A glass quality defect detection method and system based on AI image recognition

Abstract

The present invention relates to the technical field of tempered glass image quality inspection, and in particular to a glass quality defect detection method and system based on AI image recognition. The method comprises the following steps: acquiring a glass image set from multiple angles; performing image contour background separation on the glass image set to generate a glass background separation image set; performing perspective stereoscopic splicing on the glass background separation image set to generate a glass perspective stereogram; performing glass high stress area segmentation on the glass perspective stereogram to generate a glass high stress area image; performing vibration detection on the glass perspective stereogram, and using a vibration sensor to perform vibration pattern image acquisition to obtain a glass vibration pattern image. The present invention improves the accuracy, comprehensiveness and automation level of quality defect detection through multi-angle image acquisition, comprehensive defect detection, intelligent identification marking and systematic management.

Description

Technical Field

[0001] The present invention relates to the technical field of tempered glass image quality inspection, and in particular to a glass quality defect detection method and system based on AI image recognition. Background Art

[0002] Traditional methods for detecting quality defects in tempered glass mainly rely on manual inspection and simple image processing techniques, which are not only time-consuming and labor-intensive, but also difficult to ensure accuracy and consistency. With the development of computer vision and deep learning technology, AI image recognition has gradually become an important means in the field of quality defect detection. Basic image processing algorithms such as edge detection and texture analysis have begun to be applied to industrial inspection, but their application scope is limited due to the complexity of the algorithm and computing power.

[0003] Entering the 21st century, the improvement of computing power and the introduction of machine learning, especially the application of algorithms such as support vector machines and decision trees, have enabled image recognition to gradually emerge in defect detection; however, the current traditional methods may detect multiple defects in tempered glass at the same time, such as internal stress, surface scratches, bubbles, etc., and often only detect a certain type of defect in tempered glass, and are unable to comprehensively detect various types of defects. At the same time, there is a lack of intelligent analysis and systematic management of the test results, making it difficult to achieve real-time feedback and historical data management of quality inspections, which in turn leads to low accuracy, comprehensiveness and automation levels in quality defect detection. Summary of the invention

[0004] Based on this, it is necessary to provide a glass quality defect detection method and system based on AI image recognition to solve at least one of the above technical problems.

[0005] To achieve the above object, the present invention discloses a glass quality defect detection method based on AI image recognition, the method comprising the following steps:

[0006] Step S1: Acquire a glass image set from multiple angles; perform image contour and background separation on the glass image set to generate a glass background separation image set; perform perspective stereoscopic stitching on the glass background separation image set to generate a glass perspective stereogram;

[0007] Step S2: segmenting the glass high stress area of ​​the glass viewing angle stereogram to generate a glass high stress area image; performing vibration detection on the glass viewing angle stereogram, and using a vibration sensor to collect vibration pattern images to obtain a glass vibration pattern image; performing abnormal vibration pattern defect detection on the glass high stress area image through the glass vibration pattern image and the glass high stress area image, thereby generating a glass first abnormal area image;

[0008] Step S3: segmenting the glass perspective stereogram into intact areas through the first abnormal area image of the glass to obtain an intact area segmentation image; performing hot melt detection on the glass according to the intact area segmentation image, and collecting thermal images using an infrared camera to obtain a thermal map of the intact area of ​​the glass; performing abnormal deformation defect detection on the thermal map of the intact area of ​​the glass to generate an image of the second abnormal area of ​​the glass;

[0009] Step S4: optically transmit and reflect the glass according to the intact area segmentation image, and collect optical transmission and reflection images through a camera to obtain a glass surface reflection image; perform peak intensity defect analysis on the glass surface reflection image to generate optical defect analysis data; create a mask based on the optical defect analysis data to obtain an abnormal optical defect mask; use the abnormal optical defect mask to perform optical abnormal area edge segmentation on the intact area segmentation image, thereby generating a third abnormal area image of the glass;

[0010] Step S5: Mark the abnormal image areas of the first abnormal area image of the glass, the second abnormal area image of the glass, and the third abnormal area image of the glass to generate a defect area marked image; construct an intelligent identifier for the defect area marked image to obtain a defect intelligent identifier; perform intelligent quality defect detection on the glass perspective stereo image according to the defect intelligent identifier to obtain a quality defect inspection map; upload the quality defect inspection map to the MES system for image storage, and generate a glass quality defect detection storage map to execute quality control management operations for the entire glass production process.

[0011] Preferably, step S1 comprises the following steps:

[0012] Step S11: using a camera to obtain a glass image set from multiple angles;

[0013] Step S12: performing image preprocessing on the glass image set to generate a standard glass image set, wherein the image preprocessing includes image filtering, image brightness enhancement and image grayscale conversion;

[0014] Step S13: extracting glass contour features from the standard glass image set to obtain glass contour feature data; performing image background separation on the standard glass image set using the glass contour feature data to generate a glass background separation image set;

[0015] Step S14: Perform perspective stereoscopic stitching on the glass-background separated image set to generate a glass perspective stereoscopic image.

[0016] The present invention uses a camera to shoot glass from multiple angles to obtain a multi-angle glass image set. These images provide rich viewing angle information for subsequent processing and can more comprehensively reflect the characteristics of the glass. The acquired glass image set is preprocessed, including the following sub-steps: image filtering: remove noise in the image to improve image quality; image brightness enhancement: adjust the brightness of the image to make the image details clearer; image grayscale conversion: convert the color image into a grayscale image to facilitate subsequent feature extraction and background separation. After these preprocessing steps, a standard glass image set is generated to ensure image quality and consistency. The contour features of the glass are extracted from the standard glass image set to obtain glass contour feature data.

[0017] Using these feature data, the image background of the standard glass image set is separated to generate a glass background separation image set, which can effectively separate the glass from the background and highlight the characteristics of the glass. The glass background separation image set is subjected to perspective stereo stitching to generate a glass perspective stereogram. By stitching multi-angle images, a glass image with a three-dimensional sense is generated, which more comprehensively and realistically reflects the three-dimensional structure of the glass. The implementation of the above steps can effectively improve the processing quality and accuracy of the glass image. The images acquired from multiple angles provide rich information, image preprocessing improves the image quality and consistency, contour feature extraction and background separation highlight the characteristics of the glass, and the stereo image generated by perspective stereo stitching can better display the three-dimensional structure of the glass. The combination of these processing steps makes the final generated glass perspective stereogram have higher quality and accuracy, which can be applied to glass detection, analysis and display and other fields.

[0018] Preferably, step S14 comprises the following steps:

[0019] Step S141: calculating the shooting perspective angle of the glass based on the glass-background separation image set to obtain shooting perspective angle data; grouping the glass-background separation image set according to the shooting perspective angle data to generate perspective group images, wherein the perspective image groups include a front view group, a rear view group, a left view group, a right view group, a top view group, and a bottom view group;

[0020] Step S142: performing group image overlap on the view group images to generate a view group overlap image; performing feature point detection on the view group overlap image to generate feature points of the view group overlap image; using the feature points of the view group overlap image to perform repeated area recognition on the view group overlap image to obtain a group overlap area image;

[0021] Step S143: performing non-overlapping area image registration on the view angle grouping overlapping area image by grouping overlapping area images to obtain a glass view angle grouping registered image;

[0022] Step S144: using the Cartesian space coordinate system to perform spatial perspective stereoscopic stitching on the glass perspective grouping registration images to generate a glass perspective stereoscopic image.

[0023] The present invention accurately calculates the shooting angle of each image, so that subsequent image grouping and processing can be performed based on actual shooting conditions, which helps to reduce image stitching errors caused by differences in viewing angles. Images are sorted and organized according to viewing angles, which facilitates subsequent processing and analysis. Grouping images from different viewing angles can effectively organize image data and ensure that data from each viewing angle is reasonably processed during stitching and overlap.

[0024] Through image coincidence processing, images from different perspectives can be aligned to generate continuous perspective group coincidence images, which helps to perform more accurate feature point detection and repeated area identification in subsequent steps. The detected feature points can be used to identify key areas in the image, thereby helping to determine repeated areas and perform image registration.

[0025] High-precision feature point detection improves the accuracy of image overlap. By identifying repeated areas, the overlapping parts in the perspective grouping overlapped images can be effectively found, providing the necessary information for the registration of non-overlapping areas and reducing errors in image processing. Registration of non-overlapping areas ensures that images at different perspectives can be accurately aligned during image stitching. This process reduces the difference between perspectives and optimizes the quality and consistency of the final image. Stereo stitching of images using the Cartesian space coordinate system can generate high-precision three-dimensional perspective images. This stereo stitching provides a complete three-dimensional perspective and enhances the understanding and analysis of the glass surface structure.

[0026] Preferably, step S2 comprises the following steps:

[0027] Step S21: performing glass region stress analysis on the glass viewing angle stereogram to generate glass region stress data; performing high stress region segmentation on the glass viewing angle stereogram according to the glass region stress data to generate a glass high stress region image;

[0028] Step S22: performing vibration detection on the glass based on the image of the high stress area of ​​the glass, and collecting vibration wrinkle images using a vibration sensor to obtain a glass vibration wrinkle map;

[0029] Step S23: Correlate the glass vibration pattern with the glass high stress region image to generate the glass high stress region vibration pattern;

[0030] Step S24: performing abnormal chattering defect detection on the glass high stress area image through the glass high stress area chattering pattern, thereby generating a glass first abnormal area image.

[0031] The present invention can accurately identify the stress distribution on the glass surface or structure by performing stress analysis on the glass viewing angle stereogram. This analysis helps determine potential high stress areas and provides basic data for subsequent high stress area segmentation.

[0032] By segmenting the high stress areas of the glass perspective stereogram through stress data, it is possible to clearly show the areas with greater stress in the glass, which helps to focus attention on the areas with problems and facilitates subsequent vibration detection and abnormality detection. The vibration pattern images obtained by the vibration detection sensor can record the response characteristics of the glass under vibration conditions. This data is very important for understanding the vibration characteristics and possible failure modes of the glass in actual use.

[0033] The present invention associates the glass vibration pattern with the image of the high stress area, which can combine the vibration response information with the stress distribution information, help identify abnormal vibration modes in the high stress area, and further analyze potential defects or damaged areas. By analyzing the vibration pattern image of the high stress area of ​​the glass, abnormal vibration pattern defects can be detected. This process helps locate and identify structural defects or problem areas in the glass, providing key information for maintenance and repair.

[0034] Preferably, step S24 includes the following steps:

[0035] Step S241: performing regional vibration wave amplitude fluctuation analysis on the vibration wave pattern of the glass high stress region to generate a regional vibration wave amplitude fluctuation curve;

[0036] Step S242: calculating the amplitude fluctuation range of the regional chattering amplitude fluctuation curve to obtain amplitude fluctuation range data;

[0037] Step S243: comparing the amplitude fluctuation data with a preset standard amplitude fluctuation threshold, and when the amplitude fluctuation data is greater than the preset standard amplitude fluctuation threshold, confirming the abnormal crack area of ​​the vibration pattern of the glass high stress area based on the amplitude fluctuation data to obtain abnormal crack area data;

[0038] Step S244: segmenting the glass high stress region image into an abnormal crack region image according to the abnormal crack region data to generate a first abnormal region image of the glass.

[0039] The present invention can reveal small changes in vibration patterns by analyzing the amplitude fluctuations of vibration patterns in high stress areas of glass. These fluctuation curves help identify abnormal features in vibration responses and provide data support for subsequent defect detection. The amplitude of the regional vibration fluctuations is calculated so that the intensity of the amplitude fluctuations can be quantified.

[0040] Accurate amplitude fluctuation data helps evaluate the vibration stability of the glass area and provides a quantitative basis for abnormality detection. Comparing the calculated amplitude fluctuation data with the preset standard amplitude fluctuation threshold can effectively screen out vibration fluctuations that exceed the normal range. When the amplitude data exceeds the threshold, the abnormal crack area can be confirmed, providing a basis for accurately locating the crack. By comparing the abnormal crack area data obtained, the crack area in the glass can be accurately identified. This confirmation helps locate structural defects and make timely repairs or further analysis.

[0041] Furthermore, the image of the high stress area of ​​the glass is segmented according to the data of the abnormal crack area to generate the first abnormal area image. This process helps to clearly identify the area where cracks exist in the glass, provides clear visual information, and facilitates subsequent defect handling and maintenance decisions.

[0042] Preferably, step S3 comprises the following steps:

[0043] Step S31: segmenting the glass viewing angle stereogram into intact regions using the first abnormal region image of the glass to obtain an intact region segmentation image;

[0044] Step S32: performing hot melt detection on the glass according to the intact area segmentation image, and collecting a thermal map using an infrared camera to obtain a thermal map of the intact area of ​​the glass; extracting peak temperature information from the thermal map of the intact area of ​​the glass to obtain peak temperature information data;

[0045] Step S33: extracting the high heat-melting area from the intact area segmentation image based on the peak temperature information data to obtain a high heat-melting area image;

[0046] Step S34: performing abnormal deformation defect detection on the high-thermal melting area image to generate a second abnormal area image of the glass.

[0047] The present invention can clearly distinguish the area without defects from the overall image by segmenting the image of the first abnormal area of ​​the glass into intact areas, which helps to exclude known abnormal areas, concentrate resources on further analysis and detection of normal areas, and improve detection efficiency.

[0048] Using an infrared camera to collect thermal images of intact areas can detect the thermal melt properties of glass under heating conditions. This process can reveal the thermal response characteristics of glass and provide basic data for subsequent extraction of thermal melt areas. Extracting peak temperature information from the thermal map can accurately obtain the highest temperature of intact glass areas. This data helps identify abnormal hot spots in the glass during the thermal melt process and provides a basis for the detection of high thermal melt areas.

[0049] The present invention extracts high-melt regions from intact region segmented images based on peak temperature information data, which can identify regions with abnormally high temperatures, help find existing melt defect regions, and provide key information for further defect analysis. Abnormal deformation defect detection on high-melt region images can identify deformations or defects that occur under high temperature conditions. This process helps locate potential problem areas in glass and provides accurate data support for glass quality assessment and maintenance.

[0050] Preferably, step S34 includes the following steps:

[0051] Step S341: performing isothermal time-series image analysis on the high thermal melting area image to generate a high thermal melting area time-series image set;

[0052] Step S342: performing image pixel position calculation on the high thermal melt region time series image set to obtain an image pixel position data set; performing adjacent time series image pixel offset calculation on the image pixel position data set to obtain a high thermal melt region pixel offset map;

[0053] Step S343: generating a displacement vector field for the pixel offset map of the high thermal melting area by using an optical flow method to obtain a pixel displacement vector field;

[0054] Step S344: Analyze the displacement amplitude of the high thermal melt area image through the pixel displacement vector field to generate abnormal deformation detection data; perform image edge segmentation on the high thermal melt area image based on the abnormal deformation detection data to generate a second abnormal area image of the glass.

[0055] The present invention generates a high-melt region time-series image set by performing isothermal time-series image analysis on the high-melt region image, which can capture the dynamic changes in the thermal melting process, help observe the performance of the thermal melting region at different time points, identify heat-related time-series features, and thus better understand the impact of thermal melting. The pixel position of the high-melt region time-series image set is calculated to obtain an image pixel position data set, which provides basic data for subsequent offset calculations. This process ensures the accuracy and reliability of subsequent image processing.

[0056] The present invention generates a pixel offset map of the high heat-melting area by calculating the pixel offset of adjacent time-series images. This step can reveal the displacement information of pixels during the heat-melting process and help identify image deformation or anomalies caused by thermal stress. The optical flow method is used to generate a displacement vector field for the pixel offset map, which can accurately calculate the motion vector of each pixel. This method can describe the displacement pattern of the heat-melting area in detail, providing key data for further analysis of abnormal deformation.

[0057] The present invention can detect abnormal deformation in the hot melt area by performing displacement amplitude analysis through the pixel displacement vector field. This analysis helps identify deformation problems caused by thermal stress, thereby accurately locating potential defect areas. Image edge segmentation of the high hot melt area image based on abnormal deformation detection data can clearly identify abnormal edges in the hot melt area, help generate an accurate second abnormal area image, and provide specific area information for subsequent defect processing and repair.

[0058] Preferably, step S4 comprises the following steps:

[0059] Step S41: optically transmit and reflect the glass according to the intact area segmentation image, and collect optical transmission and reflection images through a camera to obtain a glass surface reflection image;

[0060] Step S42: performing fast Fourier transform on the glass surface reflection image to obtain a surface reflection spectrum diagram; performing amplitude spectrum and phase spectrum calculation on the surface reflection spectrum diagram to obtain reflection amplitude data and reflection phase data;

[0061] Step S43: performing peak intensity defect analysis according to the reflection amplitude data and the reflection phase data to generate optical defect analysis data; creating a mask based on the optical defect analysis data to obtain an abnormal optical defect mask;

[0062] Step S44: performing inverse Fourier transform on the abnormal optical defect mask to obtain the abnormal optical defect area; mapping the abnormal optical defect area to the intact area segmentation image to perform optical abnormal area edge segmentation, thereby generating a third abnormal area image of the glass.

[0063] The present invention obtains the reflected image of the glass surface through optical transmission reflection and camera acquisition, and can obtain detailed information of the reflected light on the glass surface, which provides a high-quality basic image for subsequent optical analysis and helps to identify surface defects and irregularities. Fast Fourier transform of the reflected image of the glass surface can convert the image from the spatial domain to the frequency domain, thereby revealing the frequency components in the image. This conversion helps to identify potential periodic or non-periodic defect patterns.

[0064] The present invention can extract the reflection intensity and phase information in the image by calculating the amplitude spectrum and phase spectrum of the spectrum graph. These data are helpful for analyzing the reflection characteristics and identifying optical defects, and provide detailed frequency domain information for subsequent defect analysis. Peak intensity defect analysis based on reflection amplitude data and phase data can identify intensity abnormalities in the reflection image. These peak intensities usually indicate potential optical defect areas, such as scratches, bubbles, etc. Creating a mask based on the optical defect analysis data can separate the identified abnormal optical defect area from the background. This process helps to highlight the defect area and provide a specific target area for subsequent processing and repair.

[0065] The present invention can convert the defect information in the frequency domain back to the spatial domain by performing an inverse Fourier transform on the abnormal optical defect mask, and obtain the actual abnormal optical defect area. This process enables the abnormal area to be clearly displayed and analyzed in the actual image. Mapping the abnormal optical defect area to the intact area segmentation image for edge segmentation can accurately locate the boundary of the optical abnormal area, which helps to generate a detailed third abnormal area image, facilitating further detection and repair work.

[0066] Preferably, step S5 comprises the following steps:

[0067] Step S51: integrating the first abnormal area image of the glass, the second abnormal area image of the glass and the third abnormal area image of the glass to generate an abnormal area image; marking defect areas of the glass viewing angle stereogram based on the abnormal area image to obtain a defect area marking image;

[0068] Step S52: dividing the defect area marked image into a data set to obtain a model training set and a model test set; using a convolutional neural network algorithm to perform model training on the model training set to generate an anomaly recognition trainer; using the model test set to optimize and iterate the anomaly recognition trainer to generate an intelligent defect identifier; performing intelligent quality defect detection on the glass perspective stereogram according to the intelligent defect identifier to obtain a quality defect inspection map;

[0069] Step S53: Upload the quality defect inspection image to the MES system for image storage, generate a glass quality defect detection storage image, and perform quality control management operations throughout the glass production process.

[0070] The present invention integrates the first, second, and third abnormal area images of the glass to generate a comprehensive abnormal area image. This process summarizes the abnormal area information obtained by different detection methods, provides a comprehensive view of the overall defects of the glass, and helps to fully understand the defect distribution on the glass surface. Based on the abnormal area image, the defect area is marked on the glass perspective stereogram, and the defect area can be clearly indicated in the stereogram. This marked image provides a clear target area for subsequent quality inspection and repair work, helping to improve the accuracy and efficiency of defect detection.

[0071] The present invention provides basic data for the training and evaluation of machine learning models by dividing defect area marked images into model training sets and model test sets. This division ensures the independence of model training and testing and improves the reliability of model evaluation. The training set is trained using a convolutional neural network algorithm to generate an anomaly recognition trainer specifically for glass defect recognition. The high efficiency of CNN in image recognition tasks helps to improve the accuracy and robustness of defect detection. The anomaly recognition trainer is optimized and iterated using the model test set to generate an intelligent defect identifier. This process can improve the performance of the identifier, reduce false positives and false negatives, and enhance the intelligent detection capability of glass defects.

[0072] The present invention generates a quality defect inspection map by intelligently detecting quality defects on the glass viewing angle stereogram according to the defect intelligent identifier. The inspection map can accurately identify and mark the quality defects in the glass, providing reliable data support for quality control. The quality defect inspection map is uploaded to the MES system for image storage, and a glass quality defect detection storage map is generated. This process can realize the quality control management operation of the entire production process, ensure the recording and tracking of quality inspection data, help improve the level of production quality control, and support subsequent quality tracing and improvement measures.

[0073] Furthermore, the present invention also discloses a glass quality defect detection system based on AI image recognition, which is used to execute the above-mentioned glass quality defect detection method based on AI image recognition. The glass quality defect detection system based on AI image recognition includes:

[0074] A stereoscopic stitching module is used to obtain a glass image set from multiple angles; perform image contour and background separation on the glass image set to generate a glass background separation image set; perform perspective stereoscopic stitching on the glass background separation image set to generate a glass perspective stereogram;

[0075] The vibration detection module is used to segment the glass high stress area of ​​the glass perspective stereogram to generate a glass high stress area image; perform vibration detection on the glass perspective stereogram, and use a vibration sensor to collect vibration pattern images to obtain a glass vibration pattern image; perform abnormal vibration pattern defect detection on the glass high stress area image through the glass vibration pattern image and the glass high stress area image, thereby generating a glass first abnormal area image;

[0076] The hot melt detection module is used to segment the glass perspective stereogram into intact areas through the first abnormal area image of the glass to obtain the segmented image of the intact area; perform hot melt detection on the glass according to the segmented image of the intact area, and use an infrared camera to collect the thermal image to obtain the thermal image of the intact area of ​​the glass; perform abnormal deformation defect detection on the thermal image of the intact area of ​​the glass to generate the second abnormal area image of the glass;

[0077] The optical inspection module is used to perform optical transmission and reflection on the glass according to the intact area segmentation image, and collect optical transmission and reflection images through a camera to obtain a glass surface reflection image; perform peak intensity defect analysis on the glass surface reflection image to generate optical defect analysis data; create a mask based on the optical defect analysis data to obtain an abnormal optical defect mask; use the abnormal optical defect mask to perform optical abnormal area edge segmentation on the intact area segmentation image, thereby generating a glass third abnormal area image;

[0078] The intelligent recognition module is used to mark the abnormal image areas of the first abnormal area image of the glass, the second abnormal area image of the glass and the third abnormal area image of the glass to generate a defect area marked image; construct an intelligent identifier for the defect area marked image to obtain a defect intelligent identifier; perform intelligent quality defect detection on the glass perspective stereo image according to the defect intelligent identifier to obtain a quality defect inspection map; upload the quality defect inspection map to the MES system for image storage to generate a glass quality defect detection storage map to execute quality control management operations for the entire glass production process.

[0079] The beneficial effects of the present invention are:

[0080] The present invention is to obtain images from multiple angles to fully understand the surface conditions of the glass. The generated perspective stereogram can provide a comprehensive perspective, which is convenient for subsequent defect detection and analysis. The segmentation of high stress areas helps to locate the weak points of the glass. Through vibration detection and vibration pattern image acquisition, potential defects caused by stress can be found. Abnormal vibration pattern detection helps to identify potential risks of glass during use. The segmentation of flawless areas can help to clarify which parts of the glass surface are flawless. Hot melt detection can check the stability of glass under high temperature conditions.

[0081] At the same time, the present invention can further ensure the durability and stability of glass in practical applications through the collection of thermal maps and abnormal deformation detection. Optical transmission and reflection image acquisition can detect abnormal reflections on the glass surface. Peak intensity defect analysis and optical abnormal area edge segmentation help to find tiny optical defects that affect the visual effects and optical properties of the glass. Abnormal area marking can accurately record various defect types. The intelligent identifier can automatically identify defective areas and improve detection efficiency. The generation and storage of quality defect inspection maps can ensure quality control in the production process and support subsequent quality tracing and improvement. Therefore, the present invention improves the accuracy, comprehensiveness and automation level of quality defect detection through multi-angle image acquisition, comprehensive defect detection, intelligent identification marking and systematic management. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] Figure 1 A schematic diagram of the steps of a glass quality defect detection method based on AI image recognition;

[0083] Figure 2 for Figure 1 Detailed implementation steps of step S2 in the flowchart;

[0084] Figure 3 for Figure 1 Detailed implementation steps of step S3 in FIG.

[0085] Figure 4 for Figure 1 Detailed implementation steps of step S4 in FIG. DETAILED DESCRIPTION

[0086] The technical method of the present invention is clearly and completely described below in conjunction with the accompanying drawings.

[0087] Example 1, please refer to Figures 1 to 4 , a glass quality defect detection method based on AI image recognition, the method comprising the following steps:

[0088] Step S1: Acquire a glass image set from multiple angles; perform image contour and background separation on the glass image set to generate a glass background separation image set; perform perspective stereoscopic stitching on the glass background separation image set to generate a glass perspective stereogram;

[0089] Step S2: segmenting the glass high stress area of ​​the glass viewing angle stereogram to generate a glass high stress area image; performing vibration detection on the glass viewing angle stereogram, and using a vibration sensor to collect vibration pattern images to obtain a glass vibration pattern image; performing abnormal vibration pattern defect detection on the glass high stress area image through the glass vibration pattern image and the glass high stress area image, thereby generating a glass first abnormal area image;

[0090] Step S3: segmenting the glass perspective stereogram into intact areas through the first abnormal area image of the glass to obtain an intact area segmentation image; performing hot melt detection on the glass according to the intact area segmentation image, and collecting thermal images using an infrared camera to obtain a thermal map of the intact area of ​​the glass; performing abnormal deformation defect detection on the thermal map of the intact area of ​​the glass to generate an image of the second abnormal area of ​​the glass;

[0091] Step S4: optically transmit and reflect the glass according to the intact area segmentation image, and collect optical transmission and reflection images through a camera to obtain a glass surface reflection image; perform peak intensity defect analysis on the glass surface reflection image to generate optical defect analysis data; create a mask based on the optical defect analysis data to obtain an abnormal optical defect mask; use the abnormal optical defect mask to perform optical abnormal area edge segmentation on the intact area segmentation image, thereby generating a third abnormal area image of the glass;

[0092] Step S5: Mark the abnormal image areas of the first abnormal area image of the glass, the second abnormal area image of the glass, and the third abnormal area image of the glass to generate a defect area marked image; construct an intelligent identifier for the defect area marked image to obtain a defect intelligent identifier; perform intelligent quality defect detection on the glass perspective stereo image according to the defect intelligent identifier to obtain a quality defect inspection map; upload the quality defect inspection map to the MES system for image storage, and generate a glass quality defect detection storage map to execute quality control management operations for the entire glass production process.

[0093] As described above, the present invention comprehensively covers various areas of the glass surface through multi-angle image acquisition, thereby improving the coverage and accuracy of defect detection.

[0094] This step improves the contrast and clarity of the image by separating the image contour from the background, which helps with subsequent defect detection. The stereo image generated by stereo stitching provides three-dimensional information about the glass and enhances the ability to identify defects with complex shapes and locations.

[0095] This step accurately locates the areas on the glass that are prone to cracks or breakage by segmenting the high stress areas, which enhances the pertinence of defect detection. Combined with the vibration detection method of the vibration sensor, it can identify the slight vibration anomaly of the glass, improving the detection accuracy of slight cracks and defects.

[0096] This step generates the first abnormal area image by comprehensively analyzing the high stress area and the vibration pattern image, effectively identifying and marking the preliminary defect area of ​​the glass. The intact area segmentation effectively distinguishes the intact part of the glass and provides a clean reference image for further defect detection.

[0097] This step combines the hot melt test with the thermal image acquisition of an infrared camera to identify the thermal stress and deformation defects inside the glass, especially the internal defects that are difficult to detect with the naked eye. Abnormal deformation detection is performed through thermal images to generate a second abnormal area image, which improves the ability to detect thermal stress and deformation defects in the glass. Optical transmission and reflection image acquisition fully captures the optical properties of the glass surface, making it easier to identify optical defects.

[0098] This step can accurately detect surface defects and optical anomalies by analyzing the optical feature data through peak intensity defect analysis. Mask creation and optical anomaly area edge segmentation accurately locate and mark the optical defect area, and generate a third abnormal area image, which is helpful for subsequent defect analysis and processing.

[0099] This step integrates various defect information by marking abnormal image areas, generating a unified defect area marking image for comprehensive analysis. Build an intelligent identifier to improve the automation level and accuracy of defect detection, reduce manual intervention and misjudgment. Through intelligent detection of quality defects, a comprehensive quality inspection map is obtained to achieve a comprehensive assessment of glass quality.

[0100] This step uploads the quality inspection image to the MES system for systematic storage and management, thus achieving quality control management of the entire production process and improving production quality and efficiency. Therefore, the present invention improves the accuracy, comprehensiveness and automation level of quality defect detection through multi-angle image acquisition, comprehensive defect detection, intelligent identification marking and systematic management.

[0101] Figure 1 : is a schematic diagram of the steps of a glass quality defect detection method based on AI image recognition in this embodiment. In this example, the glass quality defect detection method based on AI image recognition includes the following steps:

[0102] Step S1: Acquire a glass image set from multiple angles; perform image contour and background separation on the glass image set to generate a glass background separation image set; perform perspective stereoscopic stitching on the glass background separation image set to generate a glass perspective stereogram;

[0103] In this step, select multiple high-resolution cameras to ensure that the cameras can capture images of the glass from different angles. Ensure that the position and angle of the cameras are fixed to avoid image stitching errors caused by position changes. Adjust the camera's exposure, focus, white balance and other parameters to ensure that the captured images are clear and have consistent colors.

[0104] In this step, you can use automated software to synchronously control the shooting parameters of multiple cameras. Take pictures of the glass from different angles, making sure that there is overlap at each angle for subsequent image stitching. You can use a turntable or other automated equipment to fix and rotate the glass to obtain images from multiple angles.

[0105] This step improves the image quality by preprocessing the acquired image, including noise removal, contrast adjustment, etc. Image enhancement techniques, such as Gaussian filtering and median filtering, are used to remove impurities and noise in the image. Image segmentation algorithms, such as threshold segmentation, edge detection, GrabCut, etc., are used to separate the outline of the glass.

[0106] In this step, you can use deep learning methods, such as U-Net, to train a specific model to automatically separate the glass background. Apply the selected segmentation algorithm to process each image and generate the corresponding glass background separation image. Make sure that the separated image only retains the outline and structure of the glass and removes the background.

[0107] This step uses feature point detection algorithms such as SIFT, SURF, and ORB to extract key feature points of each image. Use methods such as RANSAC to match feature points and filter out mismatched feature points. Based on the matched feature points, use affine transformation or perspective transformation to align images at different angles. Bundle Adjustment can be used to optimize image registration results and reduce cumulative errors.

[0108] This step stitches the aligned images and uses weighted average, Poisson fusion and other methods to process the transition area of ​​the image and reduce the seam marks. Ensure that the final generated glass perspective stereo image has no obvious stitching marks and the visual effect is natural. Further optimize the stitching results through deep learning methods to enhance the stereo effect. Generate a depth map and render the stereo image to enhance the three-dimensional perception of the glass perspective stereo image.

[0109] Step S2: segmenting the glass high stress area of ​​the glass viewing angle stereogram to generate a glass high stress area image; performing vibration detection on the glass viewing angle stereogram, and using a vibration sensor to collect vibration pattern images to obtain a glass vibration pattern image; performing abnormal vibration pattern defect detection on the glass high stress area image through the glass vibration pattern image and the glass high stress area image, thereby generating a glass first abnormal area image;

[0110] In the embodiment of the present invention, stress distribution simulation is performed by using finite element analysis software (such as ANSYS, Abaqus). The material properties and boundary conditions of the glass are input into the model, the loading conditions are simulated, and a stress distribution map is obtained. The high stress area is extracted from the stress distribution map, and a corresponding image is generated. The obtained stress image is preprocessed, such as graying, denoising, edge detection, etc.

[0111] This step further refines the contours of high stress areas by using morphological processing methods (such as corrosion and expansion). Generate images of high stress areas of the glass and mark the high stress areas. Install vibration sensors on the glass to collect vibration data of the glass at different locations. Apply vibrations of known frequency and amplitude through the control system and record the response of the glass. Use signal processing techniques (such as fast Fourier transform, FFT) to analyze the vibration data and identify vibration patterns.

[0112] In this step, the vibration sensor data is combined to capture the surface vibration image of the glass using a high-speed camera during the vibration process. The collected vibration images are aligned and processed through image processing technology to generate a clear glass vibration image. Feature extraction is performed on the glass high stress area image and vibration image. Vibration features and stress features are extracted using image processing technology (such as Canny edge detection, Hough transform, etc.). Compare and analyze the glass high stress area image and vibration image. Use machine learning or deep learning methods (such as convolutional neural network, CNN) to train the model to identify abnormal vibration defects.

[0113] In this step, the model automatically detects and marks abnormal areas by inputting feature extraction results. The detected abnormal vibration defects are marked on the high stress area image. The first abnormal area image of the glass is generated, which contains all the detected abnormal vibration defects.

[0114] Step S3: segmenting the glass perspective stereogram into intact areas through the first abnormal area image of the glass to obtain an intact area segmentation image; performing hot melt detection on the glass according to the intact area segmentation image, and collecting thermal images using an infrared camera to obtain a thermal map of the intact area of ​​the glass; performing abnormal deformation defect detection on the thermal map of the intact area of ​​the glass to generate an image of the second abnormal area of ​​the glass;

[0115] In the embodiment of the present invention, the consistency and contrast of the image are ensured by standardizing the glass perspective stereogram. The image of the first abnormal area of ​​the glass is used as a reference to locate the abnormal area. The glass perspective stereogram is segmented using an image segmentation algorithm (such as GrabCut, region growing, K-means clustering) to exclude the marked abnormal area.

[0116] In this step, the intact area is extracted from the original image based on the abnormal area image to generate the intact area segmentation image. The segmented intact area is subjected to morphological processing (such as dilation and erosion) to further optimize the segmentation result. The edge detection algorithm (such as Canny edge detection) is used to check the edge of the segmented area to ensure the accuracy of the segmentation.

[0117] In this step, an infrared camera is installed in the experimental environment and the equipment is debugged to ensure the accuracy of data collection. The glass sample is placed on a controllable heating device to prepare for the hot melt test. The glass is gradually heated and the hot melt behavior of the glass at different temperatures is recorded. The infrared camera is used to collect the thermal image data of the glass in real time during the heating process. The collected thermal images are preprocessed, such as denoising and contrast enhancement.

[0118] This step ensures the accuracy of temperature information by calibrating the temperature of each frame of thermal image. Extract the intact area from the preprocessed thermal image to generate a thermal image of the intact area of ​​the glass. Use image processing technology (such as edge detection and texture analysis) to extract features from the thermal image.

[0119] This step uses machine learning algorithms (such as PCA and LDA) to reduce and select features. Use deep learning methods (such as convolutional neural networks, CNN) to train the model and detect abnormal deformations in the thermal map. Input the extracted features into the trained model to identify and annotate the abnormal areas in the thermal map. Annotate the detected abnormal deformation defects on the intact area thermal map. Generate an image of the second abnormal area of ​​the glass and annotate all detected abnormal deformation defects.

[0120] Step S4: optically transmit and reflect the glass according to the intact area segmentation image, and collect optical transmission and reflection images through a camera to obtain a glass surface reflection image; perform peak intensity defect analysis on the glass surface reflection image to generate optical defect analysis data; create a mask based on the optical defect analysis data to obtain an abnormal optical defect mask; use the abnormal optical defect mask to perform optical abnormal area edge segmentation on the intact area segmentation image, thereby generating a third abnormal area image of the glass;

[0121] In the embodiment of the present invention, optical equipment including a light source (such as a laser or LED light) and a high-resolution camera are prepared, and a glass sample is placed in an optical transmission and reflection experimental device to ensure that the positions of the light source and the camera are fixed to avoid position deviation during the experiment.

[0122] In this step, the angle and intensity of the light source are adjusted to allow light to pass through the glass and reflect on its surface. Use a camera to simultaneously capture optical transmission images and reflection images. Perform optical transmission and reflection tests on multiple positions and angles of the glass to ensure the comprehensiveness of the data. Preprocess the collected optical transmission and reflection images, such as noise removal and contrast adjustment. Register and align the optical transmission image and the reflection image to ensure the accuracy of subsequent analysis. Use image processing techniques (such as high-pass filtering and edge detection) to extract features from the reflection image on the glass surface.

[0123] This step enhances the visibility of peak intensity defects by grayscale transformation and normalization of the image. The peak intensity regions in the image are identified by statistical analysis methods (such as histogram analysis and local maximum detection). The peak intensity defect regions are separated from the background using a threshold segmentation algorithm. The detected peak intensity defect data are annotated and classified to generate optical defect analysis data. Based on the optical defect analysis data, an initial mask is generated to mark all detected peak intensity defect regions.

[0124] This step optimizes the mask shape by using morphological processing techniques (such as dilation and corrosion) to ensure the accuracy of the defect area boundary. Image fusion technology is used to fuse the defect detection results under multiple angles and multiple light sources to generate the final abnormal optical defect mask. The abnormal optical defect mask is applied to the intact area segmentation image to perform edge segmentation of the optical abnormal area.

[0125] This step refines the segmentation results by using edge detection algorithms (such as Canny edge detection and Sobel operator). The detected optical abnormal areas are marked on the intact area segmentation image. A glass third abnormal area image containing all optical abnormal areas is generated.

[0126] Step S5: Mark the abnormal image areas of the first abnormal area image of the glass, the second abnormal area image of the glass, and the third abnormal area image of the glass to generate a defect area marked image; construct an intelligent identifier for the defect area marked image to obtain a defect intelligent identifier; perform intelligent quality defect detection on the glass perspective stereo image according to the defect intelligent identifier to obtain a quality defect inspection map; upload the quality defect inspection map to the MES system for image storage, and generate a glass quality defect detection storage map to execute quality control management operations for the entire glass production process.

[0127] In the embodiment of the present invention, the image of the first abnormal area of ​​the glass, the image of the second abnormal area of ​​the glass, and the image of the third abnormal area of ​​the glass are registered to ensure that they are aligned in space. Image processing software (such as OpenCV) is used to perform image fusion to superimpose the images of the various abnormal areas. The fused image is marked to mark various abnormal areas.

[0128] In this step, a marking box or outline is drawn on the defect area by using a marking tool or a custom algorithm (such as threshold segmentation, region growing). The marking information is merged into an image file to generate a defect area marking image. The defect area marking image is formatted (such as saved in PNG or JPEG format) for subsequent analysis and processing. The defect area marking images are collected and labeled to extract the type and location information of each defect area.

[0129] This step facilitates model training and evaluation by dividing the labeled image dataset into training and test sets. Select a specific deep learning algorithm (such as convolutional neural network, CNN) to build the smart recognizer. Use deep learning frameworks (such as TensorFlow, PyTorch) to design and train the smart recognition model. Input the features and labels of the labeled images of the defect area and train the smart recognizer to identify and classify different types of defects. Use the test set to evaluate the trained smart recognizer and check its accuracy, recall, and F1 score.

[0130] Optimize the model based on the evaluation results, adjust parameters or add training data to improve the performance of the recognizer. Use the trained defect intelligent recognizer to perform quality inspection on the new glass perspective stereogram. Input the glass perspective stereogram to be inspected, and the intelligent recognizer automatically identifies the defect areas in the image. Overlay the defect areas marked by the intelligent recognizer on the glass perspective stereogram. Generate a quality defect inspection map, clearly marking all quality defect areas in the glass.

[0131] Upload quality defect inspection pictures to the Manufacturing Execution System (MES) through the programming interface (API) or file transfer protocol (FTP). Ensure the security and integrity of data transmission during the upload process. Store quality defect inspection pictures in the MES system for subsequent quality control and analysis. Generate glass quality defect detection storage pictures and record relevant quality inspection information, such as inspection date, production batch number, etc. Use the MES system to monitor and manage the quality of the glass production process. Perform production optimization and defect analysis based on the stored defect data to ensure that product quality meets the standards.

[0132] Preferably, step S1 comprises the following steps:

[0133] Step S11: using a camera to obtain a glass image set from multiple angles;

[0134] Step S12: performing image preprocessing on the glass image set to generate a standard glass image set, wherein the image preprocessing includes image filtering, image brightness enhancement and image grayscale conversion;

[0135] Step S13: extracting glass contour features from the standard glass image set to obtain glass contour feature data; performing image background separation on the standard glass image set using the glass contour feature data to generate a glass background separation image set;

[0136] Step S14: Perform perspective stereoscopic stitching on the glass-background separated image set to generate a glass perspective stereoscopic image.

[0137] In the embodiment of the present invention, a high-resolution camera is installed to ensure that it can capture multiple angles of the glass. The shooting parameters of the camera (such as focal length, aperture, shutter speed) are configured to adapt to the light transmission and reflection characteristics of the glass. The glass sample is fixed on a rotatable platform to ensure that it is photographed from multiple angles. The glass image is taken from different angles (such as 0 degrees, 45 degrees, 90 degrees, etc.) to ensure that all parts of the glass surface are covered.

[0138] Keep the lighting of the shooting environment uniform to avoid reflections and shadows that interfere with image quality. Name and store the acquired images to ensure that images from each angle are clearly marked. Organize the image data into an image set for subsequent processing. Use filtering algorithms (such as Gaussian filtering and mean filtering) to denoise the original image. Remove noise and stray light from the image to improve image quality. Apply brightness adjustment techniques (such as histogram equalization) to enhance the brightness and contrast of the image. Ensure that image details are more obvious, especially the edges and surface features of the glass.

[0139] Convert color images to grayscale images to reduce computational complexity. Use grayscale transformation techniques (such as histogram equalization) to further enhance image contrast. Save the preprocessed images as a standard glass image set for subsequent feature extraction and background separation. Use edge detection algorithms (such as Canny edge detection and Sobel operator) to extract glass contours. Obtain key feature data of glass through contour detection and feature point extraction algorithms (such as Hough transform). Store the extracted contour feature data in a data file, including the coordinates of the feature points and the shape information of the contour, format the feature data to facilitate subsequent background separation operations, and determine the area and parameters for background separation based on the contour feature data.

[0140] This step separates the background area from the glass image in the standard glass image set by using background separation algorithms (such as background modeling and foreground extraction). Generate a glass background separation image set to ensure that the glass part in each image is accurately extracted. Register the glass background separation image set and use feature matching algorithms (such as SIFT and SURF) to ensure that images from different perspectives can be accurately aligned.

[0141] Use stitching algorithms (such as panoramic image stitching technology) to synthesize the aligned images to generate a glass perspective stereogram. Optimize the stitched images, adjust the brightness, contrast and color balance to ensure the visual consistency of the stitching results, and save the stitched images as glass perspective stereograms for subsequent analysis and processing.

[0142] Preferably, step S14 comprises the following steps:

[0143] Step S141: calculating the shooting perspective angle of the glass based on the glass-background separation image set to obtain shooting perspective angle data; grouping the glass-background separation image set according to the shooting perspective angle data to generate perspective group images, wherein the perspective image groups include a front view group, a rear view group, a left view group, a right view group, a top view group, and a bottom view group;

[0144] Step S142: performing group image overlap on the view group images to generate a view group overlap image; performing feature point detection on the view group overlap image to generate feature points of the view group overlap image; using the feature points of the view group overlap image to perform repeated area recognition on the view group overlap image to obtain a group overlap area image;

[0145] Step S143: performing non-overlapping area image registration on the view angle grouping overlapping area image by grouping overlapping area images to obtain a glass view angle grouping registered image;

[0146] Step S144: using the Cartesian space coordinate system to perform spatial perspective stereoscopic stitching on the glass perspective grouping registration images to generate a glass perspective stereoscopic image.

[0147] In the embodiment of the present invention, the shooting angle of each image is calculated by analyzing the relative position and direction of each image in the glass background separation image set. The angle of view error caused by lens distortion or inaccurate camera position is corrected by using a geometric correction algorithm (such as camera calibration). The shooting angle data of each image is recorded, including front view, rear view, left view, right view, top view and bottom view.

[0148] According to the shooting perspective angle data, the images in the glass background separation image set are divided into different perspective groups to generate perspective group images, including: front view group, rear view group, left view group, right view group, top view group and bottom view group. Use image registration algorithms (such as feature matching and mutual information registration) to align the images in the same perspective group, eliminate the perspective difference, and fuse the aligned images to generate perspective group overlap images.

[0149] Perform feature point detection (such as SIFT, SURF, ORB) on the perspective grouping overlapped images, extract significant feature points, and generate descriptors for each feature point for subsequent matching and registration. Use feature point descriptors for region matching, identify repeated areas in the image, mark repeated areas, separate them from non-repeated areas, and obtain grouping overlapped area images. Perform precise registration of non-overlapping areas in the grouping overlapped area images to eliminate displacement and distortion between images of different perspectives, fuse the registered non-overlapping area images with the overlapped area images, and generate glass perspective grouping registered images.

[0150] Further adjustments are made to the generated registered images to ensure smoothness and consistency of the stitching, and to check if there are any seams or overlaps in the registered images and repair them. Define the Cartesian space coordinate system, including the X, Y, and Z axes, which are used to locate each view image in three-dimensional space, and convert the glass view group registered images into the Cartesian coordinate system.

[0151] Spatial alignment is performed on the grouped registered images of each perspective to ensure that they are accurately connected in three-dimensional space. The aligned images are synthesized using three-dimensional image processing technology (such as voxelization and three-dimensional reconstruction) to generate a glass perspective stereogram. The generated stereogram is adjusted to ensure perspective consistency and detail accuracy. The final generated glass perspective stereogram is quality checked to confirm its completeness and accuracy.

[0152] As an example of the present invention, refer to Figure 2 As shown, in this example, step S2 includes:

[0153] Step S21: performing glass region stress analysis on the glass viewing angle stereogram to generate glass region stress data; performing high stress region segmentation on the glass viewing angle stereogram according to the glass region stress data to generate a glass high stress region image;

[0154] Step S22: performing vibration detection on the glass based on the image of the high stress area of ​​the glass, and collecting vibration wrinkle images using a vibration sensor to obtain a glass vibration wrinkle map;

[0155] Step S23: Correlate the glass vibration pattern with the glass high stress region image to generate the glass high stress region vibration pattern;

[0156] Step S24: performing abnormal chattering defect detection on the glass high stress area image through the glass high stress area chattering pattern, thereby generating a glass first abnormal area image.

[0157] In the embodiment of the present invention, the glass perspective stereogram is used as input data. Finite element analysis (FEA) or other stress analysis algorithms are used to calculate the stress of the glass area. The algorithm needs to consider the geometric shape, material properties and external loading of the glass. The stress value of each area is calculated to generate glass area stress data, which includes information such as the intensity and distribution of the stress.

[0158] According to the stress data, a threshold is set to identify high stress areas, and an image segmentation algorithm (such as threshold segmentation and region growing) is applied to segment the glass perspective stereogram, marking the areas above the threshold as high stress areas, creating a glass high stress area image, and showing all high stress areas in the glass. A vibration sensor is installed on or near the glass surface to capture the vibration characteristics of the glass, and the glass is subjected to vibration testing, the vibration signal is recorded, and the vibration data is analyzed using a signal processing algorithm to identify the vibration pattern.

[0159] Use a vibration sensor to collect vibration pattern images, that is, images that reflect vibration patterns. Process the collected vibration pattern images to ensure that the images are clear and can accurately reflect vibration characteristics. Convert the processed vibration data into vibration pattern images and save them as glass vibration pattern maps. Align the glass vibration pattern images with the glass high stress area images to ensure that the two are aligned in space. Use image registration techniques (such as feature point matching and image transformation) to align the two images. Overlay or fuse the glass high stress area images with the glass vibration pattern images to generate glass high stress area vibration pattern maps. Associate the vibration patterns in the vibration pattern maps with the high stress areas to highlight the vibration characteristics in the high stress areas.

[0160] Extract vibration characteristics from vibration patterns in high-stress areas of glass, including vibration intensity, mode, and frequency, and use anomaly detection algorithms (such as threshold detection and machine learning classifiers) to identify abnormal features in vibration patterns. These abnormal features indicate potential defects, mark abnormal vibration detection results in the image of high-stress areas of glass, and generate markers for abnormal areas. Visualize the detected abnormal areas and generate images of the first abnormal areas of glass to ensure that the display of abnormal areas is clear and accurate, which is convenient for subsequent analysis and processing.

[0161] Preferably, step S24 includes the following steps:

[0162] Step S241: performing regional vibration wave amplitude fluctuation analysis on the vibration wave pattern of the glass high stress region to generate a regional vibration wave amplitude fluctuation curve;

[0163] Step S242: calculating the amplitude fluctuation range of the regional chattering amplitude fluctuation curve to obtain amplitude fluctuation range data;

[0164] Step S243: comparing the amplitude fluctuation data with a preset standard amplitude fluctuation threshold, and when the amplitude fluctuation data is greater than the preset standard amplitude fluctuation threshold, confirming the abnormal crack area of ​​the vibration pattern of the glass high stress area based on the amplitude fluctuation data to obtain abnormal crack area data;

[0165] Step S244: segmenting the glass high stress region image into an abnormal crack region image according to the abnormal crack region data to generate a first abnormal region image of the glass.

[0166] In the embodiment of the present invention, the amplitude data of the vibration area is extracted from the vibration pattern of the high stress area of ​​the glass. Image analysis tools (such as image segmentation and region extraction) can be used to identify the region of interest, and the amplitude of the extracted vibration area is calculated to generate the amplitude data of each area. The fluctuation of the amplitude data is analyzed using a signal processing algorithm (such as Fourier transform and time series analysis), and a regional vibration amplitude fluctuation curve is drawn to show the trend of amplitude changes over time or space.

[0167] Calculate the amplitude fluctuation using statistical analysis methods (such as standard deviation, peak-to-peak amplitude) to generate amplitude fluctuation data, which reflects the stability and fluctuation of the vibration area. Record the calculated amplitude fluctuation data for subsequent comparison and analysis. Set a preset standard amplitude fluctuation threshold based on experimental data or industry standards, and compare the amplitude fluctuation data with the preset threshold. If the data is greater than the threshold, mark it as an abnormal area. Mark the area where the amplitude fluctuation is greater than the threshold in the vibration pattern of the high stress area of ​​the glass, and generate abnormal crack area data for subsequent image segmentation.

[0168] Use image segmentation techniques (such as region growing, edge detection, and watershed algorithm) to segment the glass high stress area image and separate the abnormal crack area. Extract the area marked according to the abnormal crack area data from the glass high stress area image to generate the glass first abnormal area image. The generated glass first abnormal area image highlights the abnormal crack area, which is convenient for further analysis and processing.

[0169] As an example of the present invention, refer to Figure 3 As shown, in this example, step S3 includes:

[0170] Step S31: segmenting the glass viewing angle stereogram into intact regions using the first abnormal region image of the glass to obtain an intact region segmentation image;

[0171] Step S32: performing hot melt detection on the glass according to the intact area segmentation image, and collecting a thermal map using an infrared camera to obtain a thermal map of the intact area of ​​the glass; extracting peak temperature information from the thermal map of the intact area of ​​the glass to obtain peak temperature information data;

[0172] Step S33: extracting the high heat-melting area from the intact area segmentation image based on the peak temperature information data to obtain a high heat-melting area image;

[0173] Step S34: performing abnormal deformation defect detection on the high-thermal melting area image to generate a second abnormal area image of the glass.

[0174] In an embodiment of the present invention, an image of the first abnormal area of ​​the glass is used, and the image marks the defective area. The intact area is separated from the glass perspective stereogram. Image subtraction or other segmentation methods can be used to exclude the defective area to obtain a segmented image of the intact area, showing all the areas marked as intact in the glass. Hot melt testing of glass usually involves a heating and cooling process. An infrared camera is used to photograph the glass, record thermal data, and generate a thermal map of the intact area of ​​the glass. Peak temperature information is extracted from the thermal map. Image processing techniques (such as thresholding and regional analysis) can be used to identify and record the highest temperature value of each area, and create peak temperature information data, which reflects the hot melt characteristics of different areas.

[0175] Set a threshold for the peak temperature to identify high-melting areas. The threshold can be set based on standard data or experimental data. Extract areas with temperatures above the threshold from the intact area segmentation image. This can be achieved through image segmentation techniques (such as threshold segmentation and region growing) to create a high-melting area image that shows areas with higher thermal melting properties in the glass. Perform a detailed deformation analysis on the high-melting area image. Image processing techniques (such as edge detection and morphological operations) can be used to identify abnormal deformations, and defect detection algorithms (such as machine learning models and template matching) can be applied to identify deformation defects. These algorithms need to detect abnormal areas based on preset defect features. Mark the detected abnormal deformation areas and generate a second abnormal area image of the glass to ensure that the abnormal area is displayed clearly and accurately reflects the deformation defects in the glass.

[0176] Preferably, step S34 includes the following steps:

[0177] Step S341: performing isothermal time-series image analysis on the high thermal melting area image to generate a high thermal melting area time-series image set;

[0178] Step S342: performing image pixel position calculation on the high thermal melt region time series image set to obtain an image pixel position data set; performing adjacent time series image pixel offset calculation on the image pixel position data set to obtain a high thermal melt region pixel offset map;

[0179] Step S343: generating a displacement vector field for the pixel offset map of the high thermal melting area by using an optical flow method to obtain a pixel displacement vector field;

[0180] Step S344: Analyze the displacement amplitude of the high thermal melt area image through the pixel displacement vector field to generate abnormal deformation detection data; perform image edge segmentation on the high thermal melt area image based on the abnormal deformation detection data to generate a second abnormal area image of the glass.

[0181] In the embodiment of the present invention, by periodically capturing images of the high-melting area during the glass heating process, it is ensured that multiple time-series images under the same thermal state are collected. These images are organized into a time-series image set, each image records the state of the hot-melting area at different time points, and the images are arranged in chronological order to facilitate subsequent analysis. Feature points are extracted from each image, which are usually high-contrast points or texture points, and the pixel position of each feature point in the image is recorded to generate an image pixel position data set.

[0182] Compare adjacent time-series images, calculate the pixel offset of the same feature point at different time points, and plot the calculated pixel offset into a high-melt area pixel offset map to show the displacement of each feature point. Use optical flow methods (such as the Lucas-Kanade method and the Horn-Schunck method) to calculate the motion vector of the pixel. These methods estimate the movement of the pixel by analyzing the brightness change of the image, generate the displacement vector of each pixel, form a pixel displacement vector field, and display the pixel displacement vector field, using arrows or vector field diagrams to indicate the displacement direction and amplitude of each pixel.

[0183] Calculate the displacement amplitude of each pixel in the pixel displacement vector field. The vector modulus can be used to quantify the displacement amplitude and determine abnormal displacement amplitudes, which usually exceed the preset threshold and indicate deformation defects. Apply image edge detection algorithms (such as Canny edge detection) to process the displacement amplitude analysis results, highlight the edges of the abnormal deformation area, and generate the second abnormal area image of the glass based on the edge segmentation results to mark the abnormal deformation area.

[0184] As an example of the present invention, refer to Figure 4 As shown, in this example, step S4 includes:

[0185] Step S41: optically transmit and reflect the glass according to the intact area segmentation image, and collect optical transmission and reflection images through a camera to obtain a glass surface reflection image;

[0186] Step S42: performing fast Fourier transform on the glass surface reflection image to obtain a surface reflection spectrum diagram; performing amplitude spectrum and phase spectrum calculation on the surface reflection spectrum diagram to obtain reflection amplitude data and reflection phase data;

[0187] Step S43: performing peak intensity defect analysis according to the reflection amplitude data and the reflection phase data to generate optical defect analysis data; creating a mask based on the optical defect analysis data to obtain an abnormal optical defect mask;

[0188] Step S44: performing inverse Fourier transform on the abnormal optical defect mask to obtain the abnormal optical defect area; mapping the abnormal optical defect area to the intact area segmentation image to perform optical abnormal area edge segmentation, thereby generating a third abnormal area image of the glass.

[0189] In the embodiment of the present invention, the transmission and reflection images of the glass surface are captured by using an optical device and a camera. The device should ensure that detailed optical information of the glass surface can be captured. According to the characteristics of the glass, the optical device is adjusted to obtain the best transmission and reflection images.

[0190] The reflected images of the glass surface are recorded by a camera. These images should include reflection data at different angles and lighting conditions. Fast Fourier transform is performed on the reflected images of the glass surface to convert the spatial domain images to the frequency domain, which helps to analyze the frequency components in the image. The amplitude of the spectrum is calculated to display the intensity of the frequency components in the image, and the phase of the spectrum is calculated to display the phase information of the frequency components in the image. The reflection amplitude data and reflection phase data are generated, which will be used for further defect analysis. Identify the peaks in the amplitude spectrum, which usually represent abnormal optical defect areas.

[0191] The intensity and position of the spike can indicate potential defects. Combined with the data of the phase spectrum, further analysis is performed to see if the spike has specific phase characteristics, which helps to accurately locate the defect. Based on the results of the spike intensity defect analysis, a mask image is created, which highlights the location of the defect area. The abnormal optical defect mask is inversely Fourier transformed to convert the frequency domain data back to the spatial domain, which can show the location and morphology of the abnormal optical defect area in the original image.

[0192] In this step, the abnormal optical defect area obtained by inverse Fourier transform is mapped to the segmented image of the intact area. The mapped image is edge detected to highlight the edge of the optical abnormal area. Based on the edge segmentation result, the third abnormal area image of the glass is generated to mark the optical abnormal area.

[0193] Preferably, step S5 comprises the following steps:

[0194] Step S51: integrating the first abnormal area image of the glass, the second abnormal area image of the glass and the third abnormal area image of the glass to generate an abnormal area image; marking defect areas of the glass viewing angle stereogram based on the abnormal area image to obtain a defect area marking image;

[0195] Step S52: dividing the defect area marked image into a data set to obtain a model training set and a model test set; using a convolutional neural network algorithm to perform model training on the model training set to generate an anomaly recognition trainer; using the model test set to optimize and iterate the anomaly recognition trainer to generate an intelligent defect identifier; performing intelligent quality defect detection on the glass perspective stereogram according to the intelligent defect identifier to obtain a quality defect inspection map;

[0196] Step S53: Upload the quality defect inspection image to the MES system for image storage, generate a glass quality defect detection storage image, and perform quality control management operations throughout the glass production process.

[0197] In the embodiment of the present invention, the first abnormal area image of the glass (crack detection), the second abnormal area image of the glass (hot melt detection), and the third abnormal area image of the glass (optical defect detection) are taken as input. The three abnormal area images are integrated into a comprehensive abnormal area image using image fusion technology (such as weighted average or image stitching), and the synthesized image should highlight the positions and shapes of all abnormal areas.

[0198] Based on the generated abnormal area image, mark all abnormal areas in the glass perspective stereogram. Use a marking tool (such as graphic annotation software) to mark the defective areas in the glass perspective stereogram to generate a defective area marked image. The image should clearly show all detected defective areas for subsequent intelligent detection model training. Integrate the defective area marked image with its corresponding label data to form a training and test data set.

[0199] In this step, the dataset is divided into a model training set and a model test set according to a certain ratio (e.g., 80% training set, 20% test set). Select a convolutional neural network architecture suitable for image recognition, such as VGG, ResNet, or Inception. Set training parameters (e.g., learning rate, batch size, number of training rounds) and initialize the network. Input the training set into the CNN for model training, optimize the network parameters to minimize the loss function, and after training, generate a trainer for anomaly recognition. Use the model test set to evaluate the trainer and test the performance of the model on unseen data.

[0200] According to the test results, the model parameters and structure are adjusted, and multiple iterations of optimization are performed. After the optimization is completed, an intelligent defect identifier is obtained for high-precision defect detection. The intelligent defect identifier is used to perform intelligent quality defect detection on the glass perspective stereogram, and a quality defect inspection map is generated based on the test results, marking all detected defect areas. The quality defect inspection map is uploaded to the system through the interface of the MES system.

[0201] The MES system stores the quality inspection images in the database for subsequent quality control and recording, and generates a storage map containing all quality defect information to facilitate the execution of quality control management tasks for the entire glass production process. Through the MES system, quality problems in the production process can be monitored in real time and corresponding quality control measures can be implemented.

[0202] Embodiment 2: This embodiment provides a glass quality defect detection system based on AI image recognition, which is used to execute the glass quality defect detection method based on AI image recognition described in the above embodiment 1. The glass quality defect detection system based on AI image recognition includes:

[0203] A stereoscopic stitching module is used to obtain a glass image set from multiple angles; perform image contour and background separation on the glass image set to generate a glass background separation image set; perform perspective stereoscopic stitching on the glass background separation image set to generate a glass perspective stereogram;

[0204] The vibration detection module is used to segment the glass high stress area of ​​the glass perspective stereogram to generate a glass high stress area image; perform vibration detection on the glass perspective stereogram, and use a vibration sensor to collect vibration pattern images to obtain a glass vibration pattern image; perform abnormal vibration pattern defect detection on the glass high stress area image through the glass vibration pattern image and the glass high stress area image, thereby generating a glass first abnormal area image;

[0205] The hot melt detection module is used to segment the glass perspective stereogram into intact areas through the first abnormal area image of the glass to obtain the segmented image of the intact area; perform hot melt detection on the glass according to the segmented image of the intact area, and use an infrared camera to collect the thermal image to obtain the thermal image of the intact area of ​​the glass; perform abnormal deformation defect detection on the thermal image of the intact area of ​​the glass to generate the second abnormal area image of the glass;

[0206] The optical inspection module is used to perform optical transmission and reflection on the glass according to the intact area segmentation image, and collect optical transmission and reflection images through a camera to obtain a glass surface reflection image; perform peak intensity defect analysis on the glass surface reflection image to generate optical defect analysis data; create a mask based on the optical defect analysis data to obtain an abnormal optical defect mask; use the abnormal optical defect mask to perform optical abnormal area edge segmentation on the intact area segmentation image, thereby generating a glass third abnormal area image;

[0207] The intelligent recognition module is used to mark the abnormal image areas of the first abnormal area image of the glass, the second abnormal area image of the glass and the third abnormal area image of the glass to generate a defect area marked image; construct an intelligent identifier for the defect area marked image to obtain a defect intelligent identifier; perform intelligent quality defect detection on the glass perspective stereo image according to the defect intelligent identifier to obtain a quality defect inspection map; upload the quality defect inspection map to the MES system for image storage to generate a glass quality defect detection storage map to execute quality control management operations for the entire glass production process.

[0208] As described above, the present invention can fully understand the surface condition of the glass by acquiring images from multiple angles. The generated perspective stereogram can provide a comprehensive perspective, which is convenient for subsequent defect detection and analysis. The segmentation of high stress areas helps to locate the weak points of the glass.

[0209] This system can detect potential defects caused by stress through vibration detection and vibration pattern image acquisition. Abnormal vibration pattern detection helps to identify potential risks of glass during use. The segmentation of flawless areas can help to clarify which parts of the glass surface are flawless.

[0210] This system can check the stability of glass under high temperature conditions through hot melt testing. The durability and stability of glass in practical applications can be further ensured through the acquisition of thermal images and abnormal deformation detection. Optical transmission and reflection image acquisition can detect abnormal reflections on the glass surface.

[0211] This system helps to find tiny optical defects through peak intensity defect analysis and optical abnormal area edge segmentation, which affect the visual effect and optical performance of the glass. Abnormal area marking can accurately record various defect types. The intelligent identifier can automatically identify defect areas and improve detection efficiency. The generation and storage of quality defect inspection pictures can ensure quality control in the production process and support subsequent quality traceability and improvement.

[0212] Therefore, the present invention improves the accuracy, comprehensiveness and automation level of quality defect detection through multi-angle image acquisition, comprehensive defect detection, intelligent identification marking and systematic management.

[0213] As described above, only preferred embodiments are disclosed in the present invention, but it is not limited thereto. A person skilled in the art can easily understand the spirit of the present invention based on the above embodiments and make different extensions and changes. However, as long as they do not deviate from the spirit of the present invention, they are all within the protection scope of the present invention.

Claims

1. A glass quality defect detection method based on AI image recognition, characterized in that: The following steps are involved: Step S1: Acquire a glass image set from multiple angles; perform image contour and background separation on the glass image set to generate a glass background separation image set; Perform perspective stereo stitching on the glass-background separated image set to generate a glass perspective stereogram; Step S2: segmenting the glass high stress area of ​​the glass viewing angle stereogram to generate a glass high stress area image; performing vibration detection on the glass viewing angle stereogram, and using a vibration sensor to collect vibration pattern images to obtain a glass vibration pattern image; performing abnormal vibration pattern defect detection on the glass high stress area image through the glass vibration pattern image and the glass high stress area image, thereby generating a glass first abnormal area image; Step S3: segmenting the glass perspective stereogram into intact areas through the first abnormal area image of the glass to obtain an intact area segmentation image; performing hot melt detection on the glass according to the intact area segmentation image, and collecting thermal images using an infrared camera to obtain a thermal map of the intact area of ​​the glass; performing abnormal deformation defect detection on the thermal map of the intact area of ​​the glass to generate an image of the second abnormal area of ​​the glass; Step S4: optically transmit and reflect the glass according to the intact area segmentation image, and collect optical transmission and reflection images through a camera to obtain a glass surface reflection image; perform peak intensity defect analysis on the glass surface reflection image to generate optical defect analysis data; A mask is created based on the optical defect analysis data to obtain an abnormal optical defect mask; the optical abnormal region edge segmentation is performed in the intact region segmentation image using the abnormal optical defect mask, thereby generating a third abnormal region image of the glass; Step S5: marking the abnormal image regions of the first abnormal glass region image, the second abnormal glass region image and the third abnormal glass region image to generate a defect region marking image; An intelligent identifier is constructed for the defect area marked image to obtain a defect intelligent identifier; Perform intelligent quality defect detection on the glass viewing angle stereogram according to the defect intelligent identifier to obtain a quality defect inspection map; Upload the quality defect inspection image to the MES system for image storage, generate a glass quality defect detection storage image, and perform quality control management of the entire glass production process.

2. The glass quality defect detection method based on AI image recognition according to claim 1, characterized in that: Step S1 includes the following steps: Step S11: using a camera to obtain a glass image set from multiple angles; Step S12: performing image preprocessing on the glass image set to generate a standard glass image set, wherein the image preprocessing includes image filtering, image brightness enhancement and image grayscale conversion; Step S13: extracting glass contour features from the standard glass image set to obtain glass contour feature data; performing image background separation on the standard glass image set using the glass contour feature data to generate a glass background separation image set; Step S14: Perform perspective stereoscopic stitching on the glass-background separated image set to generate a glass perspective stereoscopic image.

3. The glass quality defect detection method based on AI image recognition according to claim 2, characterized in that: Step S14 includes the following steps: Step S141: calculating the shooting perspective angle of the glass based on the glass-background separation image set to obtain shooting perspective angle data; grouping the glass-background separation image set according to the shooting perspective angle data to generate perspective group images, wherein the perspective image groups include a front view group, a rear view group, a left view group, a right view group, a top view group, and a bottom view group; Step S142: performing group image overlap on the view group images to generate a view group overlap image; performing feature point detection on the view group overlap image to generate feature points of the view group overlap image; using the feature points of the view group overlap image to perform repeated area recognition on the view group overlap image to obtain a group overlap area image; Step S143: performing non-overlapping area image registration on the view angle grouping overlapping area image by grouping overlapping area images to obtain a glass view angle grouping registered image; Step S144: using the Cartesian space coordinate system to perform spatial perspective stereoscopic stitching on the glass perspective grouping registration images to generate a glass perspective stereoscopic image.

4. The glass quality defect detection method based on AI image recognition according to claim 1, characterized in that: Step S2 includes the following steps: Step S21: performing glass region stress analysis on the glass viewing angle stereogram to generate glass region stress data; performing high stress region segmentation on the glass viewing angle stereogram according to the glass region stress data to generate a glass high stress region image; Step S22: performing vibration detection on the glass based on the image of the high stress area of ​​the glass, and collecting vibration wrinkle images using a vibration sensor to obtain a glass vibration wrinkle map; Step S23: Correlate the glass vibration pattern with the glass high stress region image to generate the glass high stress region vibration pattern; Step S24: performing abnormal chattering defect detection on the glass high stress area image through the glass high stress area chattering pattern, thereby generating a glass first abnormal area image.

5. The glass quality defect detection method based on AI image recognition according to claim 4, characterized in that: Step S24 includes the following steps: Step S241: performing regional vibration wave amplitude fluctuation analysis on the vibration wave pattern of the glass high stress region to generate a regional vibration wave amplitude fluctuation curve; Step S242: calculating the amplitude fluctuation range of the regional chattering amplitude fluctuation curve to obtain amplitude fluctuation range data; Step S243: comparing the amplitude fluctuation data with a preset standard amplitude fluctuation threshold, and when the amplitude fluctuation data is greater than the preset standard amplitude fluctuation threshold, confirming the abnormal crack area of ​​the vibration pattern of the glass high stress area based on the amplitude fluctuation data to obtain abnormal crack area data; Step S244: segmenting the glass high stress region image into an abnormal crack region image according to the abnormal crack region data to generate a first abnormal region image of the glass.

6. The glass quality defect detection method based on AI image recognition according to claim 1, characterized in that: Step S3 includes the following steps: Step S31: segmenting the glass viewing angle stereogram into intact regions using the first abnormal region image of the glass to obtain an intact region segmentation image; Step S32: performing hot melt detection on the glass according to the intact area segmentation image, and collecting a thermal map using an infrared camera to obtain a thermal map of the intact area of ​​the glass; extracting peak temperature information from the thermal map of the intact area of ​​the glass to obtain peak temperature information data; Step S33: extracting the high heat-melting area from the intact area segmentation image based on the peak temperature information data to obtain a high heat-melting area image; Step S34: performing abnormal deformation defect detection on the high-thermal melting area image to generate a second abnormal area image of the glass.

7. The glass quality defect detection method based on AI image recognition according to claim 6, characterized in that: Step S34 includes the following steps: Step S341: performing isothermal time-series image analysis on the high thermal melting area image to generate a high thermal melting area time-series image set; Step S342: performing image pixel position calculation on the high thermal melt region time series image set to obtain an image pixel position data set; performing adjacent time series image pixel offset calculation on the image pixel position data set to obtain a high thermal melt region pixel offset map; Step S343: generating a displacement vector field for the pixel offset map of the high thermal melting area by using an optical flow method to obtain a pixel displacement vector field; Step S344: Analyze the displacement amplitude of the high thermal melt area image through the pixel displacement vector field to generate abnormal deformation detection data; perform image edge segmentation on the high thermal melt area image based on the abnormal deformation detection data to generate a second abnormal area image of the glass.

8. The glass quality defect detection method based on AI image recognition according to claim 1, characterized in that: Step S4 includes the following steps: Step S41: optically transmit and reflect the glass according to the intact area segmentation image, and collect optical transmission and reflection images through a camera to obtain a glass surface reflection image; Step S42: performing fast Fourier transform on the glass surface reflection image to obtain a surface reflection spectrum diagram; performing amplitude spectrum and phase spectrum calculation on the surface reflection spectrum diagram to obtain reflection amplitude data and reflection phase data; Step S43: performing peak intensity defect analysis according to the reflection amplitude data and the reflection phase data to generate optical defect analysis data; creating a mask based on the optical defect analysis data to obtain an abnormal optical defect mask; Step S44: performing inverse Fourier transform on the abnormal optical defect mask to obtain the abnormal optical defect area; mapping the abnormal optical defect area to the intact area segmentation image to perform optical abnormal area edge segmentation, thereby generating a third abnormal area image of the glass.

9. The glass quality defect detection method based on AI image recognition according to claim 1, characterized in that: Step S5 includes the following steps: Step S51: integrating the first abnormal area image of the glass, the second abnormal area image of the glass and the third abnormal area image of the glass to generate an abnormal area image; marking defect areas of the glass viewing angle stereogram based on the abnormal area image to obtain a defect area marking image; Step S52: dividing the defect area marked image into a data set to obtain a model training set and a model test set; using a convolutional neural network algorithm to perform model training on the model training set to generate an anomaly recognition trainer; using the model test set to optimize and iterate the anomaly recognition trainer to generate an intelligent defect identifier; performing intelligent quality defect detection on the glass perspective stereogram according to the intelligent defect identifier to obtain a quality defect inspection map; Step S53: Upload the quality defect inspection image to the MES system for image storage, generate a glass quality defect detection storage image, and perform quality control management operations throughout the glass production process.

10. A glass quality defect detection system based on AI image recognition, characterized in that: A system for performing a glass quality defect detection method based on AI image recognition as claimed in any one of claims 1 to 9, the system comprising: A stereoscopic stitching module is used to obtain a glass image set from multiple angles; perform image contour and background separation on the glass image set to generate a glass background separation image set; perform perspective stereoscopic stitching on the glass background separation image set to generate a glass perspective stereogram; The vibration detection module is used to segment the glass high stress area of ​​the glass perspective stereogram to generate a glass high stress area image; perform vibration detection on the glass perspective stereogram, and use a vibration sensor to collect vibration pattern images to obtain a glass vibration pattern image; perform abnormal vibration pattern defect detection on the glass high stress area image through the glass vibration pattern image and the glass high stress area image, thereby generating a glass first abnormal area image; The hot melt detection module is used to segment the glass perspective stereogram into intact areas through the first abnormal area image of the glass to obtain the segmented image of the intact area; perform hot melt detection on the glass according to the segmented image of the intact area, and use an infrared camera to collect the thermal image to obtain the thermal image of the intact area of ​​the glass; perform abnormal deformation defect detection on the thermal image of the intact area of ​​the glass to generate the second abnormal area image of the glass; The optical inspection module is used to perform optical transmission and reflection on the glass according to the intact area segmentation image, and collect optical transmission and reflection images through a camera to obtain a glass surface reflection image; perform peak intensity defect analysis on the glass surface reflection image to generate optical defect analysis data; create a mask based on the optical defect analysis data to obtain an abnormal optical defect mask; use the abnormal optical defect mask to perform optical abnormal area edge segmentation on the intact area segmentation image, thereby generating a glass third abnormal area image; The intelligent recognition module is used to mark the abnormal image areas of the first abnormal area image of the glass, the second abnormal area image of the glass and the third abnormal area image of the glass to generate a defect area marked image; construct an intelligent identifier for the defect area marked image to obtain a defect intelligent identifier; perform intelligent quality defect detection on the glass perspective stereo image according to the defect intelligent identifier to obtain a quality defect inspection map; upload the quality defect inspection map to the MES system for image storage to generate a glass quality defect detection storage map to execute quality control management operations for the entire glass production process.

Images