sequence relative to the initial image data, and quantifying the multi-dimensional imaging optimization effect of the mesh crack texture; using a cyclic feedback method to process the ambient light intensity and vibration frequency fluctuation data to generate a light source parameter adjustment sequence; executing the light source parameter adjustment sequence to obtain high-precision imaging detection results of laser glass defects.

8. The method for real-time detection of laser glass defects according to claim 1, characterized in that, The acquisition of laser glass defect imaging detection results includes: The pulse frequency configuration scheme is executed to drive the light source to emit a pulse beam, acquire a laser glass defect detection image sequence, calculate the difference in grayscale distribution between the image sequence and the initial image data, and determine the signal-to-noise ratio improvement and contrast enhancement rate. Based on the signal-to-noise ratio improvement and contrast enhancement rate, the geometric feature parameters and photometric feature parameters of the mesh crack texture are extracted, the change in feature parameters before and after optimization is calculated, and the multidimensional imaging optimization effect score is determined. Based on the multidimensional imaging optimization effect score, monitor the changes in ambient light intensity and vibration frequency, calculate the light source polarization angle and wavelength adjustment amount, and generate a light source parameter adjustment sequence. The light source parameter adjustment sequence is executed to calculate the grayscale standard deviation of crack intersections, the inter-frame displacement of branch textures, and the pixel ratio of background light artifacts, thereby determining the laser glass defect imaging detection results.

Images