On-line detection method and system for surface defects of vehicle-mounted curved-surface anti-dazzle glass

Abstract

The invention relates to the technical field of glass detection, in particular to an on-line detection method and system for surface defects of vehicle-mounted curved-surface anti-dazzle glass, and the method comprises the steps: collecting surface images of the vehicle-mounted curved-surface anti-dazzle glass through a linear array camera, an annular light source and a polarized light assembly at different angles; preprocessing the collected surface image, including curved surface distortion correction, de-noising enhancement and region extraction, eliminating detection interference caused by a curved surface form and an anti-dazzle coating, and outputting a standardized image; performing defect candidate region positioning on the standardized image based on an improved deep learning model, extracting geometric features, gray features and texture features of defects, and forming defect feature vectors; accurate classification and grade evaluation of the defects are completed through multi-dimensional feature fusion and threshold self-adaptive judgment, and a detection report and a defect visual map are generated. Through multi-link optimization of the visual inspection technology, the detection efficiency and the detection precision are improved, and the quality inspection requirements of large-scale production of the vehicle-mounted glass are met.

Description

Technical Field

[0001] This invention relates to the field of glass inspection technology, specifically to an online detection method and system for surface defects in automotive curved anti-glare glass. Background Technology

[0002] As a core component of automotive windshields and side windows, the surface quality of curved anti-glare glass directly affects driving safety and comfort. This type of glass is characterized by its complex curved shape and anti-glare coating, making it prone to defects such as scratches, bubbles, pitting, and coating peeling during production. Strict online inspection is necessary to eliminate substandard products.

[0003] Existing methods for detecting surface defects in automotive curved anti-glare glass have many technical shortcomings: Poor adaptability to curved surfaces: Traditional visual inspection uses a flat camera to acquire images. The curvature of curved glass causes image stretching and distortion, resulting in low defect positioning accuracy, especially in edge areas where defects are easily missed. Anti-glare coatings cause significant interference: The microstructure of anti-glare coatings generates diffuse reflection and texture noise, making it difficult for traditional detection methods to distinguish between coating textures and real defects, resulting in a high false detection rate. Incomplete defect detection: Under single-view and single-light source modes, the imaging contrast of different types of defects (such as shallow scratches and micro bubbles) is low, making it impossible to simultaneously meet the detection needs of various types of defects. The contradiction between testing efficiency and accuracy: manual testing is inefficient (testing time for a single piece of glass ≥30s) and highly subjective. Existing automated testing methods sacrifice testing accuracy in pursuit of efficiency, making it difficult to meet the high requirements of large-scale production. Fixed threshold: Using a fixed threshold to determine the defect level cannot adapt to the fluctuations in glass quality under different production batches and different process parameters, resulting in poor flexibility.

[0004] Therefore, there is an urgent need for an optimized solution based on vision inspection to solve problems such as curved surface adaptation, coating interference, and full coverage defect detection, so as to achieve rapid, accurate, and online detection of surface defects of automotive curved anti-glare glass. Summary of the Invention

[0005] To address the aforementioned technical shortcomings, this invention provides an online detection method and system for surface defects in automotive curved anti-glare glass. Through multi-view visual acquisition, targeted image preprocessing, improved deep learning detection, and adaptive evaluation, it achieves high efficiency and accuracy in defect detection.

[0006] This invention is achieved through the following technical solution: Step S10: Build a multi-view visual inspection platform. Based on the surface parameters of the vehicle-mounted curved anti-glare glass and the production line speed, configure inspection equipment and collect multi-dimensional images of the glass surface. Transmit the collected raw image data to the image processing unit. Step S20: The image processing unit preprocesses the original image data, sequentially performing surface distortion correction, image denoising enhancement, and region extraction, and outputs a standardized image; Step S30: Based on the improved deep learning model, locate the candidate region of the defect in the standardized image, extract the geometric features, grayscale features and texture features of the defect, and form a defect feature vector; Step S40: Analyze the defect feature vector using a multi-dimensional feature fusion algorithm, identify and classify the defect type by combining adaptive threshold judgment, generate detection results and feed them back to the production control system.

[0007] Preferably, the construction of the multi-view visual inspection platform in step S10 includes: The system includes a linear array camera, a ring diffuse light source, a polarizing filter, a laser ranging unit, and a synchronization control unit. The linear array camera is distributed along the normal direction of the glass surface and at ±30° and ±45° side views, with the acquisition frame rate matched to the pipeline speed. The ring diffuse light source uses an LED light source. The polarizing filter has a dynamically adjustable polarization angle, with an adjustment range of 0° to 90°, used to eliminate the reflection interference of the anti-glare coating. The laser ranging unit acquires the distance data from the glass surface to the linear array camera in real time, providing a coordinate reference for surface distortion correction.

[0008] Preferably, the step of acquiring a multi-dimensional image of the glass surface in step S10 includes: The synchronous control unit triggers the linear array camera and the light source to work together based on the signal from the production line encoder. The laser ranging unit synchronously collects the object distance data corresponding to each pixel. Taking into account the coating characteristics of the anti-glare glass, the polarization filter angle is adjusted to suppress surface specular reflection. The angle is adjusted from 30° to 60°. Simultaneously, images in three modes—no polarization, low polarization, and high polarization—are collected to form a multimodal image set. The no polarization image retains the overall information of the defect, the low polarization image highlights linear defects such as shallow scratches, and the high polarization image enhances point defects such as bubbles and pits, ensuring the imaging clarity of different types of defects.

[0009] Preferably, the step of preprocessing the original image data by the image processing unit in step S20 includes: Curved surface distortion correction: Based on the object distance data obtained by the laser ranging unit, a three-dimensional coordinate model of the glass surface is established. Perspective transformation combined with nonlinear interpolation algorithm is used to map the curved surface image to a two-dimensional plane to correct the image stretching and deformation caused by the curvature of the curved surface. The camera's intrinsic and extrinsic parameters are calibrated by a calibration plate to ensure that the dimensional accuracy error of the corrected image is ≤0.1mm. Image denoising and enhancement: For Gaussian noise and salt-and-pepper noise in the multimodal image set, adaptive median filtering is used to remove salt-and-pepper noise, with the window size dynamically adjusted from 3×3 to 7×7. Gaussian noise is suppressed by wavelet threshold denoising algorithm. The illumination component and reflectance component of the image are separated based on the Retinex algorithm to enhance the grayscale contrast between the defect area and the background. Pixel-level fusion is performed on images under three polarization modes, including no polarization, low polarization and high polarization. The weighted average method is used to improve the overall signal-to-noise ratio of the image. The weight of the no polarization image is set to 0.4, and the weights of the low polarization and high polarization images are each 0.3. Region extraction: Based on the contour features of the curved anti-glare glass in the vehicle, the glass region contour is extracted through edge detection algorithm. The threshold is adaptively adjusted to remove the background region in the image. According to the size and specifications of the glass, the effective detection region is cropped to reduce the amount of invalid data processing and improve detection efficiency.

[0010] Preferably, step S30, which involves locating defect candidate regions in a standardized image based on an improved deep learning model, includes: Defect candidate region localization: An improved YOLOv9 model is used to localize the defect candidate region. By adding an adaptive receptive field unit (ARF) to the neck, the receptive field size is dynamically adjusted to adapt to the detection requirements of defects of different sizes. Defects ≤100μm correspond to small receptive fields, and defects >1mm correspond to large receptive fields. Multi-scale feature pyramids are constructed on the standardized images. The bounding boxes of defect samples are clustered by the anchor box clustering algorithm to optimize the anchor box parameters and improve the defect recall rate. Defect feature extraction: For the located candidate regions, geometric features, grayscale features, and texture features are extracted. Geometric features include defect area, perimeter, roundness, and aspect ratio. Grayscale features include the average grayscale of the defect region, grayscale variance, and grayscale difference between the defect region and the background. Texture features are extracted based on the gray-level co-occurrence matrix (GLCM) to extract four parameters: energy, entropy, contrast, and correlation. At the same time, the texture characteristics of the anti-glare coating are combined to add coating consistency features. The texture similarity between the defect region and the surrounding normal coating is calculated, which is used to distinguish coating defects from surface contamination. Feature vector construction: All extracted features are normalized to the [0,1] interval using Min-Max normalization to eliminate dimensional differences. Redundant features are removed using variance screening, and features with variance ≥0.05 are retained to construct a defect feature vector with dimensions of 20-30.

[0011] Preferably, the training process of the improved YOLOv9 model includes: constructing a defect dataset for automotive curved anti-glare glass, containing at least 100,000 labeled images, covering 6 types of defects and different lighting and angle scenes; expanding the dataset using methods such as Mosaic data augmentation, random flipping, and brightness and contrast adjustment; adopting an adaptive learning rate strategy during training, setting an initial learning rate of 0.01, and decaying it to 0.1 times the original rate every 10 epochs; and using CIoU Loss combined with FocalLoss as the loss function to solve the problem of imbalanced defect samples.

[0012] Preferably, step S40, which involves analyzing the defect feature vector using a multi-dimensional feature fusion algorithm and combining it with adaptive threshold determination to identify and classify the defect type, includes: Multi-dimensional feature fusion: The AMFN attention mechanism is used to perform weighted fusion of defect feature vectors. The importance weight of each feature is calculated through the self-attention module to strengthen the focus on key features for defect identification. Defect type identification: The fused feature vector is input into the support vector machine (SVM) classifier. The preset defect types include six categories: scratches, bubbles, pits, coating peeling, dents, and contamination residue. The classifier parameters are optimized through cross-validation. Defect level classification: Based on the geometric characteristics of defects and the requirements of usage scenarios, a level classification standard is formulated, which is divided into level A, level B, level C and level D. An adaptive threshold algorithm is adopted to dynamically adjust the level judgment threshold according to the inspection data of glass batches to adapt to the quality requirements of different production processes. Inspection result output: Generates an inspection report containing glass number, inspection time, defect type, defect level, defect coordinates and defect image. Simultaneously generates a defect visualization map, marking the defect location and range in the map. The inspection results are fed back to the production control system in real time. For defects of level C and above, an alarm mechanism is triggered. For defects of level D, the production line is stopped for timely handling.

[0013] Furthermore, to achieve the above objectives, the present invention also proposes an online detection system for surface defects of automotive curved anti-glare glass, the online detection system for surface defects of automotive curved anti-glare glass comprising: Multi-dimensional image acquisition module for automotive curved anti-glare glass surface: used to build a multi-view visual inspection platform. Based on the surface parameters of automotive curved anti-glare glass and the production line speed, it configures inspection equipment and acquires multi-dimensional images of the glass surface, and transmits the acquired raw image data to the image processing unit. Image preprocessing module: Used by the image processing unit to preprocess the raw image data, sequentially completing surface distortion correction, image denoising enhancement and region extraction, and outputting a standardized image; Vehicle-mounted curved anti-glare glass defect feature extraction module: used to locate defect candidate regions in standardized images based on an improved deep learning model, extract the geometric features, grayscale features and texture features of defects, and form a defect feature vector; Vehicle-mounted curved anti-glare glass defect identification and assessment module: It analyzes the defect feature vector through a multi-dimensional feature fusion algorithm, identifies and classifies the defect type by combining adaptive threshold judgment, generates detection results and feeds them back to the production control system.

[0014] Furthermore, to achieve the above objectives, the present invention also proposes an online detection device for surface defects of vehicle-mounted curved anti-glare glass. The device includes: a memory, a processor, and programs such as an online detection algorithm for surface defects of vehicle-mounted curved anti-glare glass stored in the memory and executable on the processor. The online detection algorithm for surface defects of vehicle-mounted curved anti-glare glass comprises the steps for implementing the online detection method for surface defects of vehicle-mounted curved anti-glare glass as described above.

[0015] In addition, to achieve the above objectives, the present invention also provides a computer program product, which includes programs such as an online detection algorithm for surface defects of vehicle-mounted curved anti-glare glass. When the online detection algorithm for surface defects of vehicle-mounted curved anti-glare glass is executed by a processor, it implements the online detection method for surface defects of vehicle-mounted curved anti-glare glass as described above.

[0016] The advantages and effects of this invention are: This invention proposes an online detection method and system for surface defects on automotive curved anti-glare glass. It acquires precise object distance data through a laser ranging module and combines perspective transformation and nonlinear interpolation to correct surface distortion, solving the problems of image stretching and inaccurate defect localization in curved glass. Simultaneously, it employs multi-polarization mode image acquisition combined with the Retinex enhancement algorithm to effectively suppress reflections and texture noise in the anti-glare coating. Furthermore, by extracting coating consistency features, it accurately distinguishes between real defects and surface contamination, reducing the false detection rate. The multi-view camera combined with a multi-polarization mode acquisition scheme, along with the multi-scale detection capabilities of the improved YOLOv9 model, achieves accurate defect detection. In addition, it uses an adaptive threshold to determine the defect level, dynamically adapting to quality fluctuations in different production batches. The detection results are fed back to the production control system in real time, enabling timely interception of non-conforming products and improving production quality control. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of an online detection method for surface defects of vehicle-mounted curved anti-glare glass according to the present invention.

[0019] Figure 2 This is a schematic diagram of the structure of an online detection system for surface defects of vehicle-mounted curved anti-glare glass according to the present invention.

[0020] Figure 3 This is a schematic block diagram of an electronic device for online detection of surface defects in vehicle-mounted curved anti-glare glass according to the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] like Figure 1 As shown, in one embodiment of the present invention, an online detection method for surface defects of automotive curved anti-glare glass includes the following steps: Step S10: Build a multi-view visual inspection platform. Based on the surface parameters of the vehicle-mounted curved anti-glare glass and the production line speed, configure inspection equipment and collect multi-dimensional images of the glass surface. Transmit the collected raw image data to the image processing unit.

[0023] Specifically, step S10 involves building a multi-view visual inspection platform, including: The system includes 3-5 sets of linear scan cameras, a ring diffuse light source, a polarizing filter, a laser ranging unit, and a synchronization control unit. The linear scan cameras are distributed along the normal direction of the glass surface and at ±30° and ±45° side views, with a pixel resolution of no less than 12 million. The acquisition frame rate is matched to the pipeline speed, for example, 5-20m / min corresponds to a frame rate of 500-1500fps. The ring diffuse light source uses a high-uniformity LED light source with a color temperature of 5500K±500K. The polarizing filter has a dynamically adjustable polarization angle with an adjustment range of 0° to 90° to eliminate the reflection interference of the anti-glare coating. The laser ranging unit acquires the distance data from the glass surface to the linear scan cameras in real time, providing a coordinate reference for surface distortion correction.

[0024] Specifically, step S10, which involves acquiring a multi-dimensional image of the glass surface, includes: The synchronous control unit triggers the linear array camera and the light source to work together based on the signal from the production line encoder. The laser ranging unit synchronously collects the object distance data corresponding to each pixel. Taking into account the coating characteristics of the anti-glare glass, the polarization filter angle is adjusted to suppress surface specular reflection. The angle is adjusted from 30° to 60°. Simultaneously, images in three modes—no polarization, low polarization, and high polarization—are collected to form a multimodal image set. The no polarization image retains the overall information of the defect, the low polarization image highlights linear defects such as shallow scratches, and the high polarization image enhances point defects such as bubbles and pits, ensuring the imaging clarity of different types of defects.

[0025] Step S20: The image processing unit preprocesses the original image data, sequentially performing surface distortion correction, image denoising enhancement, and region extraction, and outputs a standardized image.

[0026] Specifically, the preprocessing step of the image processing unit on the original image data in step S20 includes: Curved surface distortion correction: Based on the object distance data obtained by the laser ranging unit, a three-dimensional coordinate model of the glass curved surface is established. Perspective transformation combined with nonlinear interpolation algorithm is used to map the curved surface image to a two-dimensional plane to correct the image stretching and deformation caused by the curvature of the curved surface. Camera intrinsic and extrinsic parameters are calibrated by a calibration plate, such as a checkerboard calibration plate with a grid size of 10mm×10mm, to ensure that the dimensional accuracy error of the corrected image is ≤0.1mm. Image denoising and enhancement: For Gaussian noise and salt-and-pepper noise in the multimodal image set, adaptive median filtering is used to remove salt-and-pepper noise, with the window size dynamically adjusted from 3×3 to 7×7. Gaussian noise is suppressed by wavelet threshold denoising algorithm. The illumination component and reflectance component of the image are separated based on Retinex algorithm to enhance the gray-level contrast between the defect area and the background. Pixel-level fusion is performed on the images under three polarization modes. The weighted average method is used to improve the overall signal-to-noise ratio of the image. The weight of the unpolarized image is set to 0.4, and the weights of the low-polarized and high-polarized images are each 0.3. Region Extraction: Based on the contour features of the curved anti-glare glass in the vehicle, the glass region contour is extracted through edge detection algorithms, such as the Canny algorithm. The threshold is adaptively adjusted to remove background areas in the image, such as conveyor belts and fixtures. According to the size specifications of the glass, such as preset length and width thresholds, the effective detection area is cropped to reduce the amount of invalid data processing and improve detection efficiency.

[0027] Step S30: Based on the improved deep learning model, locate the candidate region of the defect in the standardized image, extract the geometric features, grayscale features and texture features of the defect, and form a defect feature vector.

[0028] Specifically, step S30, which involves locating defect candidate regions in the standardized image based on an improved deep learning model, includes: Defect candidate region localization: An improved YOLOv9 model is used to localize the defect candidate region. By adding an adaptive receptive field unit (ARF) to the neck area, the receptive field size is dynamically adjusted to adapt to the detection requirements of defects of different sizes. For example, the defect size is 10μm-5mm, defects ≤100μm correspond to a small receptive field, and defects >1mm correspond to a large receptive field. Multi-scale feature pyramids are constructed on the standardized images with feature map scales of 1 / 8, 1 / 16, and 1 / 32. The bounding boxes of defect samples are clustered using an anchor box clustering algorithm, and the anchor box parameters are optimized, such as K-means++, with a cluster number of 9, to improve the recall rate of defects ≤100μm. Defect Feature Extraction: For the located candidate regions, geometric features, grayscale features, and texture features are extracted. Geometric features include defect area, perimeter, roundness, and aspect ratio. Grayscale features include the average grayscale of the defect region, grayscale variance, and grayscale difference between the defect region and the background. Texture features are extracted based on the Gray-Level Co-occurrence Matrix (GLCM) using four parameters: energy, entropy, contrast, and correlation. The distance is 1, and the angles are 0°, 45°, 90°, and 135°. In addition, the texture characteristics of the anti-glare coating are combined to add coating consistency features. The texture similarity between the defect region and the surrounding normal coating is calculated, which is used to distinguish coating defects from surface contamination. Feature vector construction: All extracted features are normalized to the [0,1] interval using Min-Max normalization to eliminate dimensional differences. Redundant features are removed using variance screening, and features with variance ≥0.05 are retained to construct a defect feature vector with dimensions of 20-30.

[0029] The training process of the improved YOLOv9 model includes: constructing a defect dataset for automotive curved anti-glare glass, containing at least 100,000 labeled images, covering 6 types of defects and different lighting and angle scenes; expanding the dataset using methods such as Mosaic data augmentation, random flipping, and brightness and contrast adjustment; adopting an adaptive learning rate strategy during training, setting the initial learning rate to 0.01, and decaying it to 0.1 times the original rate every 10 epochs; using CIoU Loss combined with Focal Loss as the loss function to solve the problem of imbalanced defect samples; and achieving an average accuracy mAP ≥ 97.5% after model training.

[0030] Step S40: Analyze the defect feature vector using a multi-dimensional feature fusion algorithm, identify and classify the defect type by combining adaptive threshold judgment, generate detection results and feed them back to the production control system.

[0031] Specifically, step S40, which involves analyzing the defect feature vector using a multi-dimensional feature fusion algorithm and combining it with adaptive threshold determination to identify and classify defect types, includes the following steps: Multi-dimensional feature fusion: The AMFN attention mechanism is used to perform weighted fusion of defect feature vectors. The importance weight of each feature is calculated through the self-attention module to strengthen the focus on key features of defect identification, such as the aspect ratio of scratches, the roundness of bubbles, and the texture abrupt features of coating peeling. Defect type identification: The fused feature vector is input into the support vector machine (SVM) classifier. The preset defect types include six categories: scratches, bubbles, pits, coating peeling, dents, and contamination residue. The classifier parameters are optimized through cross-validation, with a penalty coefficient C=10 and an RBF kernel function to ensure a classification accuracy of no less than 98%. Defect severity classification: Based on the geometric characteristics of defects and usage scenario requirements, a severity classification standard is established, divided into four levels: A, B, C, and D. Level A represents no defects, and Level B represents minor defects with a defect area ≤ 0.1 mm. 2 Furthermore, the defect length must be ≤1mm; Grade C is a general defect, 0.1mm. 2 <Defect area ≤1mm 2 Or, if the defect length is 1mm or less and the defect area is ≤3mm, grade D is a severe defect, and the defect area is >1mm². 2 If the defect length is >3mm, an adaptive threshold algorithm is adopted to dynamically adjust the grade judgment threshold based on the inspection data of the glass batch, so as to adapt to the quality requirements of different production processes. Inspection result output: Generates an inspection report containing glass number, inspection time, defect type, defect level, defect coordinates and defect image. Simultaneously generates a defect visualization map, marking the defect location and range in the map. The inspection results are fed back to the production control system in real time. For defects of level C and above, an alarm mechanism is triggered. For defects of level D, the production line is stopped for timely handling.

[0032] In addition, such as Figure 2 As shown, in one embodiment of the present invention, an online detection system for surface defects of automotive curved anti-glare glass is proposed. The system includes: Multi-dimensional image acquisition module for automotive curved anti-glare glass surface: used to build a multi-view visual inspection platform. Based on the surface parameters of automotive curved anti-glare glass and the production line speed, it configures inspection equipment and acquires multi-dimensional images of the glass surface, and transmits the acquired raw image data to the image processing unit. Image preprocessing module: Used by the image processing unit to preprocess the raw image data, sequentially completing surface distortion correction, image denoising enhancement and region extraction, and outputting a standardized image; Vehicle-mounted curved anti-glare glass defect feature extraction module: used to locate defect candidate regions in standardized images based on an improved deep learning model, extract the geometric features, grayscale features and texture features of defects, and form a defect feature vector; Vehicle-mounted curved anti-glare glass defect identification and assessment module: It analyzes the defect feature vector through a multi-dimensional feature fusion algorithm, identifies and classifies the defect type by combining adaptive threshold judgment, generates detection results and feeds them back to the production control system.

[0033] This application provides an online detection system for surface defects of automotive curved anti-glare glass, employing an online detection method for surface defects of automotive curved anti-glare glass as described in the above embodiments. This system addresses the technical problems of poor surface adaptability, significant interference from the anti-glare coating, and high rates of missed and false detections of defects in existing detection methods. Compared with the prior art, the beneficial effects of the online detection system for surface defects of automotive curved anti-glare glass provided in this application are the same as those of the online detection method for surface defects of automotive curved anti-glare glass provided in the above embodiments. Furthermore, other technical features of the online detection system for surface defects of automotive curved anti-glare glass are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0034] This application provides an online detection device for surface defects of automotive curved anti-glare glass. The online detection device for surface defects of automotive curved anti-glare glass includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the online detection method for surface defects of automotive curved anti-glare glass as described in Embodiment 1 above.

[0035] like Figure 3 As shown in the illustration, in one embodiment of the present invention, a structural schematic diagram of an online detection device for surface defects of automotive curved anti-glare glass suitable for implementing the embodiments of this application is presented. The online detection device for surface defects of automotive curved anti-glare glass in this application embodiment may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital radio receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), etc., as well as fixed terminals such as digital TVs, desktop computers, etc. Figure 3 The online detection device for surface defects of vehicle-mounted curved anti-glare glass shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0036] Figure 3 The illustrated online inspection device for surface defects on automotive curved anti-glare glass may include a processor 1001 (e.g., a central processing unit, graphics processor, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage device 1003 into a machine-readable storage medium (RAM) 1004. The RAM 1004 also stores various programs and data required for the operation of the online inspection device for surface defects on automotive curved anti-glare glass. The processor 1001, ROM 1002, and RAM 1004 are interconnected via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to I / O interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication unit 1009. Communication unit 1009 allows an online inspection device for surface defects on automotive curved anti-glare glass to exchange data wirelessly or via wired communication with other devices. Although the figure shows an online inspection device for surface defects on automotive curved anti-glare glass with various systems, it should be understood that it is not required to implement or possess all the systems shown. More or fewer systems may be implemented alternatively.

[0037] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication system, or installed from storage device 1003, or installed from ROM 1002. When the computer program is executed by processor 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0038] This application provides an online detection device for surface defects of vehicle-mounted curved anti-glare glass, employing an online detection method for surface defects of vehicle-mounted curved anti-glare glass as described in the above embodiments. This method solves the technical problems of poor surface adaptability, significant interference from the anti-glare coating, and high rates of missed and false detections of defects in existing detection methods. Compared with the prior art, the beneficial effects of the online detection device for surface defects of vehicle-mounted curved anti-glare glass provided in this application are the same as those of the online detection method for surface defects of vehicle-mounted curved anti-glare glass provided in the above embodiments. Furthermore, other technical features of this online detection device for surface defects of vehicle-mounted curved anti-glare glass are the same as those disclosed in the previous embodiment method, and will not be repeated here.

[0039] The various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0040] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the above-described online detection method for surface defects of automotive curved anti-glare glass.

[0041] The computer program product provided in this application can solve the technical problems of poor surface adaptability, large interference from anti-glare coatings, and high rate of missed and false detections of defects in existing detection methods. Compared with the prior art, the beneficial effects of the computer program product provided in this application are the same as those of the online detection method for surface defects of vehicle-mounted curved anti-glare glass provided in the above embodiments, and will not be repeated here.

[0042] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. An online detection method for surface defects of vehicle-mounted curved anti-glare glass, characterized in that, The method includes the following steps: Step S10: Build a multi-view visual inspection platform. Based on the surface parameters of the vehicle-mounted curved anti-glare glass and the production line speed, configure inspection equipment and collect multi-dimensional images of the glass surface. Transmit the collected raw image data to the image processing unit. Step S20: The image processing unit preprocesses the original image data, sequentially performing surface distortion correction, image denoising enhancement, and region extraction, and outputs a standardized image; Step S30: Based on the improved deep learning model, locate the candidate region of the defect in the standardized image, extract the geometric features, grayscale features and texture features of the defect, and form a defect feature vector; Step S40: Analyze the defect feature vector using a multi-dimensional feature fusion algorithm, identify and classify the defect type by combining adaptive threshold judgment, generate detection results and feed them back to the production control system.

2. The online detection method for surface defects of vehicle-mounted curved anti-glare glass according to claim 1, characterized in that, The step S10, which involves building a multi-view visual inspection platform, includes: The system includes a linear array camera, a ring diffuse light source, a polarizing filter, a laser ranging unit, and a synchronization control unit. The linear array camera is distributed along the normal direction of the glass surface and in the ±30° and ±45° side view directions. The acquisition frame rate is matched with the pipeline speed. The ring diffuse light source uses an LED light source. The polarizing filter has a dynamically adjustable polarization angle. The laser ranging unit acquires the distance data from the glass surface to the linear array camera in real time, providing a coordinate reference for surface distortion correction.

3. The online detection method for surface defects of vehicle-mounted curved anti-glare glass according to claim 1, characterized in that, The step of acquiring a multi-dimensional image of the glass surface in step S10 includes: The synchronous control unit triggers the linear array camera and the light source to work together based on the signal from the production line encoder. The laser ranging unit synchronously collects the object distance data corresponding to each pixel. Based on the coating characteristics of the anti-glare glass, the surface specular reflection is suppressed by adjusting the angle of the polarization filter. At the same time, images under different polarization modes are collected to form a multimodal image set to ensure the imaging clarity of different types of defects.

4. The online detection method for surface defects of vehicle-mounted curved anti-glare glass according to claim 1, characterized in that, The step S20, in which the image processing unit preprocesses the original image data, includes: Curved surface distortion correction: Based on the object distance data obtained by the laser ranging unit, a three-dimensional coordinate model of the glass surface is established. Perspective transformation combined with nonlinear interpolation algorithm is used to map the curved surface image to a two-dimensional plane to correct the image stretching and deformation caused by the curvature of the curved surface. Camera intrinsic and extrinsic parameters are calibrated through a calibration plate. Image denoising and enhancement: For Gaussian noise and salt-and-pepper noise in multimodal image sets, adaptive median filtering is used to remove salt-and-pepper noise, wavelet threshold denoising algorithm is used to suppress Gaussian noise, Retinex algorithm is used to separate the illumination component and reflectance component of the image, the gray-level contrast between the defect area and the background is enhanced, pixel-level fusion is performed on images under different polarization modes, and weighted averaging method is used to improve the overall signal-to-noise ratio of the image. Region extraction: Based on the contour features of the curved anti-glare glass in the vehicle, the glass region contour is extracted through edge detection algorithm, the threshold is adaptively adjusted, the background region in the image is removed, and the detection region is cropped according to the size specifications of the glass.

5. The online detection method for surface defects of vehicle-mounted curved anti-glare glass according to claim 1, characterized in that, The step S30, which involves locating defect candidate regions in the standardized image based on an improved deep learning model, includes: Defect candidate region localization: An improved YOLOv9 model is used to localize the defect candidate region. By adding an adaptive receptive field unit (ARF) to the neck, it is adapted to the detection requirements of defects of different sizes. Multi-scale feature pyramids are constructed on standardized images, and anchor box parameters are optimized by anchor box clustering algorithm to improve the defect recall rate. Defect feature extraction: For the located candidate regions, extract geometric features, grayscale features and texture features, and combine the texture characteristics of the anti-glare coating to increase coating consistency features; Feature vector construction: All extracted features are normalized to the [0,1] interval using Min-Max normalization, and redundant features are removed using variance screening, retaining features with variance ≥0.05, and constructing a defect feature vector with dimensions of 20-30.

6. The online detection method for surface defects of vehicle-mounted curved anti-glare glass according to claim 5, characterized in that, The training process of the improved YOLOv9 model includes: constructing a dataset of defects in automotive curved anti-glare glass, expanding the dataset using Mosaic data augmentation, random flipping, and brightness / contrast adjustment, and adopting an adaptive learning rate strategy during training. The initial learning rate is set to 0.01, and it is reduced to 0.1 times the original rate every 10 epochs. The loss function is a combination of CIoU Loss and FocalLoss.

7. The online detection method for surface defects of vehicle-mounted curved anti-glare glass according to claim 1, characterized in that, The step S40, which involves analyzing the defect feature vector using a multi-dimensional feature fusion algorithm and combining it with adaptive threshold determination to identify and classify defect types, includes: Multi-dimensional feature fusion: The attention mechanism fusion network AMFN is used to weight and fuse the defect feature vectors, strengthening the focus on key features for defect identification; Defect type identification: The fused feature vector is input into the support vector machine (SVM) classifier. The preset defect types include six categories: scratches, bubbles, pits, coating peeling, dents, and contamination residue. The classifier parameters are optimized through cross-validation. Defect classification: Based on the geometric characteristics of defects and the requirements of usage scenarios, classification standards are formulated, and an adaptive threshold algorithm is adopted to dynamically adjust the classification threshold according to the inspection data of glass batches to adapt to the quality requirements of different production processes. Inspection result output: Generate an inspection report containing glass number, inspection time, defect type, defect level, defect coordinates and defect image, and simultaneously generate a defect visualization map, marking the defect location and range in the map, and feed the inspection results back to the production control system in real time.

8. An online detection system for surface defects of vehicle-mounted curved anti-glare glass, comprising the method described in any one of claims 1 to 7, characterized in that, include: Multi-dimensional image acquisition module for automotive curved anti-glare glass surface: used to build a multi-view visual inspection platform. Based on the surface parameters of automotive curved anti-glare glass and the production line speed, it configures inspection equipment and acquires multi-dimensional images of the glass surface, and transmits the acquired raw image data to the image processing unit. Image preprocessing module: Used by the image processing unit to preprocess the raw image data, sequentially completing surface distortion correction, image denoising enhancement and region extraction, and outputting a standardized image; The defect feature extraction module for automotive curved anti-glare glass is used to locate candidate defect regions in standardized images based on an improved deep learning model, and extract the geometric features, grayscale features, and texture features of defects to form a defect feature vector. Vehicle-mounted curved anti-glare glass defect identification and assessment module: It analyzes the defect feature vector through a multi-dimensional feature fusion algorithm, identifies and classifies the defect type by combining adaptive threshold judgment, generates detection results and feeds them back to the production control system.

9. An online inspection device for surface defects of vehicle-mounted curved anti-glare glass, characterized in that, include: The system includes a memory, a processor, and an online detection program for surface defects of automotive curved anti-glare glass stored in the memory and executable on the processor. When the processor executes the online detection program for surface defects of automotive curved anti-glare glass, it implements an online detection method for surface defects of automotive curved anti-glare glass as described in any one of claims 1 to 7.

10. A computer program product, characterized in that, The computer program product includes an online detection program for surface defects of vehicle-mounted curved anti-glare glass, which, when executed by a processor, implements an online detection method for surface defects of vehicle-mounted curved anti-glare glass as described in any one of claims 1 to 7.

Images