Transmittance detection method and system based on image detection, and storage medium

Abstract

This application discloses a transparency detection method, system, and storage medium based on image detection, relating to the field of machine vision optical image transmittance performance detection technology. It constructs an original dataset by acquiring image sequences from multiple incident angles, and reconstructs the basic geometric contour of the object by combining the light angle and surface normal vector. It analyzes the refraction distribution and delineates distortion regions through light propagation simulation, and performs image distortion correction based on pixel intensity deviation. The image is then optimized by adjusting the intensity correction coefficient, and the local transparency feature distribution is obtained by combining distortion residual evaluation. The weights of internal scattering influence factors are adaptively adjusted by fusing optical path and thickness data to generate the final transparency performance distribution map. Finally, an overall transparency performance vector is constructed, and distortion correction and optical parameters are optimized to achieve accurate quantitative evaluation of the transparency of non-planar transparent objects. This application effectively eliminates interference from curved surfaces, thickness, and optical distortion, achieving high detection accuracy, strong anti-interference capability, and precise quantification of the overall and local transparency performance of objects.

Description

Technical Field

[0001] This application belongs to the field of machine vision optical image transmittance performance detection technology, specifically a method, system and storage medium for transmittance detection based on image detection. Background Technology

[0002] In modern industry and technology, transparency testing of transparent objects is a crucial technology, playing an indispensable role, especially in product quality control and optical performance evaluation. This technology is widely used in industries such as glass manufacturing and lens production, directly impacting product safety and performance. Its research and application have profound significance for improving industrial production levels. However, despite progress in transparency testing of planar objects, many challenges remain when dealing with objects of complex shapes.

[0003] Current techniques often struggle to adapt to the interference caused by shape variations in non-planar transparent objects. Many existing solutions focus more on the direct imaging effect on the object's surface, neglecting the impact of shape differences on the light propagation path, leading to significant discrepancies between detection results and actual performance. This discrepancy not only affects the reliability of the detection but may also cause resource waste and quality risks during the production process.

[0004] In this field, the core technical challenge lies in the interference caused by variations in the geometry of an object's surface. In particular, the varying degrees of curvature on curved surfaces lead to complex refraction and distortion of light as it passes through, preventing the acquired images from accurately reflecting the object's transparency. Furthermore, these refractions and distortions are exacerbated by uneven thickness across different parts of the object, ultimately resulting in distorted detection data. For example, when inspecting a glass bottle, differences in curvature and thickness at different locations alter the light path, potentially causing some areas to be incorrectly identified as opaque or defective. This error directly impacts subsequent quality assessments.

[0005] Therefore, accurately reproducing the true path of light propagation when dealing with curved objects and eliminating the interference of shape differences on the test results has become a key problem that urgently needs to be solved in the field of transparency testing. Summary of the Invention

[0006] To address the above issues, this application provides an image-based transparency detection method, system, and storage medium. This addresses the problem that most existing traditional light transmittance detection technologies are designed for planar, regular light-transmitting objects, have simple detection methods, and can only achieve a rough calculation of the overall average light transmittance. They cannot meet the detection needs of curved, irregular, or non-planar light-transmitting objects with uneven thickness.

[0007] To achieve the above objectives, in a first aspect, this application provides a transparency detection method based on image detection, the method comprising the following steps: S1. Use an image acquisition device to acquire an image sequence of the object to be detected, and record the incident light angle distribution in layers to obtain a raw data set containing the shape changes and related parameters of the object to be detected; S2. Based on the original data set, a matching technique is used to combine the incident light angle distribution and the surface normal vector calculation to reconstruct the preliminary geometric contour of the object to be detected and obtain the basic surface structure data. S3. The light propagation process is simulated by light propagation simulation method. Combined with preset refractive index parameters and light path processing, the boundary reflection intensity and internal scattering influence factor are analyzed to obtain the refraction distribution and distortion region boundary division results corresponding to the light path. S4. Based on the refraction distribution and the boundary division results of the distortion region, extract the light intensity distribution mapping corresponding to each pixel, calculate the light intensity deviation of each pixel in combination with the preset deviation threshold, and when the light intensity deviation exceeds the preset range, use the distortion correction algorithm to eliminate the interference of the local distortion region and obtain the corrected clear image. S5. Based on the corrected clear image, the local transparency feature distribution of the object to be detected is determined by adjusting the light intensity correction coefficient and image optimization processing, and by combining the distortion residual error evaluation data analysis. S6. Based on the local transparency feature distribution, the light propagation distance and the surface thickness change data of the object to be detected are fused. When the light intensity correction coefficient of the uneven thickness area deviates from the standard value, the weight of the internal scattering influence factor is adaptively adjusted to generate the final transparency performance distribution map. S7. Extract the overall transparency vector of the object to be tested based on the final transparency performance distribution map, and achieve a quantitative assessment of the transparency of the object to be tested by optimizing distortion correction and optical calculation parameters.

[0008] Preferably, step S1 includes: Adjust the shooting focal length, exposure parameters, sampling frame rate and shooting position of the image acquisition device, fix the object to be detected at the preset detection position, and keep the detection environment lighting stable and free from stray light interference. The illumination sources are arranged in a layered gradient according to different incident light pitch angles and horizontal deflection angles, and the light incident angles are adjusted step by step while the image acquisition device is triggered to continuously sample. Collect a continuous frame image sequence of the object to be detected, and synchronously associate, mark and archive the incident light angle parameters and shooting coordinate parameters corresponding to each frame image; By filtering out blurry, overexposed, and underexposed frames, invalid image data is obtained, resulting in a set of valid original images and associated parameters that contain features of changes in the shape of the object to be detected, the refraction and shift of light on curved surfaces, and changes in light and shadow.

[0009] Preferably, step S2 includes: The valid image data and associated parameters in the original dataset are retrieved, and feature point matching and contour matching techniques are used to accurately match and align the edge features and surface feature points of the object to be detected in the image. Based on the effective image data after matching and alignment, and combined with the light incident angle parameter and the image pixel coordinate conversion relationship, the normal vector value of each position on the surface of the object to be detected is calculated point by point. By fitting normal vectors and using a 3D contour reconstruction algorithm, combined with feature point spatial coordinate conversion, the 3D geometric contour model of the object to be detected is initially reconstructed. After performing contour denoising, edge calibration, and surface fitting correction on the three-dimensional geometric contour, the core parameters of the surface curvature, contour boundary, and surface undulation of the object to be detected are extracted to obtain basic surface structure data.

[0010] Preferably, step S3 specifically includes: Based on the material properties of the object to be tested, preset the corresponding basic optical refractive index parameters, light absorption coefficient and interface reflection coefficient, and build a light propagation simulation environment for the object to be tested. Based on the basic surface structure data of the object to be detected, the light propagation simulation system is imported to simulate the entire light path of light rays passing through the object at different incident angles, including incident, refraction, reflection, and exit. Quantitatively calculate the light reflection intensity value at the boundary of the surface of the object to be detected, calibrate the degree of light scattering attenuation in different regions inside the object, and quantify the corresponding internal scattering influence factor. By combining the light path offset, the change in the light refraction angle, and the degree of light and shadow distortion, the distribution law of light path refraction across the entire region is statistically analyzed to define the critical boundary between the distorted area and the normal transparent area of ​​the image, and to obtain the boundary division result of the distorted area.

[0011] Preferably, step S4 includes: Based on the boundary division results of the distortion region, the acquired image is divided into a global pixel grid, and the real-time light intensity value corresponding to each pixel is extracted pixel by pixel to construct a global pixel light intensity distribution mapping map. Pre-calibrate the pixel light intensity reference value under standard transparent conditions, compare the real-time light intensity value of each pixel with the reference value, and calculate the real-time light intensity deviation of each pixel one by one. By comparing the light intensity deviation of each pixel with the preset deviation threshold, local distorted pixel areas with light intensity deviation exceeding the preset range are selected, and abnormal areas of the image are accurately located. The preset distortion correction algorithm is invoked to perform targeted correction and repair on the abnormal area, eliminating recognition interference caused by refraction distortion and ambient light and shadow, and outputting a clear image after correction.

[0012] Preferably, step S5 includes: Based on the corrected clear image, according to the mapping table between optical parameters and light intensity correction coefficients, the light intensity correction coefficients corresponding to the optical parameters are obtained, and the light intensity of all pixels in the image is adjusted for equalization based on the light intensity correction coefficients. The image after light intensity adjustment is optimized to eliminate slight noise and residual light and shadow defects, resulting in an optimized image. The distortion residual data of the corrected and optimized image is collected and compared with the standard distortion-free image sample to obtain the image residual distortion residual error assessment result. Based on the residual error assessment results, the light transmittance, light attenuation, and scattering loss characteristics of each local area are analyzed according to different curved areas and different thicknesses of the object to be tested, so as to determine the refined local transparency characteristics distribution of the object to be tested.

[0013] Preferably, step S6 includes: Simultaneously correlate local transparency feature distribution, actual light propagation distance parameters of light penetrating the object under test, and measured surface thickness distribution variation data of the object under test; Pre-set the standard light intensity correction coefficient reference values ​​corresponding to different thickness ranges, and compare the real-time light intensity correction coefficient of each thickness uneven area with the corresponding standard light intensity correction coefficient reference value to calculate the coefficient deviation difference of each thickness uneven area. Determine whether the real-time light intensity correction coefficient of each uneven thickness region deviates from the corresponding standard value. If there is a deviation, adjust the weight ratio of the scattering influence factor inside the corresponding detection area according to the magnitude of the coefficient deviation. Based on the weighted internal scattering influence factor, the transparency data of each region is recalculated, and the transparency data of each region is integrated and stitched together to generate the final transparency performance distribution map of the object to be detected.

[0014] Preferably, step S7 includes: Based on the final transparency performance distribution map, all feature parameters are integrated to construct an overall transparency performance vector that characterizes the overall light transmission properties of the object to be detected. Based on the current lighting conditions and transparency detection error feedback data, the distortion correction algorithm transformation parameters and optical calculation parameters related to optical refractive index and scattering attenuation are iteratively optimized to reduce residual image distortion error and optical calculation system deviation. The optimized distortion correction parameters and optical calculation parameters are input into the transparency evaluation calculation model. Combined with the overall transparency performance vector, a comprehensive weighted calculation is performed to complete the accurate quantitative evaluation of the transparency of the object under the current illumination conditions and output the quantitative detection results.

[0015] Secondly, this application provides a transparency detection system based on image detection, the system comprising: The acquisition module uses an image acquisition device to acquire image sequences of the object to be detected, and records the angle distribution of the incident light rays in layers to obtain a raw data set containing the shape changes and related parameters of the object to be detected. The reconstruction module, based on the original data set, uses matching technology combined with the incident light angle distribution and surface normal vector calculation to reconstruct the preliminary geometric contour of the object to be detected and obtain basic surface structure data. The module is divided into sections. The light propagation process is simulated by the light propagation simulation method. Combined with the preset refractive index parameters and light path processing, the boundary reflection intensity and internal scattering influence factor are analyzed to obtain the refraction distribution and distortion region boundary division results corresponding to the light path. The correction module extracts the light intensity distribution mapping corresponding to each pixel based on the refraction distribution and the boundary division results of the distortion region. It calculates the light intensity deviation of each pixel in combination with the preset deviation threshold. When the light intensity deviation exceeds the preset range, the distortion correction algorithm is used to eliminate the interference of the local distortion region and obtain a clear image after correction. The local transparency feature acquisition module, based on the corrected clear image, determines the distribution of local transparency features of the object to be detected by adjusting the light intensity correction coefficient and image optimization processing, and combining the distortion residual error evaluation data analysis. Transparency performance distribution map acquisition module. Based on the local transparency feature distribution, it integrates the light propagation distance and surface thickness variation data of the object to be detected. When the light intensity correction coefficient of the uneven thickness area deviates from the standard value, it adaptively adjusts the weight of the internal scattering influence factor to generate the final transparency performance distribution map. The evaluation module extracts the overall transparency vector of the object under test based on the final transparency performance distribution map, and achieves a precise quantitative evaluation of the transparency of the object under test by optimizing distortion correction and optical calculation parameters.

[0016] Thirdly, this application provides a computer-readable storage medium storing a transformer test strategy generation program, which, when executed by a processor, implements the image detection-based transparency detection method described above.

[0017] This application targets non-planar, curved, irregularly shaped, and unevenly thick translucent objects. It employs multi-angle, layered image sequence acquisition combined with geometric contour reconstruction to accurately establish the basic surface structure data of the object under inspection, effectively solving the problems of traditional detection methods that cannot adapt to curved structures, suffer from significant shape interference, and lack basic modeling. Through light propagation simulation combined with refraction distribution statistics and precise distortion region segmentation, it can accurately distinguish between normal translucent areas and optical distortion interference areas, pre-locating the sources of detection errors such as refraction shift and light and shadow distortion. By using pixel-level light intensity deviation judgment and targeted distortion correction algorithms, it effectively eliminates image interference caused by curved surface refraction, ambient light and shadow, and imaging distortion, significantly improving the image quality clarity and the authenticity of light intensity data. This application uses gradient light intensity correction coefficients for global light intensity equalization adjustment combined with distortion residual evaluation to accurately obtain refined local transparency feature distributions, overcoming the limitations of traditional techniques that can only detect overall average transparency and cannot distinguish local transparency differences. Simultaneously, this application combines data on light propagation distance and surface thickness variation, and adaptively adjusts the weight of the internal scattering influence factor based on the coefficient deviation magnitude. This effectively offsets the detection deviation caused by uneven object thickness and differences in internal scattering, improving the stability and fit of the transparency detection results. Finally, by constructing an overall transparency performance vector and iteratively optimizing distortion correction and optical calculation parameters, accurate quantitative evaluation of transparency is achieved. The overall detection exhibits strong anti-interference capability, high detection accuracy, and good versatility, meeting the practical engineering application needs for high-precision quality detection and performance grading of various non-planar transparent objects. Attached Figure Description

[0018] Figure 1 This is a flowchart of the image detection-based transparency detection method in the embodiments of this application. Detailed Implementation

[0019] To enable those skilled in the art to better understand the technical solution, the present application will be described in detail below with reference to the embodiments. The description in this section is only exemplary and explanatory, and should not be used to limit the scope of protection of the present application in any way.

[0020] This application provides a transparency detection method based on image detection, referring to... Figure 1 , Figure 1 This is a schematic flowchart of a transparency detection method based on image detection provided in an embodiment of this application. The method includes steps S1 to S7, as follows: S1. Use an image acquisition device to acquire an image sequence of the object to be detected, and record the incident light angle distribution in layers to obtain a raw data set containing the shape changes and related parameters of the object to be detected.

[0021] In this embodiment, an industrial high-definition color camera is selected as the image acquisition device. The camera's focal length, exposure gain, shutter speed, and sampling frame rate are pre-calibrated and adjusted to unify the camera's focus distance and shooting resolution. After calibration, the camera bracket is locked to fix the shooting position, shooting height, and shooting angle, ensuring no displacement deviation throughout the shooting process. The non-planar translucent object to be tested is stably placed and clamped onto a pre-set dedicated testing fixture. A limiting positioning structure restricts the object's translation and rotation, ensuring consistent posture for each sampling test. Simultaneously, excess ambient light sources are turned off to maintain a stable lighting environment at the testing station, free of light spots and stray light interference, providing a stable imaging foundation for subsequent multi-angle unified sampling. Multi-dimensional layered gradient lighting is used, with step-by-step angle adjustment and synchronous sampling. Multi-angle adjustable standard color temperature supplementary lighting sources are deployed around the testing station. Layered gradient lighting is arranged according to the pitch and horizontal deflection angles of the incident light, forming a multi-angle, multi-condition incident lighting system. During implementation, the incident angle of the light source is adjusted step-by-step and systematically at fixed angle intervals. After each adjustment is completed and the illumination stabilizes, the image acquisition equipment is immediately triggered to continuously sample in batches. This ensures that multiple consecutive images are acquired for each incident light angle condition, achieving full coverage of the curved surface of the object under test under different incident angles, and completely capturing the changes in light and shadow and refraction differences of the object's shape at different angles. Image sequence acquisition and parameter association are recorded and archived. During the continuous sampling and acquisition of a sequence of consecutive frame images of the object under test from multiple perspectives and illumination angles, each acquired image is bound and associated with a marker in real time. The real-time incident ray pitch angle parameters, horizontal deflection angle parameters, camera shooting coordinate parameters, and light source position parameters corresponding to that frame are synchronously recorded and archived one-to-one. This establishes a correlation between image pixel data and illumination geometry parameters and shooting position parameters, ensuring that subsequent contour reconstruction, normal calculation, and light propagation simulation can be traced and matched. Invalid image removal and original data set generation are also performed. All acquired image sequences undergo unified screening and filtering, automatically identifying and removing blurry frames with poor focus, overexposed frames with excessive brightness, underexposed frames with excessive brightness, and invalid image data containing noise or shadow interference. Valid original images that are clear, have complete textures, distinct edge contours, and accurately reflect the shape changes of the object under test, the light refraction shift caused by curved surface structures, and the characteristics of light and shadow changes at different angles are retained. All valid images and their corresponding incident light angles, shooting coordinates, and other related parameters are uniformly integrated and archived to construct a structured and standardized raw data set. This provides reliable and consistent raw input data for subsequent geometric contour reconstruction, light path simulation, distortion region segmentation, and accurate detection of transparency features.

[0022] S2. Based on the original dataset, a matching technique is used to combine the incident light angle distribution with the surface normal vector calculation to reconstruct the preliminary geometric contour of the object to be detected and obtain basic surface structure data. In this embodiment, the original data is retrieved, and feature points and contours are precisely matched and aligned. In specific implementation, the filtered valid original image data is directly called, and the incident light pitch angle, horizontal deflection angle, shooting coordinates, and other related parameters corresponding to each frame of the image are loaded synchronously. Feature point matching technology is used to extract stable surface feature points, edge contour corner points, and contour edge lines of the object to be detected. Frame-by-frame feature matching and contour alignment are performed on multiple images under different lighting angles and shooting conditions to correct image offset and contour misalignment caused by imaging at different lighting angles. This ensures that the object edge position and surface feature position are uniformly corresponding in all images, providing consistent image base data for subsequent normal calculation.

[0023] After image matching and alignment are completed, a precise mapping relationship between the two-dimensional coordinates of the pixel plane and the actual three-dimensional coordinates of the object is established based on the pre-fixed conversion relationship between image pixel coordinates and physical space coordinates obtained through camera calibration experiments. This eliminates scale conversion errors caused by shooting distance, lens distortion, and imaging angle, ensuring that the image pixel position accurately corresponds to the actual physical surface position of the object. Simultaneously, the incident light angle distribution parameters corresponding to each frame of the image are archived to clarify the specific optical incident conditions of the light's incident pitch angle and horizontal deflection angle during each frame's imaging. This determines the true incident orientation and incident angle reference of the light entering the surface of the object to be detected, providing an accurate optical geometric premise for subsequent normal numerical calculation. On this basis, the entire effective detection area of ​​the object to be detected is traversed row by row and column by column according to the pixel positions, fully covering all light-transmitting detection positions on the object's curved surface, without missing any thin-thickness transition areas, curved surface bending areas, and edge deformation areas. During the traversal, for each pixel's corresponding spatial position on the object's surface, the geometric equations are established point-by-point, combining the geometric relationship of light incidence, the laws of imaging projection, and the optical geometric constraints of the curved surface. This accurately solves for and outputs the specific values ​​of the 3D surface normal vector corresponding to each spatial position on the object's surface. The normal vector possesses a clear 3D spatial orientation attribute, capable of orienting the tilt orientation, slope angle, and curvature of each local location on the curved surface. Through the global distribution of the normal vector, the differences in surface orientation at various points on a non-planar translucent object surface are realistically reproduced, accurately reflecting the actual structural features of the object's surface, such as local undulations, curvature, surface gradient, and thickness transitions. For gently curving areas, the change in the normal vector direction is minimal and the transition is uniform; for curved areas with large curvature and obvious bends, the change in the normal vector direction is significant. This accurately distinguishes the structural differences between planar areas, gently curving areas, and areas with large curvature distortion, avoiding the problems of ambiguous surface feature descriptions and large deviations in normal estimation caused by the complexity of non-planar structures. This provides reliable basic data for subsequent accurate geometric contour reconstruction and optical refraction simulation calculations.

[0024] After accurately calculating the 3D surface normal vectors at the corresponding positions of all pixels in the global domain of the object to be detected, based on the acquired global surface normal vector data and combined with the spatial coordinate conversion relationships of various feature points locked after the previous image matching and alignment, the normal vector fitting and fusion and the preliminary reconstruction of the 3D geometric contour model are carried out. First, the dataset of 3D surface normal vectors corresponding to all pixel points obtained from the previous traversal calculation is integrated. Simultaneously, the original pixel coordinates of object edge feature points, surface curvature feature points, and contour corner points calibrated in the image matching process are extracted. Based on the preset camera intrinsic and extrinsic calibration parameters and the mapping relationship between pixels and physical space, the core operation of accurate conversion of feature point spatial coordinates is carried out. This conversion process uses the horizontal and vertical coordinates of 2D image pixels as basic input parameters. Combined with the camera focal length, principal point offset coefficient, and distortion correction coefficient, a pixel normalization conversion equation is constructed. The 2D pixel coordinates of the image are first converted into normalized planar coordinates in the camera coordinate system. Then, combined with the preset detection working distance and light incident depth compensation parameters, the 2D normalized coordinates in the camera coordinate system are further converted into 3D physical space coordinates in the world coordinate system. The actual spatial coordinates of each feature point in the real physical space X, Y, and Z axes are accurately obtained. This completely eliminates the coordinate conversion errors caused by image imaging perspective deviation, shooting position differences, and lens optical distortion, ensuring that the spatial coordinates of all feature points are completely matched with the actual physical placement and true shape of the object. After accurately converting the 3D spatial coordinates of all feature points, the global surface normal vector data and the 3D point cloud coordinate data of the feature points are bidirectionally fused and preprocessed. A normal vector fitting algorithm is used to smooth the normal vectors of adjacent regions and correct for continuity constraints. Local normal vector abrupt changes and calculation noise deviations are iteratively corrected to ensure that the transition of normal vector changes in continuous curved surface regions conforms to the actual surface deformation law of the object, avoiding the interference of single-pixel normal calculation errors on the overall contour reconstruction accuracy. Subsequently, the fitted and optimized normal vector dataset and the accurately converted 3D point cloud coordinate data of the feature points are simultaneously input into the preset 3D contour reconstruction algorithm. Through point cloud coordinate topological correlation operation combined with normal vector surface interpolation and completion calculation, coordinate interpolation and surface completion operations are performed on sparse point cloud regions, surface transition blind areas, and uneven thickness deformation regions. The correspondence between normal orientation and spatial coordinate position is matched point-to-point. All local surface structure data are stitched and fused region by region to initially construct a complete and continuous 3D geometric contour model of the object to be detected. This reconstruction model can completely restore the overall actual shape and size of the object under test, the curvature of the curved surface, the distribution of surface undulations, and the original structural features of thickness transition. It accurately reproduces the original physical structure of non-planar translucent objects, preserving the accuracy of the object's edge contour boundaries while fully restoring the subtle deformations and thickness differences of the curved surface. This provides a structurally accurate and data-complete initial 3D model foundation for subsequent 3D geometric contour noise reduction and correction, surface parameter extraction, and subsequent optical simulation of light propagation.

[0025] After the initial reconstruction of the 3D geometric contour model of the object to be detected, the resulting initial 3D model, although possessing the overall shape and basic surface form, still exhibits a small number of discrete noise points, local contour protrusions and depressions, abrupt changes in surface values, jagged edges, and minor calculation distortions due to pixel calculation discreteness, local errors in normal fitting, uneven feature point sampling density, minor residual deviations in coordinate conversion, and interpolation operations in the contour reconstruction algorithm. If directly used for subsequent light propagation simulation and optical refraction calculation, it is highly likely to cause offset of the light incident point, inaccurate normal matching, and deviations in refraction angle calculation, thereby leading to distortion in the light transmission distortion classification and transparency detection results. Therefore, it is necessary to sequentially perform multi-dimensional fine-tuning and optimization processing on the initially reconstructed 3D geometric contour model to ensure that the model's geometric shape is completely consistent with the actual physical structure of the object to be detected. First, contour denoising and noise removal are performed. Isolated noise points, abnormally shaped contour points deviating from the main surface, and invalid redundant points generated by reconstruction interpolation are discretely distributed in the 3D model's point cloud and surface mesh. Spatial neighborhood threshold filtering and connected component screening are used to batch-filter these out, retaining only the valid contour data connected to the object's true structure. This completely eliminates invalid interference points and abnormal data derived from the reconstruction calculation process. Next, precise edge contour calibration is performed. The outer contour edge curve of the 3D model is fitted and calibrated point-to-point with the actual edge contour boundary coordinates of the object in the original acquired image. This corrects issues such as edge jaggedness, contour offset, and boundary misalignment, ensuring that the model's outer contour boundary dimensions, edge positions, and contour closure shape accurately correspond to the actual shape boundary of the object under test, eliminating subsequent light incidence boundary judgment errors caused by edge contour deviations. Based on this, surface smoothing and fitting correction processing is carried out. Addressing defects such as abrupt changes in local surface undulations, unevenness, and harsh transitions between adjacent areas, a continuous surface fitting constraint algorithm is used to smooth and optimize the entire surface. While preserving the original surface curvature characteristics, thickness transition structures, and key deformation regions, it eliminates local numerical abrupt changes and irregular fluctuations, ensuring a continuous overall surface trend and natural curvature transitions, closely conforming to the physical morphology of a non-planar translucent object's surface. After completing all contour correction and surface optimization processing, the corrected 3D geometric contour model has accurate structure, regular surface, standard edges, and no computational noise or distortion defects. Based on this, core geometric structural parameters of the object are uniformly extracted in batches, mainly including the curvature distribution parameters of the entire surface of the object under test, the size parameters of the inner and outer contour boundaries, the surface undulation deformation distribution parameters, the structural parameters of the thickness transition regions, and the surface gradient change characteristic parameters. All extracted geometric parameters are standardized, categorized, packaged, and formatted to form structured, normalized basic surface structure data that can be directly interfaced with optical simulation systems.The basic surface structure data accurately reflects the true shape, surface state, curvature change and thickness distribution of the object under test. It can provide accurate, stable and reliable geometric model data support for subsequent light propagation simulation, light incident refraction path solution, boundary reflection intensity calculation, internal scattering factor quantification and accurate division of image distortion areas. It avoids the adverse effects of structural model errors on the accuracy of subsequent transparency optical detection from the geometric source.

[0026] S3. The light propagation process is simulated by light propagation simulation method. Combined with preset refractive index parameters and light path processing, the boundary reflection intensity and internal scattering influence factor are analyzed to obtain the refraction distribution and distortion region boundary division results corresponding to the light path. In this embodiment, the optical refractive index, light absorption coefficient, and interface reflectance coefficient are preset according to the material properties of the object to be tested, and a dedicated light propagation simulation environment for the object to be tested is built. In specific implementation, a transparent plastic workpiece with curved and irregular shapes and uneven thickness is used as the object to be tested. First, based on the actual factory optical parameters of the plastic material of the workpiece, dedicated optical basic configuration parameters are preset to accurately match the optical response characteristics of the real material. Among them, the fixed optical refractive index is preset to 1.45 according to the inherent light transmission properties of the plastic material, the light absorption coefficient is preset to 0.025 according to the light transmission loss characteristics of the material, and the interface reflectance coefficient is preset to 0.08 according to the optical reflection law of the interface between the object and the air. All optical parameters are fixed benchmark values ​​according to the material type to prevent the distortion of subsequent simulation calculations caused by the deviation of optical parameter settings. After the parameters are preset, a light propagation simulation environment is built that is completely consistent with the actual working conditions of the testing station. The standard refractive index of the simulated environment air is uniformly calibrated to 1.0 within the simulation system, and the ambient stray light scattering basis is set to zero to match the actual closed-off light-shielding testing environment. The hardware layout, such as the light incident angle range, light source placement, and image acquisition and receiving points, is replicated simultaneously in the actual testing. The physical refraction, reflection, and scattering calculation rules and optical boundary constraints of light propagation are uniformly solidified. A standardized light propagation simulation environment is constructed that is highly consistent with the on-site testing scenario, accurately matches optical parameters, and supports full-path tracking calculation of a single light ray. This lays a solid and accurate foundation for the subsequent full-process simulation of light penetrating objects at different angles.

[0027] The system imports basic surface structure data and simulates the entire ray path of light rays at different incident angles, including incident, refraction, reflection, and exit. Specifically, after the light propagation simulation environment is built, the corrected and optimized basic surface structure data of the object to be tested is completely imported into the simulation system. The system automatically and accurately reads the core geometric structure information of the workpiece, such as its three-dimensional geometric contour, surface curvature distribution, thickness transition area position, and surface boundary undulations. Within the simulation environment, it accurately replicates the three-dimensional model and placement posture of the workpiece that are consistent with the real object. Subsequently, strictly comparing the actual light pitch angle and horizontal deflection angle gradient conditions used in the image acquisition stage, each set of incident ray parameters is retrieved sequentially from smallest to largest angle to generate independent simulated incident rays, ensuring that no detection illumination angle condition is missed. For each simulated incident ray, the entire propagation process is simulated step-by-step, strictly following the physical laws of optical propagation: first, the ray is simulated to smoothly enter the workpiece surface from the air medium to the designated incident point; then, the ray is simulated to penetrate the surface and enter the workpiece, undergoing refraction and deflection; next, the ray is simulated to scatter and attenuate within the workpiece at different thicknesses, as well as to reflect and return through internal interfaces; finally, the ray is simulated to penetrate the workpiece and exit back into the air medium. During the simulation of each ray, the incident coordinates, refraction angle, reflection point, exit position, and complete propagation trajectory are recorded in real time. All ray path data are archived one by one, accurately reproducing the real optical propagation characteristics of non-planar curved workpieces, such as ray deviation, refraction differences, and scattering intensity changes caused by uneven thickness and curvature.

[0028] The intensity of light reflection at the surface boundary is quantitatively calculated, the degree of internal scattering attenuation is calibrated, and the corresponding internal scattering influence factor is quantified. Specifically, after the full-angle light path simulation is completed, for all incident and exit boundary regions on the outer surface of the workpiece to be tested, the intensity of boundary light reflection at different curved surface positions and different thickness transition positions on the workpiece surface is quantitatively calculated point by point, based on the preset interface reflection coefficient and the initial incident energy value of each light ray. The difference in reflection intensity between high curvature surface regions, smooth curved surface regions, and regions with abrupt changes in thickness is statistically analyzed to accurately quantify the differences in light energy loss caused by surface interface reflection. Combining the actual propagation path length of each light ray inside the workpiece and the preset material light absorption coefficient, the degree of light scattering attenuation at different positions in the thin, medium, and thick areas of the workpiece is calibrated by region. The greater the workpiece thickness and the longer the light propagation path, the more severe the scattering attenuation; the smaller the workpiece thickness and the shorter the propagation path, the weaker the scattering attenuation. The measured data of scattering attenuation in all regions are normalized and quantified. The differences in scattering intensity in different regions are uniformly converted into standardized internal scattering influence factors. The influence factor value is increased for thicker regions with stronger scattering attenuation and decreased for thinner regions with weaker scattering attenuation. This achieves zonal quantification and numerical characterization of the degree of scattering interference within the workpiece, providing dedicated scattering basic parameters for subsequent light intensity correction compensation and accurate transparency calculation.

[0029] By combining light path offset, light refraction angle change, and light and shadow distortion degree, the refraction distribution pattern is statistically analyzed to define the critical boundary between the distorted and normal light-transmitting areas, and the distortion area division results are obtained. Specifically, after completing the light simulation and scattering factor quantification, three core distortion quantification indicators corresponding to all simulated light rays are uniformly extracted: the light path offset between the actual propagation path and the standard ideal straight path of each light ray, the difference in refraction angle change before and after the light ray incident, and the degree of light and shadow distortion caused by the thickness difference of the curved surface. Based on the path offset data, refraction angle change data, and light and shadow distortion data of all light rays in the entire area, the global refraction distribution pattern of the light path in the entire workpiece inspection area is statistically analyzed to clarify the distribution characteristics of optical distortion intensity at different curved surface positions and different thickness positions. By setting a predefined critical threshold for distortion judgment, curved areas with large path offsets, excessive refraction angle changes, and high degrees of light and shadow distortion, as well as areas with abrupt changes in thickness, are defined as optical distortion interference areas. Smooth curved areas with small path offsets, stable refraction angles, and uniform light and shadow without obvious distortion are defined as normal light transmission areas. This precise definition of the clear critical boundary between the distorted area and the normal light transmission area completes the accurate division of the entire region. Finally, it outputs accurate distorted area boundary division results, providing a precise region positioning basis for subsequent image pixel light intensity deviation correction and accurate elimination of local distortion.

[0030] S4. Based on the refraction distribution and distortion region boundary division results, extract the light intensity distribution map corresponding to each pixel, calculate the light intensity deviation of each pixel in combination with the preset deviation threshold, and when the light intensity deviation exceeds the preset range, use the distortion correction algorithm to eliminate the interference of local distortion regions and obtain a corrected clear image; In this embodiment, based on the distortion region boundary division results, the entire image is divided into pixel grids, and real-time light intensity values ​​are extracted pixel by pixel to construct a global pixel light intensity distribution mapping map. This step is performed after the distortion region boundary division is completed. Using non-planar, unevenly thick translucent components as the objects to be detected, the coordinates of the determined critical boundary between the distortion region and the normal translucent region, as well as the region contour data, are directly loaded. This serves as the basis for region division, avoiding accidental damage to the normal translucent imaging region during the correction process. Specifically, the original detection image of the object to be detected is processed into a full-coverage pixel grid according to the inherent row and column structure of the camera's imaging pixels. Each image pixel is used as the smallest grid unit, and the grid is strictly divided row by row and column by column according to the pixel coordinate order. This covers the effective translucent area of ​​the object to be detected, the curved edge bending area, the area of ​​sudden thickness transition, and the area surrounding the distortion critical boundary, achieving a complete and thorough grid division without blind spots or omissions. After the grid is split, relying on the image grayscale analysis algorithm, the real-time imaging grayscale light intensity value of each pixel corresponding to each grid is read one by one. The original real-time light intensity data corresponding to different curved surface positions, different thickness positions, and different distortion degrees are accurately collected. The binding relationship between the two-dimensional coordinate information of each pixel and the corresponding light intensity value is recorded simultaneously. All pixel coordinate data and light intensity measurement data are uniformly collected and organized, and the data is arranged and visualized according to the topological relationship of pixel spatial position. A full-domain pixel light intensity distribution mapping map covering the detection range of the object to be detected is constructed, which intuitively presents the differences in light intensity and darkness at various parts of the object, the fluctuation of light and shadow on curved surfaces, and the abnormal light intensity shift characteristics in distorted areas. This provides standardized and structured original light intensity data support for subsequent pixel-level accurate deviation calculation.

[0031] A pre-calibrated baseline value for the light intensity of a standard transparent pixel is established. Real-time light intensity is compared pixel-by-pixel with this baseline value, and the real-time light intensity deviation is calculated pixel-by-pixel. After the global pixel light intensity distribution mapping map is constructed, calibration is performed under standardized conditions—the same testing station, the same incident light angle, and the same camera exposure and acquisition parameters—using a standard calibration sample that is the same material as the object being tested, has uniform thickness, a smooth surface without curvature or optical refraction distortion. After acquiring the test image of the standard calibration sample, the light intensity values ​​of all pixels across the entire region are extracted and averaged. The stable, uniform, distortion-free, and scattering-free average light intensity is solidified as the pixel light intensity baseline value under standard transparent conditions, serving as a unified reference scale for ideal, error-free light transmission detection. After calibration, the real-time light intensity value of each pixel in the image of the object to be detected is compared point-to-point with the preset standard light intensity reference value. According to the light intensity deviation calculation rules, the difference between the measured light intensity value and the reference value at each point is calculated pixel by pixel. The positive deviation of high light intensity and the negative deviation of low light intensity are counted separately. The amount of light intensity distortion deviation caused by surface refraction, uneven thickness, internal scattering, and light and shadow interference for each pixel is quantified. Among them, the deviation value of areas with large surface curvature and obvious thickness changes is significantly larger, while the deviation value of flat and thin areas is smaller. Through precise point-to-point calculation, the degree of light intensity distortion of each pixel can be quantified, compared, and traced, providing an accurate deviation data basis for subsequent screening and judgment of distortion and abnormal areas.

[0032] The light intensity deviation is compared with a preset deviation threshold to filter out pixels with excessive distortion, accurately locating abnormal distortion areas in the image. After calculating the light intensity deviation pixel by pixel, a light intensity deviation judgment threshold is pre-calibrated and solidified based on the product's detection accuracy level and the allowable fluctuation range of optical distortion. This threshold serves as the critical judgment standard for distinguishing between normally transparent pixels and abnormally distorted pixels, effectively avoiding misjudgments and omissions caused by normal minor light and shadow fluctuations. The absolute value of the light intensity deviation of each pixel is compared with the preset deviation threshold. Pixels with a light intensity deviation less than the threshold are judged as normally transparent pixels and retain their original imaging data without correction. Pixels with a light intensity deviation exceeding the preset threshold are all marked as abnormally distorted pixels. Regional clustering and fusion processing is performed on discretely distributed abnormal pixels to merge adjacent and contiguous abnormal pixels into a whole local distortion region. The outline range, location, and area affected by the distortion region are precisely delineated. Various imaging anomalies caused by light refraction shift, surface shape changes, uneven thickness scattering, and residual light and shadow interference from the environment are accurately located. This achieves targeted and precise localization of the distortion region, ensuring that subsequent correction only operates on the problem area and does not affect the true imaging characteristics of the normally transparent area.

[0033] The algorithm invokes a preset distortion correction algorithm to specifically correct and repair abnormal areas, eliminating interference and outputting a clear image after correction. After accurately locating the localized abnormal pixel areas, it calls a dedicated distortion correction algorithm adapted for detecting objects with curved surfaces. Combining the previous refraction distribution pattern, distortion boundary position, and scattering influence factor parameters, it performs targeted correction and repair only on the marked abnormal distortion areas, while the original image data of normally transparent areas is preserved without modification. During the correction process, the algorithm performs reverse compensation correction, light and shadow transition smoothing, and inverse refraction distortion reduction operations on the light intensity of distorted areas based on the magnitude of light intensity deviation of each pixel and the relationship between light path offset and refraction angle change. This counteracts imaging and recognition interference caused by curved surface refraction deformation, uneven thickness scattering attenuation, and residual weak ambient light. After correction, the image seam edges are optimized for smooth transition, eliminating correction stitching marks and pixel abrupt color differences, ensuring natural transitions and consistent light intensity across the entire image. Finally, all local distortion interference and imaging distortion problems are removed, and a clear image with clear imaging, true light intensity, accurate surface restoration, and no distortion or noise is output after correction. This provides a high-quality image foundation for subsequent local transparency feature extraction and accurate transparency quantification.

[0034] S5. Based on the corrected clear image, the local transparency feature distribution of the object to be detected is determined by adjusting the light intensity correction coefficient and image optimization processing, and by combining the distortion residual error evaluation data analysis. In this embodiment, based on the corrected clear image, the corresponding coefficients are matched according to the mapping table of optical parameters and light intensity correction coefficients to complete the equalization adjustment of the light intensity of pixels across the entire image. This step is carried out after the output of the distortion-free, high-quality corrected clear image. The test object is still a translucent plastic workpiece with a curved surface and uneven thickness distribution. In the early stage, through a large number of benchmark test experiments, combined with the refractive index of the workpiece material, the interface reflection coefficient, the basic parameters of internal scattering, the optical loss law of different surface curvature conditions and different thickness ranges, a standard mapping table with one-to-one correspondence between exclusive optical parameters and light intensity correction coefficients was established. The mapping table is classified and calibrated according to the areas of smooth surface, curved surface, thinner thickness, and thicker thickness, and the exclusive light intensity correction coefficients corresponding to different working conditions are calibrated to achieve accurate coefficient matching for different optical loss scenarios and avoid the problem of local transparency distortion caused by uniform coefficient adjustment. During implementation, the rectified clear image processed by S4 is directly retrieved, and the measured optical parameters corresponding to the detection conditions of the image are synchronously associated, including real-time illumination incident angle parameters, workpiece surface curvature parameters, local thickness parameters, and light scattering parameters. Based on a preset standard mapping table, the system automatically retrieves and matches the exclusive light intensity correction coefficients that precisely correspond to the current optical parameters, avoiding the mixing of correction parameters from different conditions to ensure that the coefficients match the actual detection conditions. Subsequently, using the matched light intensity correction coefficients as the adjustment benchmark, a global light intensity equalization adjustment operation is performed on all pixels in the rectified clear image. For areas with weak scattered light intensity in thick regions, the pixel light intensity values ​​are appropriately increased by the coefficients; for areas with strong transmitted light intensity in thin regions, the pixel light intensity values ​​are appropriately decreased by the coefficients; and for curved surface transition areas, a smooth transition adjustment is performed using a gradient coefficient. This unifies and balances the uneven light intensity caused by thickness differences and residual surface refraction in different areas of the entire image, ensuring that the light intensity performance in each area of ​​the image only retains the true light transmission differences of the object itself, eliminating systematic light intensity deviations caused by optical structures, and completing the global pixel light intensity standardization and equalization adjustment process.

[0035] After adjusting the light intensity, the image is optimized to eliminate slight noise and residual light and shadow defects, resulting in an optimized image. After the global pixel light intensity equalization adjustment is completed, the overall light intensity distribution of the image is uniform and there are no obvious distortion areas. However, due to the noise of photosensitive particles in the image acquisition device itself, the slight scattering fluctuation of light, and the small calculation error of the correction operation, there are still a few discrete slight noise, slight color difference of pixels, and weak light and shadow transition defects at the corners. If it is directly used for transparency feature analysis, it is easy to cause slight fluctuations in local transparency data and a decrease in feature extraction accuracy. Therefore, the image after light intensity adjustment needs to undergo specialized and refined optimization for noise reduction and defect repair. Specifically, an adaptive neighborhood smoothing optimization algorithm is employed. While preserving the core effective features of the workpiece edge contour details, curved surface translucency texture features, and differences in transparency between thick and thin regions, discrete salt-and-pepper noise and slight Gaussian noise within the image are smoothed and filtered out. For residual weak light and shadow at the workpiece edges and the transition points between thick and thin surfaces, as well as minor defects in light and shadow transitions, local light and shadow smoothing and fusion repair processing is performed to eliminate abrupt local light and shadow, pixel gaps, and abrupt changes in color and brightness. During the optimization process, the optimization intensity is strictly controlled, removing only invalid noise and redundant light and shadow defects without altering the core translucency features and the effective light intensity differences corresponding to the curved surface structure, avoiding over-optimization that could lead to distortion by smoothing out local transparency features. After processing, an optimized image with clean image quality, eliminated noise, natural light and shadow transitions, complete detail preservation, and no redundant residual defects is obtained, providing a high-quality image sample basis for subsequent accurate comparison and evaluation of distortion residuals.

[0036] After collecting and optimizing the image distortion residual data, it is compared with a standard distortion-free sample image to generate residual error evaluation results. Once the optimized image is prepared, distortion residual data collection and error comparison evaluation are performed. First, a pre-saved standard distortion-free image sample is retrieved. This sample is prepared using qualified calibration workpieces of the same material, specifications, uniform thickness, and standard surface deformation-free characteristics, under identical lighting conditions, equipment acquisition parameters, and correction processing procedures. The image has no refractive distortion, lighting defects, or calculated residuals, serving as the standard reference for residual comparison. Then, the actual pixel grayscale data, light intensity distribution data, and surface edge contour coordinate data of the optimized image under test are collected pixel by pixel, forming the original dataset of distortion residuals for the image under test. This original dataset is then compared pixel by pixel with the standard reference data of the standard distortion-free image sample. The difference in light intensity values, contour coordinate deviations, and grayscale distribution offsets are statistically analyzed at each point. The overall residual mean, local residual fluctuation amplitude, and maximum residual deviation peak value are calculated across the entire domain. Based on the magnitude of the global residual distribution, the location of the residual concentration area, and the degree of residual fluctuation, the degree of residual distortion and the level of residual error correction in the image are comprehensively evaluated. Finally, a quantitative, regional, and traceable image residual distortion residual error assessment result is generated, which clarifies which areas have minimal residuals and have fully met the standards, which areas have slight residual errors, and the specific numerical value of the error deviation. This provides a basis for residual error correction for subsequent accurate analysis of transparency features in different regions.

[0037] Based on the residual evaluation results, the characteristics of light transmission, attenuation, and scattering loss are analyzed according to the surface and thickness regions. This determines the refined distribution of local transparency features. After obtaining the residual distortion error evaluation results, instead of a uniform and coarse judgment of transparency performance across the entire area, a refined and differentiated analysis is performed for each region. According to the differences in the surface structure and thickness distribution of the object under inspection, the overall inspection area of ​​the workpiece is divided into several independent sub-regions: a gently curving thin area, a conventional medium-thick area, a curved thick area, and a transitional area with abrupt changes in thickness. Based on the residual error evaluation values ​​corresponding to each sub-region, the original light intensity data of each region is corrected by residual inverse compensation to offset the calculation deviation of transparency data caused by residual minor distortion residuals. After correction, the intuitive characteristics of light transmission intensity, the degree of attenuation loss during light transmission, and the variation law of the magnitude of scattering loss within the object are analyzed region by region. This accurately distinguishes the differentiated transparency performance: thick areas have large scattering loss and weak light transmission performance; thin areas have small scattering loss and strong light transmission performance; gently curving areas have uniform light transmission; and curved areas have significant light transmission fluctuations. By integrating residual correction data, light intensity equalization data, scattering attenuation characteristics, and light transmission intensity differences from various regions, normalization and quantitative processing are performed to accurately characterize the true light transmission characteristics of each local location of the object under test. This results in a refined local transparency feature distribution that is fully covered, highly accurate, zone-differentiated, and precisely reflects the influence of surface changes and thickness fluctuations. This provides accurate and reliable original local transparency feature data support for subsequent adaptive adjustment of scattering factor weights and generation of the final transparency performance distribution map.

[0038] S6. Based on the local transparency feature distribution, the light propagation distance and surface thickness variation data of the object to be detected are fused. When the light intensity correction coefficient of the uneven thickness region deviates from the standard value, the weight of the internal scattering influence factor is adaptively adjusted to generate the final transparency performance distribution map; In this embodiment, the local transparency feature distribution, actual light propagation distance parameters, and measured surface thickness distribution variation data of the object are synchronously correlated. This step follows the refined local transparency feature distribution obtained in S5 to perform synchronous data correlation and fusion processing. The test object is uniformly a translucent plastic workpiece with curved and irregular shapes and uneven thickness transitions. In specific implementation, the global refined local transparency feature distribution data parsed in S5 is first fully retrieved to clarify the basic transparency feature data of light transmission intensity, light attenuation degree, and scattering loss corresponding to each local curved area and different thickness zones of the test object, forming a global transparency feature base layer. The actual light propagation distance parameters of light penetrating the test object recorded by ray tracing during the light propagation simulation in S3 are retrieved simultaneously. The light penetration path length is different in different thickness areas, with thicker areas having longer light propagation distances and thinner areas having shorter light propagation distances. All propagation distance data are matched and archived one-to-one with the image pixel positions. Simultaneously, the measured surface thickness distribution data of the object under test, obtained from previous geometric contour reconstruction and thickness detection, are retrieved to accurately label the actual thickness distribution of thin areas, normal thickness areas, thicker areas, and abrupt thickness transition areas at various locations on the workpiece. During implementation, the local transparency feature distribution data, the actual propagation distance parameters of the light path, and the measured thickness distribution data of the object surface are precisely and synchronously associated and bound according to the same pixel spatial coordinates and the same local detection area. This achieves one-to-one matching and fusion of three core data types—transparency performance, optical path length, and solid thickness—at the same location, avoiding data misalignment and region mismatch that could lead to distortion in subsequent weight adjustments. This provides multi-source integrated basic data support for subsequent thickness partition coefficient comparison, deviation difference calculation, and adaptive weight adjustment of scattering factors.

[0039] Pre-set standard light intensity correction coefficient benchmark values ​​for different thickness ranges, compare real-time and standard coefficients zone by zone, calculate the coefficient deviation difference for each thickness uneven area, and after data association, pre-divide multiple continuous thickness ranges according to the thickness variation gradient of the workpiece to be tested. For each thickness range, under the standard calibration conditions of no distortion, uniform light transmission, and no scattering interference, measure and calibrate the corresponding standard light intensity correction coefficient benchmark value, and establish a benchmark comparison table that corresponds one-to-one between different thickness ranges and standard light intensity correction coefficient benchmark values. The larger the thickness, the higher the corresponding standard benchmark coefficient, and the smaller the thickness, the lower the corresponding standard benchmark coefficient, ensuring that each thickness range has a dedicated matching standard reference benchmark, realizing zoned calibration and accurate reference. After calibration, according to the actual thickness distribution area of ​​the workpiece surface, the real-time light intensity correction coefficient under the current working condition is extracted for each uneven thickness transition area and curved surface thickness variation area. The real-time light intensity correction coefficient measured in each local area is compared and calculated point-to-point with the standard light intensity correction coefficient benchmark value corresponding to the thickness range of that area. The positive or negative deviation of the real-time light intensity coefficient of each uneven thickness area relative to the standard benchmark coefficient is counted one by one. The coefficient deviation difference corresponding to each local uneven thickness area is accurately calculated, quantifying the degree of light intensity coefficient deviation caused by thickness fluctuation and curved surface scattering differences in different areas, clarifying the magnitude and direction of deviation in each area, and providing a quantitative deviation basis for subsequent adaptive weight adjustment.

[0040] The system determines whether the coefficient deviates from the standard value. Based on the magnitude of the deviation, it adaptively adjusts the weight ratio of the internal scattering influence factor in the corresponding region. After calculating the deviation difference of the coefficients in each region, it checks whether the real-time light intensity correction coefficient in each region with uneven thickness deviates from the standard reference value of the corresponding thickness range. If the coefficients are consistent and there is no deviation, the original weight of the internal scattering influence factor in that region remains unchanged and no adjustment is needed. If the coefficients deviate positively or negatively, an adaptive weight adjustment mechanism is activated, strictly matching the weight adjustment magnitude proportionally to the actual deviation of the coefficients. The specific adjustment logic is as follows: when the light intensity correction coefficient in a locally thick region deviates negatively, the light intensity is too dim, and the internal scattering attenuation is too large, the weight ratio of the internal scattering influence factor in that detection region is adaptively increased to strengthen the scattering loss correction compensation. When the light intensity correction coefficient in a locally thin region deviates positively, the light intensity is too bright, and the internal scattering attenuation is too small, the weight ratio of the internal scattering influence factor in that detection region is adaptively decreased to weaken the influence of over-correction of scattering. The greater the deviation, the greater the weight adjustment range; the smaller the deviation, the smaller the weight fine-tuning range. This ensures that the weight is adjusted to compensate for the deviation, achieving differentiated and adaptive precise adjustment of the weights of scattering influence factors for different thicknesses and curved surfaces. This also eliminates the problems of inaccurate calculation and imbalanced compensation in thick and thin regions caused by uniform fixed weights.

[0041] Based on the recalculation of scattering factors after weight adjustment, the transparency data is integrated and stitched together to generate the final transparency performance distribution map. After adaptive adjustment of the weights of scattering influence factors within all local detection areas, the latest adjusted and optimized scattering influence factor weights are used as the core calculation parameters. Combined with the local transparency characteristics, light propagation distance, and thickness distribution data of each region, the real-time transparency quantification data of each curved surface partition and each thickness transition partition is recalculated accurately for each region. The transparency calculation values ​​corresponding to the changes in scattering weights in each region are updated, correcting the transparency calculation deviations caused by the original fixed weights. After the transparency data of all regions is recalculated, the transparency quantification data of each independent local region is integrated, smoothly connected, and seamlessly stitched. The transparency values ​​at the boundary positions of regions are optimized for gradual and smooth transition, eliminating the data stitching discontinuities and abrupt changes in boundary values ​​caused by partition calculation, ensuring that the overall transparency data is continuous, the transition is natural, and the distribution pattern closely matches the actual light transmission performance of the object. After the overall integration and optimization is completed, the overall transparency quantification value is visualized and mapped into a map according to the pixel spatial distribution law and the color gradient mapping relationship. Finally, a final transparency performance distribution map of the object under test is generated, which fully covers the entire range of the object under test, accurately reflects the thickness difference and the influence of surface scattering, and provides accurate and intuitive visualization of the data. This provides the final spectral data foundation for subsequent overall transparency performance vector extraction and accurate quantitative evaluation.

[0042] S7. Extract the overall transparency vector of the object to be tested based on the final transparency performance distribution map, and achieve a quantitative assessment of the precise transparency of the object to be tested by optimizing distortion correction and optical calculation parameters; In this embodiment, based on the final transparency performance distribution map, various transparency feature parameters are integrated to construct an overall transparency performance vector characterizing the overall light transmission characteristics of the object under test. This step follows the implementation of the final transparency performance distribution map of the object under test generated in S6. The test object is still uniformly a translucent plastic workpiece with a curved surface and uneven thickness distribution. In the specific implementation process, the final transparency performance distribution map of the whole domain processed in S6 is first fully retrieved. This distribution map has accurately integrated the quantitative distribution data of the whole domain transparency after the differences in the curved surface structure of the object, the changes in thickness, the length of the light propagation path, the adaptive weight adjustment of internal scattering, and the distortion correction compensation, completely preserving the true light transmission performance characteristics of the object's local and overall characteristics. Subsequently, core key transparency feature parameters are extracted in batches from the final transparency performance distribution map. These include the global average transparency quantification value of the object under test, peak transparency values ​​in different curved areas, average transparency attenuation in thick and thin zones, transparency fluctuation coefficient at curved edges, comprehensive quantification value of internal scattering loss, and transparency stability parameters after distortion correction. This covers multiple dimensions of core transparency indicators such as overall transparency level, local transparency extremes, regional transparency fluctuations, and scattering attenuation loss. All extracted transparency feature parameters are then uniformly integrated, normalized, quantified, and grouped into ordered vectors according to fixed parameter sorting rules and standardized data formats. This structured combination forms a unique overall transparency performance vector. This vector is a multi-dimensional digital feature vector that can comprehensively, three-dimensionally, and numerically accurately characterize the overall transparency characteristics of the object under test. It includes both the overall average transparency level and the transparency fluctuation differences caused by local curved surfaces and thicknesses, overcoming the limitations of traditional single transmittance representation methods and providing a standardized, multi-dimensional, and comprehensive vector data foundation for subsequent accurate quantitative weighted evaluation.

[0043] Based on the current lighting conditions and detection error feedback data, the core parameters of distortion correction and optical calculation are iteratively optimized to reduce residual errors and system calculation deviations. After the overall transparency performance vector is constructed, iterative optimization and calibration of distortion correction parameters and optical calculation parameters are carried out. First, real-time data of the actual lighting conditions at the testing site are collected and recorded, including the incident angle of light, illuminance intensity, light source color temperature parameters, and the interference coefficient of weak ambient stray light. Simultaneously, distortion residual error data, light intensity correction deviation data, and optical calculation fitting error feedback data retained at each step of the entire testing process are retrieved to form a dedicated error feedback dataset for this test, accurately reflecting the residual image distortion error and inherent deviation of the optical calculation system under the current conditions. Relying on the real-time lighting conditions data and detection error feedback data, an iterative optimization and correction mechanism for parameters is initiated. Iterative fine-tuning and optimization are performed on the core transformation coefficients and image distortion compensation correction parameters within the distortion correction algorithm. Simultaneously, adaptive iterative calibration and correction are performed on the optical refractive index reference parameters, scattering attenuation calculation coefficients, and internal scattering influence factor basic optical calculation parameters. During the optimization process, the goal is to minimize the residual error. The parameter values ​​are continuously and iteratively fine-tuned to gradually offset the image error caused by the residual distortion of curved surface imaging and the system deviation caused by optical modeling calculation. This continues until the residual distortion error and the system deviation of optical calculation are reduced to within the preset allowable accuracy range. This ensures that the parameters used in subsequent evaluation calculations are highly matched with the current actual lighting conditions and the true optical characteristics of the object. This reduces the evaluation calculation error from the source of the parameters and improves the accuracy and stability of quantitative transparency evaluation.

[0044] The optimized parameters are input into the evaluation calculation model. Combined with the comprehensive weighted calculation of the transparency performance vector, a precise quantitative evaluation is completed and the quantitative detection result is output. After iterative optimization and calibration of the distortion correction parameters and optical calculation parameters, all the latest optimized and calibrated distortion correction parameters, optical refractive index parameters, and scattering attenuation optical calculation parameters are precisely preset and input into the pre-built and trained transparency evaluation calculation model. This completes the real-time parameter update and working condition adaptation calibration of the model, ensuring that the model's calculation parameters conform to the current actual detection conditions and the true optical properties of the object. Subsequently, the overall transparency performance vector constructed in the first step is synchronously imported into the transparency evaluation calculation model. The model performs layered weighted fusion calculation and comprehensive quantitative conversion processing on the multi-dimensional feature data of the overall transparency performance vector according to the preset comprehensive weighted calculation rules and the weight ratio of different transparency feature parameters. This takes into account the overall transparency level of the object and the differences in local transparency fluctuations, comprehensively offsets the optical calculation deviation and residual interference of imaging distortion, and accurately calculates the comprehensive and accurate quantitative value of the transparency of the object under the current lighting conditions. After the model calculation is completed, standardized quantitative test results are automatically generated, and the comprehensive transparency score, transparency level judgment data, and full-area transparency quantitative distribution report of the object under test are intuitively output. This completes the high-precision, digital, accurate quantitative full-dimensional evaluation of the transparency of non-planar, unevenly thick, and transparent objects, realizing the detection goal of transparency from qualitative judgment to accurate quantitative numerical output, and meeting the actual production application needs of product quality grading and accurate light transmission performance testing.

[0045] Secondly, this embodiment provides a transparency detection system based on image detection, the system comprising: The acquisition module uses an image acquisition device to acquire image sequences of the object to be detected, and records the angle distribution of the incident light rays in layers to obtain a raw data set containing the shape changes and related parameters of the object to be detected. The reconstruction module, based on the original data set, uses matching technology combined with the incident light angle distribution and surface normal vector calculation to reconstruct the preliminary geometric contour of the object to be detected and obtain basic surface structure data. The module is divided into sections. The light propagation process is simulated by the light propagation simulation method. Combined with the preset refractive index parameters and light path processing, the boundary reflection intensity and internal scattering influence factor are analyzed to obtain the refraction distribution and distortion region boundary division results corresponding to the light path. The correction module extracts the light intensity distribution mapping corresponding to each pixel based on the refraction distribution and the boundary division results of the distortion region. It calculates the light intensity deviation of each pixel in combination with the preset deviation threshold. When the light intensity deviation exceeds the preset range, the distortion correction algorithm is used to eliminate the interference of the local distortion region and obtain a clear image after correction. The local transparency feature acquisition module, based on the corrected clear image, determines the distribution of local transparency features of the object to be detected by adjusting the light intensity correction coefficient and image optimization processing, and combining the distortion residual error evaluation data analysis. Transparency performance distribution map acquisition module. Based on the local transparency feature distribution, it integrates the light propagation distance and surface thickness variation data of the object to be detected. When the light intensity correction coefficient of the uneven thickness area deviates from the standard value, it adaptively adjusts the weight of the internal scattering influence factor to generate the final transparency performance distribution map. The evaluation module extracts the overall transparency vector of the object under test based on the final transparency performance distribution map, and achieves a precise quantitative evaluation of the transparency of the object under test by optimizing distortion correction and optical calculation parameters.

[0046] It should be noted that, in this document, the terms "comprising," "including," and any other variations are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Specific examples have been used in this document to illustrate the principles and implementation methods of the technical solutions of this application. The above examples are only for the purpose of helping to understand the methods and core ideas of this application. The above descriptions are merely preferred embodiments of this application. It should be pointed out that, due to the limitations of written expression and the objective existence of infinite specific structures, those skilled in the art can make several improvements, modifications, or changes without departing from the principles of this application, and can also combine the above technical features in an appropriate manner; these improvements, modifications, changes, or combinations, or the direct application of the concept and technical solutions of this application to other situations without modification, should all be considered within the scope of protection of this application.

Claims

1. A transparency detection method based on image detection, characterized in that, The method includes the following steps: S1. Use an image acquisition device to acquire an image sequence of the object to be detected, and record the incident light angle distribution in layers to obtain a raw data set containing the shape changes and related parameters of the object to be detected; S2. Based on the original data set, a matching technique is used to combine the incident light angle distribution and the surface normal vector calculation to reconstruct the preliminary geometric contour of the object to be detected and obtain the basic surface structure data; S3. The light propagation process is simulated by light propagation simulation method. Combined with preset refractive index parameters and light path processing, the boundary reflection intensity and internal scattering influence factor are analyzed to obtain the refraction distribution and distortion region boundary division results corresponding to the light path. S4. Based on the refraction distribution and the boundary division results of the distortion region, extract the light intensity distribution mapping corresponding to each pixel, calculate the light intensity deviation of each pixel in combination with the preset deviation threshold, and when the light intensity deviation exceeds the preset range, use the distortion correction algorithm to eliminate the interference of the local distortion region and obtain the corrected clear image. S5. Based on the corrected clear image, the local transparency feature distribution of the object to be detected is determined by adjusting the light intensity correction coefficient and image optimization processing, and by combining the distortion residual error evaluation data analysis. S6. Based on the local transparency feature distribution, the light propagation distance and the surface thickness change data of the object to be detected are fused. When the light intensity correction coefficient of the uneven thickness area deviates from the standard value, the weight of the internal scattering influence factor is adaptively adjusted to generate the final transparency performance distribution map. S7. Extract the overall transparency vector of the object to be tested based on the final transparency performance distribution map, and achieve a quantitative assessment of the transparency of the object to be tested by optimizing distortion correction and optical calculation parameters.

2. The transparency detection method based on image detection according to claim 1, characterized in that, Step S1 specifically includes: Adjust the shooting focal length, exposure parameters, sampling frame rate and shooting position of the image acquisition device, and fix the object to be detected at the preset detection position; The illumination sources are arranged in a layered gradient according to different incident light pitch angles and horizontal deflection angles, and the light incident angles are adjusted step by step while the image acquisition device is triggered to continuously sample. Collect a continuous frame image sequence of the object to be detected, and synchronously associate, mark and archive the incident light angle parameters and shooting coordinate parameters corresponding to each frame image; Invalid image data is filtered out to obtain a set of valid original images and associated parameters containing features of shape changes of the object to be detected, light refraction shift of curved surfaces, and changes in light and shadow.

3. The transparency detection method based on image detection according to claim 1, characterized in that, Step S2 specifically includes: The valid image data and associated parameters in the original dataset are retrieved, and feature point matching and contour matching techniques are used to accurately match and align the edge features and surface feature points of the object to be detected in the image. Based on the effective image data after matching and alignment, and combined with the light incident angle parameter and the image pixel coordinate conversion relationship, the normal vector value of each position on the surface of the object to be detected is calculated point by point. By fitting normal vectors and using a 3D contour reconstruction algorithm, combined with feature point spatial coordinate conversion, the 3D geometric contour model of the object to be detected is initially reconstructed. After performing contour denoising, edge calibration, and surface fitting correction on the three-dimensional geometric contour, the core parameters of the surface curvature, contour boundary, and surface undulation of the object to be detected are extracted to obtain basic surface structure data.

4. The transparency detection method based on image detection according to claim 1, characterized in that, Step S3 specifically includes: Based on the material properties of the object to be tested, preset the corresponding basic optical refractive index parameters, light absorption coefficient and interface reflection coefficient, and build a light propagation simulation environment for the object to be tested. Based on the basic surface structure data of the object to be detected, the light propagation simulation system is imported to simulate the entire light path of light rays passing through the object at different incident angles, including incident, refraction, reflection, and exit. Quantitatively calculate the light reflection intensity value at the boundary of the surface of the object to be detected, calibrate the degree of light scattering attenuation in different regions inside the object, and quantify the corresponding internal scattering influence factor. By combining the light path offset, the change in the light refraction angle, and the degree of light and shadow distortion, the distribution law of light path refraction across the entire region is statistically analyzed to define the critical boundary between the distorted area and the normal transparent area of ​​the image, and to obtain the boundary division result of the distorted area.

5. The transparency detection method based on image detection according to claim 1, characterized in that, Step S4 includes: Based on the boundary division results of the distortion region, the acquired image is divided into a global pixel grid, and the real-time light intensity value corresponding to each pixel is extracted pixel by pixel to construct a global pixel light intensity distribution mapping map. Pre-calibrate the pixel light intensity reference value under standard transparent conditions, compare the real-time light intensity value of each pixel with the reference value, and calculate the real-time light intensity deviation of each pixel one by one. By comparing the light intensity deviation of each pixel with the preset deviation threshold, local distorted pixel areas with light intensity deviation exceeding the preset range are selected, and abnormal areas of the image are accurately located. The preset distortion correction algorithm is invoked to perform targeted correction and repair on the abnormal area, eliminating recognition interference caused by refraction distortion and ambient light and shadow, and outputting a clear image after correction.

6. The transparency detection method based on image detection according to claim 1, characterized in that, Step S5 includes: Based on the corrected clear image, according to the mapping table between optical parameters and light intensity correction coefficients, the light intensity correction coefficients corresponding to the optical parameters are obtained, and the light intensity of all pixels in the image is adjusted for equalization based on the light intensity correction coefficients. The image after light intensity adjustment is optimized to obtain the optimized image; The distortion residual data of the corrected and optimized image is collected and compared with the standard distortion-free image sample to obtain the image residual distortion residual error assessment result. Based on the residual error assessment results, the light transmittance, light attenuation, and scattering loss characteristics of each local area are analyzed according to different curved areas and different thicknesses of the object to be tested, so as to determine the refined local transparency characteristics distribution of the object to be tested.

7. The transparency detection method based on image detection according to claim 1, characterized in that, Step S6 includes: Simultaneously correlate local transparency feature distribution, actual light propagation distance parameters of light penetrating the object under test, and measured surface thickness distribution variation data of the object under test; Pre-set the standard light intensity correction coefficient reference values ​​corresponding to different thickness ranges, and compare the real-time light intensity correction coefficient of each thickness uneven area with the corresponding standard light intensity correction coefficient reference value to calculate the coefficient deviation difference of each thickness uneven area. Determine whether the real-time light intensity correction coefficient of each uneven thickness region deviates from the corresponding standard value. If there is a deviation, adjust the weight ratio of the scattering influence factor inside the corresponding detection area according to the magnitude of the coefficient deviation. Based on the weighted internal scattering influence factor, the transparency data of each region is recalculated, and the transparency data of each region is integrated and stitched together to generate the final transparency performance distribution map of the object to be detected.

8. The transparency detection method based on image detection according to claim 1, characterized in that, Step S7 includes: Based on the final transparency performance distribution map, all feature parameters are integrated to construct an overall transparency performance vector that characterizes the overall light transmission properties of the object to be detected. Based on the current lighting conditions and transparency detection error feedback data, the distortion correction algorithm transformation parameters and optical calculation parameters related to optical refractive index and scattering attenuation are iteratively optimized to reduce residual image distortion error and optical calculation system deviation. The optimized distortion correction parameters and optical calculation parameters are input into the transparency evaluation calculation model. Combined with the overall transparency performance vector, a comprehensive weighted calculation is performed to complete the accurate quantitative evaluation of the transparency of the object under the current illumination conditions and output the quantitative detection results.

9. A transparency detection system based on image detection, characterized in that, The system includes: The acquisition module uses an image acquisition device to acquire image sequences of the object to be detected, and records the angle distribution of the incident light rays in layers to obtain a raw data set containing the shape changes and related parameters of the object to be detected. The reconstruction module, based on the original data set, uses matching technology combined with the incident light angle distribution and surface normal vector calculation to reconstruct the preliminary geometric contour of the object to be detected and obtain basic surface structure data. The module is divided into sections. The light propagation process is simulated by the light propagation simulation method. Combined with the preset refractive index parameters and light path processing, the boundary reflection intensity and internal scattering influence factor are analyzed to obtain the refraction distribution and distortion region boundary division results corresponding to the light path. The correction module extracts the light intensity distribution mapping corresponding to each pixel based on the refraction distribution and the boundary division results of the distortion region. It calculates the light intensity deviation of each pixel in combination with the preset deviation threshold. When the light intensity deviation exceeds the preset range, the distortion correction algorithm is used to eliminate the interference of the local distortion region and obtain a clear image after correction. The local transparency feature acquisition module, based on the corrected clear image, determines the distribution of local transparency features of the object to be detected by adjusting the light intensity correction coefficient and image optimization processing, and combining the distortion residual error evaluation data analysis. Transparency performance distribution map acquisition module. Based on the local transparency feature distribution, it integrates the light propagation distance and surface thickness variation data of the object to be detected. When the light intensity correction coefficient of the uneven thickness area deviates from the standard value, it adaptively adjusts the weight of the internal scattering influence factor to generate the final transparency performance distribution map. The evaluation module extracts the overall transparency vector of the object under test based on the final transparency performance distribution map, and achieves a precise quantitative evaluation of the transparency of the object under test by optimizing distortion correction and optical calculation parameters.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores an image detection transparency detection program, and when the transformer test strategy generation program is executed by the processor, it implements the image detection-based transparency detection method as described in any one of claims 1 to 8.

Images