Glass bead defect detection method and system based on visual detection

Abstract

The invention relates to the technical field of image processing, in particular to a glass bead defect detection method and system based on visual detection. According to the method, glass beads are separated one by one to a detection track and controlled to roll at a constant speed, image sequences containing multiple rotation angles are continuously collected in the rolling process, then whether movement stability, dispersion and brightness meet corresponding standards or not is determined, and when any detection result does not meet the standards, a sorting mechanism immediately removes defective products at the discharging end. According to the method, the glass beads with defects can be accurately and quickly identified and removed.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, in particular to a glass bead defect detection method and system based on visual inspection. BACKGROUND

[0002] Glass beads are widely used as precision components in medical packaging, industrial grinding and decorative materials. Surface and internal defects (such as cracks, bubbles, color difference, wear) directly affect the sealing, strength and optical performance of the product. Traditional manual visual inspection is inefficient and has a high rate of missed detection, and existing machine vision detection technology has obvious limitations: for shape defects, static images cannot capture three-dimensional irregular features; for color defects, environmental light source fluctuations can easily lead to hue misjudgment; for microcracks, conventional optical methods cannot penetrate the glass material to achieve internal imaging. In addition, the high-speed operation of the production line requires the detection system to quickly complete single-bead multi-dimensional analysis, but existing technologies either rely on complex shadow correction models to increase processing time or can only recognize a single defect type, and cannot meet the batch detection efficiency under the premise of ensuring accuracy. SUMMARY

[0003] To solve or at least partially solve the above technical problems, the embodiments of the present application provide a glass bead defect detection method and system based on visual inspection.

[0004] In a first aspect, the present application provides a glass bead defect detection method based on visual inspection, comprising the following steps:

[0005] S1, separate the glass beads to be detected one by one to a detection track, and control the glass beads to roll uniformly on the detection track;

[0006] S2, continuously capture images by an industrial camera group during the rolling of the glass beads to obtain an image sequence containing multiple rotation angles of the glass beads, wherein the industrial camera group contains multiple industrial cameras arranged at intervals along the detection track;

[0007] S3, based on the image sequence, the following detection steps are performed:

[0008] S31, calculate the center of mass displacement vector of the glass beads in the continuous image frames, and determine whether the motion stability meets the stability standard based on the change of the displacement vector;

[0009] S32, extract the color feature value of the glass bead region in each frame of the image sequence, calculate the dispersion of the color feature value, and determine whether the dispersion meets the dispersion standard;

[0010] S33, track the brightness change curve of the highlight reflection region in the image sequence, and determine whether the brightness meets the brightness standard based on the brightness change curve;

[0011] S4, determining that the glass beads have defects when any of the detection steps in S3 does not meet the corresponding standard, and controlling the sorting mechanism to remove the defective glass beads at the discharge end.

[0012] Optionally, in S31, the determination of whether the motion stability meets the stability standard based on the change in the displacement vector includes:

[0013] S311, arranging a distance sensor array on the chute side wall of the detection track to collect distance data between the glass beads and the chute side wall in real time;

[0014] S312, generating a motion trajectory envelope line in combination with the center of mass displacement vector and the distance data;

[0015] S313, determining that the motion stability does not meet the stability standard when it is detected that the envelope line presents periodic fluctuations or continuously deviates from the chute center axis or the center of mass position suddenly changes;

[0016] Optionally, in S31, the detection track has a chute structure that limits the free rolling path of the glass beads, and the contact area between the chute side wall and the glass beads forms a motion constraint boundary.

[0017] Optionally, S33 specifically includes the following steps:

[0018] S331, identifying a highlight reflection area formed by light source irradiation in each image frame during rolling of the glass beads;

[0019] S332, extracting a maximum value of brightness of the highlight reflection area in consecutive image frames to generate a brightness change time series curve;

[0020] S333, calculating a first derivative of the time series curve, and marking a brightness sudden drop point when it is detected that the absolute value of the derivative exceeds a preset mutation threshold;

[0021] S334, counting the number of non-continuous change points in the time series curve, the non-continuous change points including the brightness sudden drop point and a point where the brightness difference between adjacent frames exceeds a step threshold;

[0022] S335, determining that the brightness does not meet the brightness standard when the number of non-continuous change points exceeds a fault tolerance number.

[0023] Optionally, after identifying the highlight reflection area in S331, the following steps are further included:

[0024] S331a, extracting contour geometric features of the highlight reflection area, including contour curvature radius distribution and contour symmetry index;

[0025] S331b, establishing an optical path offset model based on the nominal refractive index of the glass beads, and calculating a theoretical deformation amount of the reflection profile in a defect-free state;

[0026] S331c, comparing the profile curvature radius distribution with the deviation of the theoretical deformation amount, and marking a potential bubble region when a local curvature abnormal mutation is detected and the deviation exceeds an optical tolerance threshold;

[0027] S331d, combining the brightness change timing curve and the deviation, and determining that there is a bubble defect when the deviation continuously exceeds the optical tolerance threshold and the brightness change timing curve of the corresponding region presents a low-frequency high-amplitude oscillation characteristic.

[0028] Optionally, the S32 specifically comprises the following steps:

[0029] S321, for each frame of image in the image sequence, segmenting out a glass bead region;

[0030] S322, converting the glass bead region to an HSV color space, and extracting a pixel value distribution of a hue channel;

[0031] S323, calculating Bhattacharyya distance of the hue distribution histogram between consecutive image frames, and calculating a mean value of the Bhattacharyya distance of three consecutive frames;

[0032] S324, calculating a kurtosis coefficient of the hue distribution histogram in the entire image sequence;

[0033] S325, determining that the dispersion degree does not meet the dispersion degree standard when the mean value of the Bhattacharyya distance exceeds a distribution change threshold or the kurtosis coefficient is lower than a flatness threshold.

[0034] Optionally, the S4 further comprises the following steps before the S4:

[0035] S51, arranging a piezoelectric sensor array on a support structure of the detection track, and collecting stress fluctuation signals generated when the glass beads roll in real time;

[0036] S52, constructing a dynamic stress distribution cloud map based on the stress fluctuation signals;

[0037] S53, extracting a spatial distribution form of a stress concentration region, and marking a potential crack when a linearly extended stress concentration band is detected;

[0038] S54, calculating a fluctuation intensity index of the stress concentration band, the fluctuation intensity index being determined by a ratio of a stress peak value to background noise;

[0039] S55, determining that the micro-crack defect exists when the stress concentration zone satisfies the spatial distribution pattern and the internal structure line of the glass bead coincides with the geometric matching threshold and / or the fluctuation intensity index presents a periodic enhancement characteristic with the rolling angle;

[0040] The S4 further comprises:

[0041] Controlling the sorting mechanism to reject the defective glass beads at the discharge end when it is determined that the glass beads have the micro-crack defect.

[0042] Optionally, the S52 specifically comprises the following steps:

[0043] S521, establishing a three-dimensional coordinate system according to the spatial coordinates of the piezoelectric sensor array, wherein the Z-axis is perpendicular to the detection track plane, and the X-axis is along the rolling direction of the glass bead;

[0044] S522, performing time-frequency decomposition on the stress fluctuation signal, extracting a fundamental frequency component synchronized with the glass bead rotation period to generate a static stress distribution layer, and separating a harmonic component higher than the fundamental frequency to generate a dynamic stress fluctuation layer;

[0045] S523, mapping the static stress distribution layer to the lower half of the glass bead surface model;

[0046] S524, marking an abnormal harmonic area in the dynamic stress fluctuation layer, and generating an initial stress concentration zone mark when the harmonic energy is concentrated in the local meridian direction and lasts for more than 3 rotation periods;

[0047] S525, fusing the static stress distribution layer and the dynamic stress fluctuation layer, superimposing the initial stress concentration zone mark, and outputting the dynamic stress distribution cloud chart.

[0048] Optionally, in the S322, the conversion of the glass bead region to the HSV color space comprises the following steps:

[0049] S3221, based on the spherical geometric characteristics of the glass bead and the light refraction law, dividing the glass bead region into a center region and an edge region, wherein the edge region includes the reflection interface of the glass bead and the detection track;

[0050] S3222, performing HSV conversion on the center region and the edge region respectively to generate a center hue distribution histogram and an edge hue distribution histogram;

[0051] S3223, integrating the center hue distribution histogram and the edge hue distribution histogram in a weighted fusion manner, wherein the weight coefficient of the center region is higher than that of the edge region;

[0052] S3224, output the hue channel pixel value distribution as the fused hue distribution histogram.

[0053] Optionally, in the S3221, the dividing the glass bead region into a center region and an edge region further comprises the following steps:

[0054] S3221a, determining a contact point refraction angle based on the chute structure parameters of the detection track; wherein the chute structure parameters include a chute inclination angle and a chute surface roughness;

[0055] S3221b, calculating an expected hue shift of the edge region according to the refractive index of the glass bead material;

[0056] S3221c, after the S3222 generates the edge hue distribution histogram, applying the expected hue shift to compensate and correct the edge hue distribution histogram;

[0057] S3221d, inputting the compensated and corrected edge hue distribution histogram to the weighted fusion of the S3223.

[0058] In a second aspect, the present application also provides a glass bead defect detection system based on visual detection, comprising:

[0059] A rolling module for separating the glass beads to be detected one by one to the detection track and controlling the glass beads to roll uniformly on the detection track;

[0060] An acquisition module for continuously acquiring images during the rolling of the glass beads by an industrial camera group to obtain an image sequence containing multiple rotation angles of the glass beads, wherein the industrial camera group contains multiple industrial cameras arranged at intervals along the detection track;

[0061] A detection module for performing the following detection steps based on the image sequence:

[0062] Calculating the center of mass displacement vector of the glass bead in the continuous image frame, and determining whether the motion stability meets the stability standard based on the change of the displacement vector;

[0063] Extracting the color feature value of the glass bead region of each frame in the image sequence, calculating the dispersion of the color feature value, and determining whether the dispersion meets the dispersion standard;

[0064] Tracking the brightness change curve of the highlight reflection region in the image sequence, and determining whether the brightness meets the brightness standard based on the brightness change curve;

[0065] The rejection module is used to determine that the glass bead has a defect and control the sorting mechanism to reject the defective glass bead at the discharge end when any detection step in the detection module fails to meet the corresponding standard.

[0066] The method and system provided by this invention have the following beneficial effects:

[0067] First, by separating each glass bead individually and rolling it at a uniform speed on a detection track, the movement of each bead is ensured to be stable and controllable. During the rolling process, industrial camera arrays are spaced along the track, continuously capturing image sequences from multiple rotation angles. This detection method fully exposes the defect characteristics of the glass beads during movement. For example, irregularly shaped glass beads will exhibit abrupt displacement or directional deviation, surface wear areas will show a sudden drop in brightness during rolling, and glass beads with abnormal colors will show abnormal fluctuations in hue distribution. This full-angle coverage detection method significantly improves the comprehensiveness of defect capture.

[0068] Secondly, dedicated detection logic is designed for different defect types: shape detection utilizes centroid displacement vector analysis to assess motion stability, accurately identifying fluctuations in the trajectory of glass beads caused by defects such as ellipses or protrusions; color detection analyzes dispersion using the HSV hue space to effectively distinguish between overall color differences and localized color spots; and surface detection tracks brightness changes in high-brightness reflective areas, exhibiting high sensitivity to defects such as scratches and wear. These three types of detection are performed simultaneously without interference, ensuring that each type of defect can be independently identified.

[0069] Finally, the system employs a tiered judgment method: when any of the motion stability, hue dispersion, or brightness variation exceeds the standard, the sorting mechanism is immediately controlled to reject defective products at the discharge end. This method avoids the multi-round screening process of traditional sorting equipment, significantly shortening the detection cycle. Simultaneously, since the detection process relies entirely on the motion characteristics and optical performance of the glass beads themselves, no additional detection medium or complex calibration is required, reducing equipment maintenance costs and avoiding secondary damage to the glass bead surface.

[0070] Furthermore, the design of uniform rolling and multi-camera detection allows the system to adapt to the detection of glass beads of different sizes without the need to adjust hardware parameters for product specifications. In actual production line operation, even for high-speed rolling glass beads, characteristics such as displacement changes, hue fluctuations, and brightness jumps can still be stably captured. This method ensures accuracy while possessing good anti-interference capabilities. Attached Figure Description

[0071] Figure 1 A schematic flowchart of a glass bead defect detection method based on visual inspection provided in an embodiment of the present invention;

[0072] Figure 2 A schematic diagram of a detection system structure provided in an embodiment of the present invention;

[0073] Figure 3 A detailed flowchart of step S31 is provided for an embodiment of the present invention;

[0074] Figure 4 A detailed flowchart of step S33 is provided for an embodiment of the present invention;

[0075] Figure 5 A schematic diagram of a glass bead defect detection method based on vision inspection provided in an embodiment of the present invention;

[0076] Figure 6 A detailed flowchart of step S32 is provided for an embodiment of the present invention;

[0077] Figure 7 A schematic diagram of a glass bead defect detection method based on vision inspection provided in an embodiment of the present invention;

[0078] Figure 8 A detailed flowchart of step S52 is provided for an embodiment of the present invention;

[0079] Figure 9 A detailed flowchart of step S322 is provided for an embodiment of the present invention;

[0080] Figure 10 This is a detailed flowchart of step S3221 provided in an embodiment of the present invention.

[0081] Explanation of reference numerals in the attached figures:

[0082] 100. Inspection track; 200. Industrial camera. Detailed Implementation

[0083] To make the objectives, technical solutions, and advantages of this application clearer, specific embodiments of this application will be described in further detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are merely for explaining this application and not for limiting it. It should also be noted that, for ease of description, only the parts relevant to this application are shown in the drawings, not all of them. Before discussing exemplary embodiments in more detail, it should be mentioned that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although the flowcharts describe operations (or steps) as sequential processes, many of these operations can be performed in parallel, concurrently, or simultaneously. Furthermore, the order of the operations can be rearranged. The process can be terminated when its operation is completed, but may also have additional steps not included in the drawings. The process can correspond to a method, function, procedure, subroutine, subprogram, etc.

[0084] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0085] Figure 1 This invention provides a schematic flowchart of a visual inspection-based glass bead defect detection method, which includes the following steps:

[0086] S1. Separate the glass beads to be tested one by one into the testing track, and control the glass beads to roll at a uniform speed on the testing track;

[0087] S2. Continuous image acquisition is performed by an industrial camera group during the rolling process of the glass bead to obtain an image sequence containing multiple rotation angles of the glass bead, wherein the industrial camera group includes multiple industrial cameras set at intervals along the detection track.

[0088] S3. Perform the following detection steps based on the image sequence:

[0089] S31. Calculate the centroid displacement vector of the glass bead in consecutive image frames, and determine whether the motion stability meets the stability criterion based on the change of the displacement vector.

[0090] S32. Extract the color feature values ​​of the glass bead region in each frame of the image sequence, calculate the dispersion of the color feature values, and determine whether the dispersion meets the dispersion standard.

[0091] S33. Track the brightness change curve of the bright reflection area in the image sequence, and determine whether the brightness meets the brightness standard based on the brightness change curve;

[0092] S4. If any detection step in S3 fails to meet the corresponding standard, it is determined that the glass bead is defective and the sorting mechanism is controlled to remove the defective glass bead at the discharge end.

[0093] Figure 2 This is a schematic diagram of a detection system structure provided in an embodiment of the present invention. Figure 2The diagram illustrates the detection track 100 and industrial cameras, serving only as a general structural illustration. After the glass beads enter the detection system, they are first separated individually by a distribution turntable, ensuring each bead enters the detection track independently. The detection track can employ a V-shaped chute structure, allowing the glass beads to maintain a uniform rolling speed under gravity. The chute surface is polished to reduce the impact of frictional resistance on the rolling speed, while side baffles restrict lateral displacement of the glass beads, ensuring a linear and stable motion trajectory. During this movement, the industrial camera group, equidistantly arranged along the detection track, activates a high-speed shooting mode. The camera frame rate matches the rolling speed of the glass beads, ensuring that clear images are captured from multiple rotation angles as each glass bead passes through the detection area, forming a complete image sequence. A ring-shaped LED light source can be installed above the track to provide a uniform and stable lighting environment.

[0094] During the image processing stage, the system simultaneously initiates three types of detection. In the motion stability detection step, consecutive image frames are first preprocessed. The outline of the glass bead is extracted using a background subtraction algorithm, and the centroid coordinates of the glass bead in each frame are calculated. The difference in centroid coordinates between adjacent frames forms a displacement vector, and the system records the change in the direction angle and the magnitude of the displacement vector change. For example, when a continuous fluctuation in the direction angle exceeds 10 degrees or a sudden increase in displacement of 50% is detected, it indicates that the glass bead has elliptical deformation or surface protrusions, causing instability in the rolling trajectory. At this time, the system generates a motion anomaly marker, which is used to control subsequent sorting actions.

[0095] In the color detection step, an HSV color space conversion strategy can be adopted. First, the glass bead region of each frame image is segmented to eliminate background interference. After the segmented region is converted to HSV space, the pixel value distribution of the hue channel is analyzed in detail. The system establishes a hue distribution histogram and calculates the hue distribution difference between consecutive frames using Bach distance. When a specific region's hue value continuously deviates from the main distribution range, or the overall hue distribution exhibits a multi-peak pattern, it indicates the presence of local color spots or uneven material texture. Hue anomaly data is stored in the detection database in real time and compared with preset color difference tolerance values.

[0096] In the surface inspection step, the changing characteristics of the high-brightness reflective areas can be focused on. Under ring light illumination, specular reflective spots are formed on the surface of the glass bead. The system locates the high-brightness reflective areas in each frame of the image through threshold segmentation. A time-series curve is generated by tracking the maximum brightness of the spots in consecutive frames. When the curve shows a steep drop or a step change, it indicates that the spot has passed through a scratch or pit area, causing a sudden change in reflected light intensity. The system counts the number of abnormal brightness changes per unit time. When the abnormal frequency exceeds the tolerance limit, the surface is determined to have wear defects.

[0097] During the result aggregation phase, the output signals from the three testing steps are connected to a logic OR gate circuit. A defect flag issued in any step will immediately trigger the sorting action. The sorting mechanism is located at the end of the testing track and employs a high-pressure gas nozzle design. When the defect flag is activated, the defective glass beads are blown into the waste channel. Qualified products fall into the collection container due to inertia, completing the sorting process.

[0098] This implementation significantly improves defect recognition rates through the synergistic analysis of motion and optical characteristics. Shape defects are captured because they cause motion instability; for example, elliptical glass beads exhibit periodic trajectory deviations during rolling. Color anomalies are effectively distinguished from ambient light interference and true color differences through hue space analysis. Surface scratches create unique brightness fingerprints by disrupting the continuity of light spots. The detection process fully utilizes the physical properties of the glass beads themselves, eliminating the need for contact probes or complex spectral analysis, thus avoiding secondary damage and reducing equipment complexity. The V-shaped structure of the detection track constrains the glass beads, enhancing the predictability of their motion trajectory and providing a physical basis for displacement vector analysis. An equidistant layout of multiple cameras ensures no surface features are missed, and a ring light source eliminates shadow interference, making color detection more reliable. The rapid-response design of the sorting mechanism adapts to high-speed production lines, enabling streamlined detection and sorting operations.

[0099] Figure 3 This is a detailed flowchart of step S31 provided in an embodiment of the present invention. In some implementations, in step S31, determining whether the motion stability meets the stability criterion based on the change in the displacement vector includes:

[0100] S311. A distance sensor array is installed on the side wall of the slide of the detection track to collect the distance data between the glass bead and the side wall of the slide in real time.

[0101] S312. Combine the centroid displacement vector and spacing data to generate the motion trajectory envelope;

[0102] S313. When the envelope is detected to exhibit periodic fluctuations or continuously deviate from the center axis of the slide or the center of mass position changes abruptly, it is determined that the motion stability does not meet the stability standard.

[0103] The groove structure of the detection track restricts the free rolling path of the glass bead, and the contact area between the sidewall of the groove and the glass bead forms a motion constraint boundary.

[0104] In the glass bead inspection process, the V-shaped chute structure of the inspection track plays a crucial role when the glass bead enters the motion stability detection stage. A distance sensor array, consisting of multiple high-precision laser rangefinders, is installed on the sidewall of the chute, arranged equidistantly along the length of the chute. Each sensor emits a laser beam in real time and receives the reflected signal, calculating the distance between the glass bead surface and the chute sidewall. The sampling frequency of the sensor array is synchronized with the frame rate of the industrial camera, ensuring that the distance data is aligned with the timestamps of the image sequence. As the glass bead rolls at a constant speed, the sensor array captures the changes in its gap with the chute, generating a continuous distance variation curve.

[0105] After preprocessing, the image sequences captured by the industrial camera group determine the geometric center point of the glass bead using a contour extraction algorithm, and calculate the centroid coordinates of each frame. The displacement vector is derived from the difference in centroid coordinates between adjacent frames, containing two components: orientation angle and displacement. These data, along with spacing data, are input into the trajectory analysis module. The system uses a coordinate transformation method to map the centroid position to the groove coordinate system, where the groove's central axis is defined as the reference datum. Combining the spacing data and displacement vector, a motion trajectory envelope is constructed: discrete centroid points are connected into a smooth curve using an interpolation algorithm, while the spacing variation data is superimposed as the envelope boundary. The envelope reflects the glass bead's motion trajectory in three-dimensional space, including the longitudinal rolling path and lateral offset magnitude.

[0106] When detecting envelope anomalies, the system scans the envelope data to detect periodic fluctuations: when the curve exhibits sinusoidal oscillations with a frequency matching the glass bead's rotation period, it indicates that elliptical deformation defects are causing rolling imbalance. Simultaneously, the system monitors the deviation between the envelope and the groove's central axis: a sustained deviation exceeding a threshold angle is identified as trajectory deviation caused by surface protrusions or pits. Furthermore, abrupt changes in the center of mass position can be detected by calculating the second derivative of the displacement vector: a sudden increase in the absolute value of the derivative indicates instantaneous bouncing of the glass bead, typically caused by internal cracks or external deposits. Upon triggering any anomaly, the system immediately flags the motion stability as failing to meet the standard.

[0107] This implementation significantly improves the robustness of defect detection. The motion state of the glass bead under the constraint of the groove is precisely quantified; shape defects such as elliptical deformation are exposed through periodic fluctuations, and surface unevenness is revealed through trajectory offset. The distance sensor array provides additional physical dimension data, compensating for the limitations of image analysis, such as reducing false positives in high-speed rolling scenes. Envelope analysis intuitively displays defect characteristics, facilitating rapid system response.

[0108] The implementation process fully utilizes the rolling characteristics of glass beads, eliminating the need for contact probes or additional excitation sources. The chute structure design takes into account the spherical geometry of the glass beads, ensuring predictable rolling paths. The fusion processing of sensor data and image data avoids errors from single sensors, enhancing the system's anti-interference capabilities. For example, even under fluctuating lighting conditions, spacing data can still reliably detect lateral offset. Finally, motion stability detection is implemented as an independent module, operating in parallel with color and surface detection without increasing overall latency. Defective glass beads are efficiently rejected, while qualified beads continue to flow to the collection end, maintaining continuous production line operation.

[0109] Figure 4 This is a detailed flowchart of step S33 provided for an embodiment of the present invention. In some embodiments, S33 specifically includes the following steps:

[0110] S331. During the rolling of the glass bead, identify the bright reflection areas formed by the light source in each frame of the image;

[0111] S332. Extract the maximum brightness of the bright reflection area in consecutive image frames and generate a brightness change time-series curve.

[0112] S333. Calculate the first derivative of the time series curve. When the absolute value of the derivative exceeds the preset abrupt change threshold, mark it as a point of sudden brightness drop.

[0113] S334. Count the number of discontinuous change points in the time-series curve. Discontinuous change points include points where brightness drops sharply and points where the brightness difference between adjacent frames exceeds the step threshold.

[0114] S335. When the number of discontinuous change points exceeds the fault tolerance limit, the brightness is determined to not meet the brightness standard.

[0115] During surface defect detection, a ring-shaped LED light source is installed above the detection track, illuminating the rolling glass beads at a fixed incident angle. As the glass beads roll at a constant speed within the chute, a specular reflection area forms on their surface. An industrial camera system captures image sequences at a frame rate matched to the rolling speed. The system first preprocesses each frame to locate the bright reflective areas: the image is converted to the HSV color space, a threshold range is set for the saturation channel to filter out ambient light interference, and a segmentation threshold is set for the brightness channel to extract the contour of the reflective area. This contour area serves as the basic unit for brightness analysis.

[0116] The system then tracks the brightness characteristics in consecutive image frames. The maximum pixel brightness within the highlighted region of each frame is extracted, forming a time-series curve of brightness variation arranged sequentially. To enhance the curve's variation characteristics, the system performs smoothing processing on the raw data: a sliding window averaging algorithm can be used, with the window width set to an integer multiple of the glass bead's rotation period, preserving effective signals while suppressing high-frequency noise. The processed brightness variation time-series curve is input into the variation analysis module, which calculates its first derivative and sets a sudden change threshold. When the absolute value of the derivative exceeds this threshold, it indicates a discontinuous brightness jump (such as a sudden drop in brightness when a light spot crosses a deep scratch), and the system marks this as a brightness drop point. Simultaneously, the system compares the brightness difference between adjacent frames; when the difference exceeds the step detection threshold, it is marked as a step change point.

[0117] The defect determination phase involves statistically analyzing abnormal events within a single rolling cycle. The system uses the complete rotation of the glass bead as the cycle unit, counting the cumulative number of all brightness drops and abrupt changes within that cycle. When the number of discontinuous abnormal points exceeds the tolerance limit, a surface wear defect is determined. This determination method considers accidental interference in actual production: the instantaneous adhesion of tiny dust particles to the chute may cause a single brightness jump, but continuous wear at multiple locations will produce dense abnormal points. The determination results are transmitted to the sorting control system in real time for subsequent rejection actions.

[0118] The implementation process fully utilizes the motion characteristics of the glass beads. Uniform rolling ensures a constant spot speed, allowing the time axis of the brightness curve to be equivalently converted into spatial position information. Surface scratches form characteristic "dimples" on the curve, while dense wear manifests as continuous sawtooth-like fluctuations. The detection system maintains high sensitivity to minute scratches (longer than the glass bead diameter) because the brightness change is significant when the spot crosses the scratch. Simultaneously, the system reduces the requirements for the cleanliness of the glass bead surface, preventing accidental stains from triggering false positives.

[0119] Under this scheme, tiny scratches that are easily overlooked in static images are captured due to sudden changes in brightness during the rolling process; and wear points with random locations are identified without omission through full-surface scanning.

[0120] Figure 5 This is a schematic flowchart of another vision-based glass bead defect detection method provided by an embodiment of the present invention. In some embodiments, after identifying the high-brightness reflection area in S331, the method further includes the following step:

[0121] S331a. Extract the contour geometric features of the high-brightness reflection area, including the contour curvature radius distribution and contour symmetry index;

[0122] S331b: Based on the nominal refractive index of glass beads, establish an optical path offset model and calculate the theoretical deformation of the reflection profile in a defect-free state.

[0123] S331c: Compare the deviation of the contour curvature radius distribution from the theoretical deformation. When a local curvature abnormal change is detected and the deviation exceeds the optical tolerance threshold, it is marked as a potential bubble region.

[0124] S331d, combining the brightness change timing curve and the deviation, when the deviation continuously exceeds the optical tolerance threshold and the brightness change timing curve of the corresponding area shows low-frequency high-amplitude oscillation characteristics, it is determined that there is a bubble defect.

[0125] During bubble defect detection, as the glass bead rolls at a constant speed on the detection track, a ring light source illuminates its surface, creating a bright reflective area. Images captured by the industrial camera array are transmitted to the processing system, which first locates the contour of the bright reflective area. By performing curve fitting on the sampling points at the contour edge, the local radius of curvature distribution is calculated. Simultaneously, the curvature difference on both sides of the contour's axis of symmetry is measured to generate a contour symmetry index. These geometric features serve as the initial input data for bubble detection.

[0126] The system constructs an optical model based on the nominal refractive index of the glass bead material. This model, based on the law of light refraction, simulates the propagation path of light within the glass sphere in a defect-free state. Through ray tracing calculations, it predicts the theoretical deformation of the ideal spherical reflection profile. The model output includes a mapping table between the reflection point location and the theoretical curvature value, which is dynamically generated based on the glass bead diameter.

[0127] During the contour comparison stage, the system matches the actual measured radius of curvature distribution with the theoretical mapping table. For each detection point, the relative deviation between the measured curvature and the theoretical value is calculated. When a local curvature abrupt change occurs in a specific area (such as a sudden increase or decrease in curvature value), and the deviation at that point continuously exceeds the optical tolerance range, the system marks that area as a potential bubble region. This process takes into account the angular changes during the rolling of the glass bead and performs multi-frame verification on the same physical location.

[0128] Brightness change time-series curves are used in the judgment process. The system analyzes the brightness curve characteristics of the marked area. Due to the light scattering effect, bubble defects produce unique low-frequency, high-amplitude oscillations: the oscillation frequency is an integer multiple of the glass bead's rotation period, and the amplitude is significantly higher than the abrupt changes caused by surface scratches. When the contour deviation continuously exceeds the standard and the brightness oscillation in the corresponding area conforms to the characteristic pattern, the system confirms the existence of bubble defects.

[0129] This implementation fully utilizes the optical properties of glass beads. The transparent material makes the effect of bubbles on the light path predictable: abrupt changes in refractive index at the bubble interface alter the reflected light path, leading to abnormal contour curvature; multiple reflections inside the bubble induce scattering, producing characteristic brightness oscillations. The uniform rolling design of the detection track ensures that defect features appear periodically, facilitating the system's capture of regular signals.

[0130] This method can effectively distinguish bubble defects from other optical interferences. The contour distortion caused by bubbles has spatial continuity and exhibits regular repetition during rolling; while temporary deposits such as dust only produce single-frame anomalies. Brightness oscillation characteristics further reduce the false positive rate, and the low-frequency patterns generated by bubble scattering are difficult to simulate by other defects. The detection sensitivity for microbubbles (diameter smaller than the glass bead radius) meets industrial requirements, and the requirements for the cleanliness of the glass bead surface are relatively low.

[0131] Figure 6 This is a detailed flowchart of step S32 provided for an embodiment of the present invention. In some embodiments, S32 specifically includes the following steps:

[0132] S321. For each frame of the image sequence, segment out the glass bead region;

[0133] S322. Convert the glass bead area to the HSV color space and extract the pixel value distribution of the hue channel;

[0134] S323. Calculate the Bartholomew distance of the hue distribution histogram between consecutive image frames, and calculate the mean Bartholomew distance of three consecutive frames.

[0135] S324. Calculate the kurtosis coefficient of the hue distribution histogram in the entire image sequence;

[0136] S325. When the mean of the Barthel distance exceeds the distribution variation threshold or the kurtosis coefficient is lower than the flatness threshold, the dispersion is determined to not meet the dispersion standard.

[0137] During the glass bead color defect detection process, a ring-shaped LED light source above the detection track provides stable color temperature illumination, and the industrial camera group uses fixed white balance parameters. As the glass bead passes through the detection area at a constant speed, multiple industrial cameras simultaneously capture image sequences from different angles. The glass bead region in each frame is extracted using dynamic segmentation technology: first, a rolling trajectory prediction model is established to predict the position of the glass bead based on its movement speed; second, edge gradient detection combined with region growing methods is used to lock the glass bead boundary; finally, morphological closing operations are used to fill the holes caused by surface reflection, forming a complete glass bead region mask.

[0138] The segmented glass bead regions undergo color space conversion. The system converts RGB pixel values ​​to HSV space. In HSV space, hue channels are extracted independently and a hue distribution histogram is generated. The system pays special attention to the clustering of hue values ​​within specific angular ranges, automatically filtering low-saturation areas to prevent grayscale pixels from interfering with hue analysis.

[0139] The system selects three frames of images in chronological order as the analysis unit and calculates the distribution difference of the hue histogram between adjacent frames. The difference is measured using a distribution similarity index in statistics, and the Bach distance is obtained through integration. The arithmetic mean of three consecutive Bach distances is calculated to reflect the intensity of short-term hue fluctuations. Simultaneously, the system statistically analyzes the hue distribution morphology characteristics of the entire image sequence and calculates the kurtosis coefficient of the distribution curve, which characterizes the degree of concentration of hue value distribution.

[0140] When the mean Bartholin's distance exceeds the distribution variation threshold, it indicates that localized color spots or impurities may appear on the glass bead surface; when the kurtosis coefficient is below the flatness threshold, it indicates that the overall hue distribution is too dispersed, possibly indicating material inhomogeneity or overall color difference. These two abnormal conditions trigger defect marking independently, avoiding missed detections by a single detection method.

[0141] This implementation optimizes the processing flow for the transparency of glass. Glass is sensitive to hue changes, and subtle color differences are amplified and identified in the HSV color space. The system utilizes the rolling characteristics of glass beads: uniform motion ensures a constant image acquisition interval for each frame, and time window analysis corresponds to the actual physical location; multi-angle acquisition covers the entire surface, avoiding blind spots from fixed viewing angles. The stable design of the light source system eliminates ambient light interference, and the fixed white balance setting of the industrial camera ensures consistent hue acquisition.

[0142] The implementation process is fully adapted to the production line environment. The processing unit adopts a pipeline architecture, with image acquisition and processing performed in parallel. Hue distribution histogram analysis requires only simple arithmetic operations, resulting in a light computational load and meeting the requirements of high-speed detection. For glass beads made of special materials (such as colored glass), the system automatically adjusts the hue analysis range to avoid filtering out effective signals. Detection results are transmitted to the sorting system in real time via shared memory and processed in parallel with motion detection and surface detection results.

[0143] Traditional methods are susceptible to misjudgments due to reflection interference. This solution separates hue information using HSV space to avoid the impact of brightness fluctuations; it also distinguishes between instantaneous reflections and true color changes through time-series analysis. The glass beads are captured multiple times at each angle during rolling, and hue anomalies are verified across multiple frames, significantly reducing the false alarm rate. After receiving defect signals, the sorting system accurately rejects products with color differences at the discharge end, ensuring color consistency in the finished product.

[0144] Figure 7 This is a schematic flowchart of another vision-based glass bead defect detection method provided by an embodiment of the present invention. In some embodiments, the following steps are included before S4:

[0145] S51. A piezoelectric sensor array is installed on the support structure of the detection track to collect stress fluctuation signals generated when the glass beads roll in real time.

[0146] S52. Construct a dynamic stress distribution cloud map based on stress fluctuation signals;

[0147] S53. Extract the spatial distribution pattern of stress concentration areas, and mark potential cracks when linearly extending stress concentration zones are detected.

[0148] S54. Calculate the fluctuation intensity index of the stress concentration zone. The fluctuation intensity index is determined by the ratio of the stress peak value to the background noise.

[0149] S55. When the stress concentration zone satisfies the condition that the spatial distribution pattern coincides with the internal structural line of the glass bead exceeding the geometric matching threshold and / or the fluctuation intensity index shows a periodic enhancement characteristic with the rolling angle, it is determined that there is a microcrack defect.

[0150] S4 also includes:

[0151] When it is determined that there are microcracks in the glass beads, the sorting mechanism is controlled to remove the defective glass beads at the discharge end.

[0152] An array of piezoelectric sensors is embedded within the support structure of the detection track. This array can consist of multiple piezoelectric ceramic units, equidistantly arranged on the track substrate along the length of the track. Each sensor unit is covered with an elastic conductive layer, making direct contact with the glass bead while maintaining electrical isolation. When the glass bead rolls at a constant speed within the track, its gravity and inertia generate dynamic pressure at the contact point. The piezoelectric sensor converts this pressure into an electrical charge signal. A signal amplifier converts the charge signal into a standard voltage signal, which is then transmitted to the acquisition system via a shielded cable.

[0153] First, a hardware filter is used to eliminate power frequency interference. Second, a sliding window integration is used to enhance the effective signal characteristics. The processed signal enters the three-dimensional reconstruction module, which establishes a spatial coordinate system based on the physical coordinates of the sensor array: the origin is set at the entrance of the detection track, the X-axis is parallel to the rolling direction, the Y-axis is perpendicular to the sidewall of the chute, and the Z-axis is vertically upward. The system uses the rotation period of the glass bead as the time unit, mapping the stress data in each period to the corresponding spatial position.

[0154] The glass bead position data acquired by the industrial camera array is synchronized with the stress signal timestamp. Based on the center coordinates of the glass bead in the image frame, the system calculates the theoretical position of the contact point at the current moment. The theoretical position of the contact point is matched with the actual position of the sensor, and a dynamic stress distribution cloud map is generated through bilinear interpolation.

[0155] The system scans the spatial distribution pattern of dynamic stress distribution cloud map and calculates the stress gradient vector of each pixel. When the gradient directions of adjacent pixels are continuous and the amplitude exceeds a threshold, they are marked as potential stress concentration zones.

[0156] Peak data from stress concentration zones are extracted, and the average stress value of the region over three rotation cycles is used as the background noise benchmark. The fluctuation intensity index is defined as the ratio of peak stress to background noise; this index eliminates systematic errors caused by differences in sensor sensitivity.

[0157] The geometric matching condition requires that the spatial orientation of the stress concentration zone coincides with the internal structural lines of the glass bead (such as molecular crystal orientation or thermal stress lines) to exceed a preset threshold. This threshold is calibrated through destructive testing of standard samples. The dynamic characteristic condition detects the periodic change of the fluctuation intensity index: when the glass bead passes through the contact point at a specific angle, the index exhibits a regular peak, indicating that the defect location is fixed. Meeting any condition triggers a defect flag. After receiving the defect signal, the sorting control unit activates the sorting mechanism at the discharge end to remove the defective glass bead.

[0158] This implementation combines physical sensing with motion characteristics to solve the challenge of detecting internal defects in transparent materials. Microcracks generate characteristic stress concentration zones in the stress field, and their spatial distribution matches the fracture mechanics properties of the material. Dynamic fluctuation characteristics eliminate false triggering factors: stress fluctuations generated by temporary attachments do not have periodic characteristics, while internal cracks continuously generate stress concentrations of the same pattern during rolling. The system maintains effective sensitivity to fine cracks and is unaffected by the surface finish of the glass beads.

[0159] Figure 8 This is a detailed flowchart of step S52 provided for an embodiment of the present invention. In some embodiments, S52 specifically includes the following steps:

[0160] S521. Establish a three-dimensional coordinate system based on the spatial coordinates of the piezoelectric sensor array, where the Z-axis is perpendicular to the detection track plane and the X-axis is along the rolling direction of the glass beads.

[0161] S522. Perform time-frequency decomposition on the stress fluctuation signal, extract the fundamental frequency component that is synchronized with the rotation period of the glass bead, generate a static stress distribution layer, and separate the resonant components higher than the fundamental frequency to generate a dynamic stress fluctuation layer.

[0162] S523. Map the static stress distribution layer onto the lower hemisphere of the glass bead surface model;

[0163] S524. Mark the abnormal resonance zone in the dynamic stress fluctuation layer. When the resonance energy is concentrated in the local meridian direction and lasts for more than 3 rotation cycles, the initial mark of the stress concentration zone is generated.

[0164] S525: Integrate the static stress distribution layer and the dynamic stress fluctuation layer, overlay the initial marker of the stress concentration zone, and output a dynamic stress distribution cloud map.

[0165] After installing the piezoelectric sensor array on the support structure of the detection track, the system begins to construct a dynamic stress distribution cloud map. The piezoelectric sensor array consists of multiple piezoelectric ceramic units, equidistantly arranged along the length of the chute. Each unit is embedded inside the chute substrate, with its surface covered by an elastic conductive layer to directly contact the glass beads. When the glass beads roll uniformly on the detection track, their gravity and inertia generate dynamic pressure at the contact point, which the sensor converts into an electrical charge signal. The signal amplifier converts the microvolt-level charge signal into a standard voltage output, which is transmitted to the processing unit via a shielded cable to avoid power frequency interference affecting data accuracy. The spatial layout of the sensor array is determined based on the physical structure of the detection track, establishing a three-dimensional coordinate system: the origin is set at the center point of the track entrance, the X-axis is parallel to the rolling direction of the glass beads, the Y-axis is perpendicular to the sidewall of the chute, and the Z-axis is vertically upward. The coordinate system definition process utilizes the reference surface of the track fixing bracket for calibration, ensuring that the coordinate axes are strictly aligned with the track's geometric features.

[0166] The acquired stress fluctuation signal undergoes time-frequency decomposition processing. The signal processing unit first applies a hardware filter to eliminate high-frequency noise, and then uses a sliding window integration to enhance signal characteristics. The decomposition process separates the signal into two components: the fundamental frequency component corresponds to the stable stress layer of the glass bead's rotation period, and the component synchronized with the rotation frequency is extracted using a bandpass filter; the resonant component corresponds to high-frequency fluctuations, and the portion higher than the fundamental frequency is separated using a high-pass filter. The fundamental frequency component generates a static stress distribution layer, reflecting the steady-state pressure when the glass bead rolls uniformly; the resonant component generates a dynamic stress fluctuation layer, capturing transient anomalies. The system uses the complete rotation of the glass bead as the time unit to ensure that component extraction is synchronized with physical motion.

[0167] The static stress distribution layer is mapped to the lower hemisphere of the glass bead surface model. The glass bead surface model is generated based on standard spherical geometry, with the diameter parameter provided in real-time by the image segmentation module. The mapping rule uses a nearest neighbor association method: the static stress value of the sensor contact point is directly associated with the nearest region on the glass bead surface, and uncovered areas are filled with data gaps using bilinear interpolation. The interpolation process considers changes in surface curvature to ensure a smooth transition in stress distribution. After mapping, the lower hemisphere is divided into grid cells, with each cell storing the corresponding stress value, forming the visualized data of the static distribution layer.

[0168] Anomalous resonance characteristics are analyzed within a dynamic stress fluctuation layer. The system scans resonance component data to detect energy concentration regions. When resonance energy continuously accumulates along a local meridian for a time span exceeding three rotation cycles, the system generates initial markers for stress concentration zones. The analysis process employs a gradient calculation method: the gradient vector of the spatial distribution of resonance energy is calculated; when the gradient direction is consistent and the amplitude exceeds a threshold, it is determined to be a persistent anomaly. The marked data records the location coordinates and energy intensity as input for subsequent fusion.

[0169] The static stress distribution layer and the dynamic stress fluctuation layer are fused together, and initial markers for stress concentration zones are superimposed to output the final dynamic stress distribution contour map. The fusion process employs a weighted addition method: the static layer is given a higher weight to highlight the overall stress distribution; the dynamic layer is given a lower weight to suppress random noise. When superimposing markers, the initial marker positions are mapped to the corresponding areas of the contour map.

[0170] This implementation process fully utilizes the structural characteristics of the detection track. The V-shaped design of the chute constrains the rolling path of the glass beads, ensuring stable and measurable stress signals. The sensor array density matches the glass bead size, ensuring that each contact point is covered by at least two sensors, improving data reliability. The processing flow avoids complex models and runs efficiently on an embedded platform: time-frequency decomposition requires only filtering operations, mapping relies on simple geometric interpolation, and fusion uses arithmetic superposition. The system is adaptable to high-speed production line environments, with extremely low signal processing latency, and runs in parallel with the image acquisition thread without increasing the overall burden. The uniform rolling characteristics of the glass beads make stress changes periodic, facilitating the capture of abnormal patterns; the spherical geometry ensures a universal surface model, and glass beads of different diameters are adapted through parameter adjustment. This method reliably identifies stress concentration features, such as local energy accumulation caused by microcracks, while suppressing transient interference caused by temporary attachments.

[0171] The processing unit is time-stamped with the industrial camera group to ensure consistency between stress data and motion status. The sorting mechanism receives the cloud image analysis results and responds instantly at the discharge end. The entire process maintains production line continuity, significantly improving defect detection rate. The characteristics of the glass beads are also fully considered: the transparent surface does not interfere with stress sensing, and dynamic pressure changes during rolling serve as the detection basis. The system has low requirements for surface cleanliness; occasional stains do not affect the stress distribution pattern, and maintenance is simple and low-cost.

[0172] Figure 9 This is a detailed flowchart of step S322 provided by an embodiment of the present invention. In some embodiments, step S322, converting the glass bead region to the HSV color space, includes the following steps:

[0173] S3221. Based on the spherical geometric characteristics of the glass bead and the law of light refraction, the glass bead region is divided into a central region and an edge region, wherein the edge region includes the reflective interface where the glass bead contacts the detection track.

[0174] S3222. Perform HSV transformation on the central region and the edge region respectively to generate a central hue distribution histogram and an edge hue distribution histogram;

[0175] S3223. The central hue distribution histogram and the edge hue distribution histogram are integrated by weighted fusion, wherein the weight coefficient of the central region is higher than that of the edge region.

[0176] S3224. Output the fused hue distribution histogram as the pixel value distribution of the hue channel.

[0177] In the color detection stage, the glass bead area processing employs a zoned optimization method. A ring-shaped LED light source above the detection track provides stable illumination. After the industrial camera captures the image, the system first locates the glass bead outline. Based on spherical geometry, the glass bead area is divided into a central region and an edge region, centered on the outline's center point. The division dynamically adjusts the radius ratio based on the glass bead diameter, with the central region occupying the majority of the area, and the edge region being a ring-shaped area containing the contact interface between the glass bead and the detection track groove. This contact interface is prone to color distortion due to light refraction.

[0178] Color space conversion is performed on both regions separately. The central region undergoes direct HSV conversion, extracting the hue channel pixel values; the edge regions have a refractive compensation layer added before conversion. This refractive compensation layer calculates optical correction parameters based on the groove tilt angle and the glass material's refractive index, eliminating hue shift caused by refraction at the contact interface. After conversion, two independent hue distribution histograms are generated: the central histogram reflects the main hue distribution of the glass bead, while the edge histogram contains the corrected hue data for the contact area.

[0179] The system assigns different weight coefficients to the two histograms, giving higher weights to the central region and lower weights to the peripheral regions. The weight ratio is adjusted according to the diameter of the glass bead: the larger the diameter, the higher the weight of the central region. The fusion process is achieved by overlaying the histograms, taking a weighted average for each hue value interval to generate the final hue distribution histogram.

[0180] This solution significantly improves color detection accuracy. Refractive distortion at the edges of the glass beads is effectively suppressed, while the main hue in the central area is preserved. Common edge color shift problems in traditional methods are resolved; for example, optical distortion in the contact area no longer affects overall hue judgment. Different glass beads are matched using refractive index parameters to maintain detection consistency. The processing is performed synchronously with image acquisition, without increasing system latency.

[0181] When the detection track wears down due to long-term use, the slide angle parameter can be adjusted in a timely manner to maintain refraction compensation accuracy. Glass beads of different sizes are adapted by automatically adjusting the partition ratio. The transparent material characteristics of the glass beads are fully considered, and the partition weighting strategy effectively balances the main hue and edge optical characteristics, suppressing interference signals while ensuring true color reproduction.

[0182] Figure 10 This is a detailed flowchart of step S3221 provided by an embodiment of the present invention. In some embodiments, dividing the glass bead region into a central region and an edge region in step S3221 further includes the following steps:

[0183] S3221a. Determine the refraction angle of the contact point based on the chute structure parameters of the detection track; wherein, the chute structure parameters include the chute inclination angle and the chute surface roughness;

[0184] S3221b: Calculate the expected hue shift in the edge region based on the refractive index of the glass bead material;

[0185] S3221c: After generating the edge hue distribution histogram in S3222, apply the expected hue offset to compensate and correct the edge hue distribution histogram.

[0186] S3221d: Input the compensated and corrected edge hue distribution histogram into the weighted fusion of S3223.

[0187] During the processing of the glass bead edge region, the system optimizes hue analysis based on the physical structure of the detection track. The chute tilt angle parameter is acquired in real time by tilt sensors mounted on the chute sidewalls; these sensors output the angle between the chute plane and the horizontal plane. The chute surface roughness is measured by an optical profilometer pre-mounted on the support structure. This device emits a laser to scan the undulations of the chute surface, generating a roughness distribution map and calculating the mean value. The refractive index of the glass bead material is retrieved from the production database; common soda-lime glass uses a fixed refractive index, while special glass is matched to a parameter library based on the batch number.

[0188] As the glass bead rolls along the groove, the system calculates the actual refraction angle by combining the groove's tilt angle and surface roughness data. A simplified optical model is established based on the Fresnel reflection principle. This model takes the groove tilt angle, surface roughness, and glass refractive index as input and outputs the expected hue shift of light at the contact interface. The calculation process considers the diffuse reflection effect of the rough surface: higher roughness increases the scattering angle range, and the hue shift increases accordingly; a smaller tilt angle enhances the total internal reflection effect, leading to a decrease in hue saturation in the edge region.

[0189] After the edge hue distribution histogram is generated, the system performs compensation correction. For the count values ​​of each hue value interval in the histogram, a translation correction is performed according to the expected offset: if the expected offset is positive, the corresponding hue value interval is shifted towards higher hues; if it is negative, it is shifted towards lower hues. The translation distance is proportional to the absolute value of the offset, and the missing intervals after translation are filled by interpolation from adjacent intervals. The correction process preserves the original histogram shape characteristics, eliminating only the systematic offset caused by refraction.

[0190] The compensated and corrected edge hue distribution histogram is input into the weighted fusion module. The histogram for the central region retains the original data, while the edge region uses the corrected histogram. During fusion, the count values ​​of the same hue value interval are multiplied by the region weight coefficient (central region coefficient > edge region coefficient), and then weighted and summed to generate the final hue distribution. This scheme ensures that refractive distortion at the contact interface is effectively suppressed while preserving the true hue information of the glass bead body.

[0191] When wear of the slide groove causes changes in the tilt angle, the system adjusts the model parameters in real time based on sensor data. The refractive index database is updated in real time to adapt to the production of new glass materials. This ensures long-term detection accuracy.

[0192] This invention also provides a vision-based glass bead defect detection system, comprising:

[0193] The rolling module is used to separate the glass beads to be tested one by one into the testing track and control the glass beads to roll at a uniform speed on the testing track.

[0194] The acquisition module is used to continuously acquire images during the rolling of the glass bead using an industrial camera group, and obtain an image sequence containing multiple rotation angles of the glass bead. The industrial camera group includes multiple industrial cameras set at intervals along the detection track.

[0195] The detection module is used to perform the following detection steps based on the image sequence:

[0196] Calculate the centroid displacement vector of the glass bead in consecutive image frames, and determine whether the motion stability meets the stability criterion based on the change of the displacement vector;

[0197] Extract the color feature values ​​of the glass bead region in each frame of the image sequence, calculate the dispersion of the color feature values, and determine whether the dispersion meets the dispersion standard.

[0198] Track the brightness change curve of the bright reflective region in the image sequence, and determine whether the brightness meets the brightness standard based on the brightness change curve;

[0199] The rejection module is used to determine that the glass beads are defective and control the sorting mechanism to reject the defective glass beads at the discharge end when any detection step in the detection module fails to meet the corresponding standard.

[0200] The system provided in this embodiment of the invention has the same technical features as the method described above, and therefore can achieve the same technical effect, which will not be elaborated here.

[0201] The above description is merely a preferred embodiment and the technical principles employed in this application. This application is not limited to the specific embodiments described herein, and various obvious changes, readjustments, and substitutions that can be made by those skilled in the art will not depart from the scope of protection of this application. Therefore, although this application has been described in detail through the above embodiments, this application is not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of this application, the scope of which is determined by the scope of the claims.

Claims

1. A method for detecting defects in glass beads based on visual inspection, characterized in that, The method includes the following steps: S1. Separate the glass beads to be tested one by one into the testing track, and control the glass beads to roll at a constant speed on the testing track; S2. Continuous image acquisition is performed by an industrial camera group during the rolling process of the glass bead to obtain an image sequence containing multiple rotation angles of the glass bead, wherein the industrial camera group includes multiple industrial cameras arranged at intervals along the detection track. S3. Perform the following detection steps based on the image sequence: S31. Calculate the centroid displacement vector of the glass bead in consecutive image frames, and determine whether the motion stability meets the stability standard based on the change of the displacement vector; S32. Extract the color feature values ​​of the glass bead region in each frame of the image sequence, calculate the dispersion of the color feature values, and determine whether the dispersion meets the dispersion standard. S33. Track the brightness change curve of the bright reflection area in the image sequence, and determine whether the brightness meets the brightness standard based on the brightness change curve; S4. If any detection step in S3 fails to meet the corresponding standard, it is determined that the glass bead has a defect and the sorting mechanism is controlled to remove the defective glass bead at the discharge end.

2. The method according to claim 1, characterized in that, In step S31, determining whether the motion smoothness meets the smoothness criterion based on the change of the displacement vector includes: S311. A distance sensor array is installed on the side wall of the slide of the detection track to collect the distance data between the glass bead and the side wall of the slide in real time. S312. Generate the motion trajectory envelope by combining the centroid displacement vector and the spacing data; S313. When the envelope is detected to exhibit periodic fluctuations or continuously deviate from the central axis of the slide or the centroid position changes abruptly, it is determined that the motion stability does not meet the stability standard. The groove structure of the detection track restricts the free rolling path of the glass bead, and the contact area between the sidewall of the groove and the glass bead forms a motion constraint boundary.

3. The method according to claim 1, characterized in that, S33 specifically includes the following steps: S331. During the rolling process of the glass bead, identify the bright reflection area formed by the light source in each frame of the image; S332. Extract the maximum brightness of the bright reflection area in consecutive image frames and generate a brightness change time-series curve. S333. Calculate the first derivative of the time-series curve. When the absolute value of the derivative exceeds a preset abrupt change threshold, mark it as a point of sudden brightness drop. S334. Count the number of discontinuous change points in the time-series curve. The discontinuous change points include the brightness drop points and the points where the brightness difference between adjacent frames exceeds the step threshold. S335. When the number of discontinuous change points exceeds the fault tolerance limit, the brightness is determined to not meet the brightness standard.

4. The method according to claim 3, characterized in that, After identifying the bright reflection area in S331, the following steps are also included: S331a. Extract the contour geometric features of the high-brightness reflection area, including the contour curvature radius distribution and contour symmetry index; S331b. Based on the nominal refractive index of the glass beads, establish an optical path offset model and calculate the theoretical deformation of the reflection profile in a defect-free state. S331c. Compare the deviation of the contour curvature radius distribution from the theoretical deformation. When a local curvature abnormal change is detected and the deviation exceeds the optical tolerance threshold, it is marked as a potential bubble region. S331d. Combining the brightness change time-series curve with the deviation, when the deviation continuously exceeds the optical tolerance threshold and the brightness change time-series curve of the corresponding region exhibits low-frequency high-amplitude oscillation characteristics, it is determined that a bubble defect exists.

5. The method according to claim 1, characterized in that, S32 specifically includes the following steps: S321. For each frame of the image sequence, segment out the glass bead region; S322. Convert the glass bead region to the HSV color space and extract the pixel value distribution of the hue channel; S323. Calculate the Bartholomew distance of the hue distribution histogram between consecutive image frames, and calculate the mean Bartholomew distance of three consecutive frames. S324. Calculate the kurtosis coefficient of the hue distribution histogram in the entire image sequence; S325. When the mean value of the Barthel distance exceeds the distribution change threshold or the kurtosis coefficient is lower than the flatness threshold, it is determined that the dispersion does not meet the dispersion standard.

6. The method according to claim 1, characterized in that, The steps preceding S4 include: S51. A piezoelectric sensor array is installed on the support structure of the detection track to collect stress fluctuation signals generated when the glass beads roll in real time. S52. Construct a dynamic stress distribution cloud map based on the stress fluctuation signal; S53. Extract the spatial distribution pattern of stress concentration areas, and mark potential cracks when linearly extending stress concentration zones are detected. S54. Calculate the fluctuation intensity index of the stress concentration zone, wherein the fluctuation intensity index is determined by the ratio of the stress peak value to the background noise. S55. When the stress concentration zone satisfies the condition that the spatial distribution pattern coincides with the internal structural line of the glass bead to an extent exceeding the geometric matching threshold and / or the fluctuation intensity index exhibits a periodic enhancement characteristic with the rolling angle, it is determined that there is a microcrack defect. S4 further includes: When it is determined that the glass beads have microcrack defects, the sorting mechanism is controlled to remove the defective glass beads at the discharge end.

7. The method according to claim 6, characterized in that, S52 specifically includes the following steps: S521. Establish a three-dimensional coordinate system based on the spatial coordinates of the piezoelectric sensor array, wherein the Z-axis is perpendicular to the detection track plane and the X-axis is along the rolling direction of the glass beads; S522. Perform time-frequency decomposition on the stress fluctuation signal, extract the fundamental frequency component that is synchronized with the rotation period of the glass bead, generate a static stress distribution layer, and separate the resonant components higher than the fundamental frequency to generate a dynamic stress fluctuation layer. S523. Map the static stress distribution layer onto the lower hemisphere of the glass bead surface model; S524. Mark the abnormal resonance region in the dynamic stress fluctuation layer. When the resonance energy is concentrated in the local meridian direction and lasts for more than 3 rotation cycles, generate the initial mark of the stress concentration zone. S525. Merge the static stress distribution layer and the dynamic stress fluctuation layer, overlay the initial mark of the stress concentration zone, and output the dynamic stress distribution cloud map.

8. The method according to claim 5, characterized in that, In step S322, converting the glass bead region to the HSV color space includes the following steps: S3221. Based on the spherical geometric characteristics and light refraction law of the glass bead, the glass bead region is divided into a central region and an edge region, wherein the edge region includes the reflective interface where the glass bead contacts the detection track. S3222. Perform HSV conversion on the central region and the edge region respectively to generate a central hue distribution histogram and an edge hue distribution histogram; S3223. The central hue distribution histogram and the edge hue distribution histogram are integrated using a weighted fusion method, wherein the weight coefficient of the central region is higher than that of the edge region. S3224. Output the fused hue distribution histogram as the pixel value distribution of the hue channel.

9. The method according to claim 8, characterized in that, In step S3221, dividing the glass bead region into a central region and an edge region further includes the following steps: S3221a. Determine the refraction angle of the contact point based on the chute structure parameters of the detection track; wherein, the chute structure parameters include the chute inclination angle and the chute surface roughness; S3221b: Calculate the expected hue shift in the edge region based on the refractive index of the glass bead material; S3221c: After generating the edge hue distribution histogram in S3222, the expected hue offset is applied to compensate and correct the edge hue distribution histogram. S3221d, Input the compensated and corrected edge hue distribution histogram into the weighted fusion of S3223.

10. A glass bead defect detection system based on vision inspection, characterized in that, include: A rolling module is used to separate the glass beads to be tested one by one into the detection track and control the glass beads to roll at a uniform speed on the detection track. The acquisition module is used to continuously acquire images during the rolling process of the glass bead using an industrial camera group, and obtain an image sequence containing multiple rotation angles of the glass bead, wherein the industrial camera group includes multiple industrial cameras arranged at intervals along the detection track. The detection module is used to perform the following detection steps based on the image sequence: Calculate the centroid displacement vector of the glass bead in consecutive image frames, and determine whether the motion stability meets the stability criterion based on the change of the displacement vector; Extract the color feature values ​​of the glass bead region in each frame of the image sequence, calculate the dispersion of the color feature values, and determine whether the dispersion meets the dispersion standard; Track the brightness change curve of the bright reflection area in the image sequence, and determine whether the brightness meets the brightness standard based on the brightness change curve; The rejection module is used to determine that the glass bead has a defect and control the sorting mechanism to reject the defective glass bead at the discharge end when any detection step in the detection module fails to meet the corresponding standard.

Images