Hand model surface defect detection method and system based on image analysis

Abstract

The invention relates to the technical field of hand model surface defect detection, in particular to a hand model surface defect detection method and system based on image analysis, and the method comprises the following steps: obtaining image information of a to-be-detected transparent or semitransparent hand model; identifying a local area which is different from the background light and shadow change in the image as an abnormal area; according to the abnormal region, determining a local region to be finely analyzed; performing multi-aspect feature analysis on the local area to be subjected to fine analysis so as to identify whether the local area to be subjected to fine analysis is a surface defect or not; and outputting the surface defect detection information of the hand model according to the distinguishing result of the surface defect. According to the method, the surface defects can be accurately recognized, the detection accuracy and reliability are remarkably improved, and the limitation of a traditional method in the aspect of distinguishing the surface defects from internal illustrations is overcome.

Description

Technical Field

[0001] This invention relates to the technical field of surface defect detection for prototype models, and specifically to a method and system for surface defect detection of prototype models based on image analysis. Background Technology

[0002] Prototype models play a crucial role in product development and verification, and their surface quality directly impacts the final product's appearance and functionality. To ensure quality, automated surface defect detection systems have emerged to overcome the low efficiency and high false negative rate of traditional manual visual inspection. These systems typically capture images using industrial cameras and employ image analysis methods to automatically identify defects such as scratches and pores. However, when the object being inspected becomes a transparent or semi-transparent prototype model, light undergoes complex refraction, reflection, and scattering as it penetrates the transparent or semi-transparent material. This results in camera images containing not only information from the model's surface but also a large amount of complex light and shadow generated by the material's internal optical effects. These internal light and shadow effects are highly similar to or mask each other in image features with inherent defects within the material (such as bubbles and impurities) and actual defects on the model's surface. This makes it difficult for the detection system to accurately distinguish between surface defects and internal artifacts, leading to a large number of false positives and false negatives, severely affecting the accuracy and reliability of the inspection. Summary of the Invention

[0003] The purpose of this invention is to address the aforementioned shortcomings by proposing a method and system for detecting surface defects in prototype models based on image analysis.

[0004] The present invention adopts the following technical solution: A method for detecting surface defects in a prototype model based on image analysis, comprising the following steps: Acquire image information of the transparent or semi-transparent hand model to be detected. The image information is obtained by changing the pose and / or lighting method of the hand model. By comparing and analyzing local regions in the image information, the results of the comparison and analysis are obtained. Based on the results of the comparison and analysis, local regions in the image that differ from the changes in light and shadow of the background are identified as abnormal regions. Based on the abnormal areas, determine the local areas that need to be analyzed in detail; Multi-faceted feature analysis is performed on the local area to be analyzed in detail to identify whether the local area to be analyzed in detail is a surface defect. The multi-faceted feature analysis includes: evaluating the light response of the local area to be analyzed in detail under different lighting conditions, analyzing the shape features of the local area to be analyzed in detail, and judging the image position change of the local area to be analyzed in detail when the pose of the prototype model changes. Based on the identification results of surface defects, the surface defect detection information of the prototype model is output.

[0005] This technical solution effectively addresses the complex optical interference in the inspection of transparent or semi-transparent prototype models. By acquiring image information from multiple angles and under multiple illumination methods, and combining it with multi-faceted feature analysis, it achieves accurate identification of surface defects, significantly improving the accuracy and reliability of the inspection, and overcoming the limitations of traditional methods in distinguishing between surface defects and internal artifacts.

[0006] This application also discloses a prototype model surface defect detection system based on image analysis, applied to the above-mentioned image analysis-based prototype model surface defect detection method. The system includes: The acquisition module acquires image information of the transparent or semi-transparent hand model to be detected. The image information is obtained by changing the pose and / or lighting method of the hand model. The recognition module compares and analyzes local regions in the image information, obtains comparison and analysis results, and identifies local regions in the image that differ from the background lighting and shadow changes as abnormal regions. The module identifies the local area to be analyzed in detail based on the abnormal area. The analysis module performs multi-faceted feature analysis on the local area to be analyzed in detail in order to identify whether the local area to be analyzed in detail is a surface defect. The multi-faceted feature analysis includes: evaluating the light response of the local area to be analyzed in detail under different lighting conditions, analyzing the shape features of the local area to be analyzed in detail, and judging the image position change of the local area to be analyzed in detail when the pose of the prototype model changes. The output module outputs surface defect detection information of the prototype model based on the surface defect identification results.

[0007] This application provides a system that can effectively perform the above-mentioned detection methods. Through modular design, it realizes an automated process of image acquisition, abnormal area identification, fine analysis and result output, providing efficient and reliable hardware and software support for the detection of surface defects in transparent or semi-transparent prototype models.

[0008] This application significantly improves the accuracy and reliability of surface defect detection for transparent or semi-transparent prototype models, effectively solving the problems of low efficiency and high missed detection rate of traditional manual visual inspection, as well as serious false alarms and missed detections in automated inspection systems, providing strong technical support for product quality control.

[0009] To further understand the features and technical content of the present invention, please refer to the following detailed description and drawings of the present invention. However, the drawings provided are for reference and illustration only and are not intended to limit the present invention. Attached Figure Description

[0010] Figure 1 This is a flowchart of a method for detecting surface defects in a prototype model based on image analysis, according to the present invention. Figure 2 This is a schematic diagram of the structure of a prototype model surface defect detection system based on image analysis according to the present invention. Detailed Implementation

[0011] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can understand the advantages and effects of the present invention from the content disclosed in this specification. The present invention can be implemented or applied through other different specific embodiments, and various details in this specification can also be modified and changed based on different viewpoints and applications without departing from the spirit of the present invention. Furthermore, the accompanying drawings of the present invention are for simple illustrative purposes only and are not depictions of actual dimensions; this is stated in advance. The following embodiments will further describe the relevant technical content of the present invention in detail, but the disclosed content is not intended to limit the scope of protection of the present invention.

[0012] This embodiment provides a method and system for detecting surface defects in hand-made prototype models based on image analysis, combined with... Figure 1 and Figure 2 As shown.

[0013] refer to Figure 1 A method for detecting surface defects in a prototype model based on image analysis, comprising the following steps: Acquire image information of the transparent or semi-transparent hand model to be detected. The image information is obtained by changing the pose and / or lighting method of the hand model. By comparing and analyzing local regions in the image information, the results of the comparison and analysis are obtained. Based on the results of the comparison and analysis, local regions in the image that differ from the changes in light and shadow of the background are identified as abnormal regions. Based on the abnormal areas, determine the local areas that need to be analyzed in detail; Multi-faceted feature analysis is performed on the local area to be analyzed in detail to identify whether the local area to be analyzed in detail is a surface defect. The multi-faceted feature analysis includes: evaluating the light response of the local area to be analyzed in detail under different lighting conditions, analyzing the shape features of the local area to be analyzed in detail, and judging the image position change of the local area to be analyzed in detail when the pose of the prototype model changes. Based on the identification results of surface defects, the surface defect detection information of the prototype model is output.

[0014] Among them, "prototype model" refers to a physical model used to verify the design, function and appearance during the product development process, and its surface quality is crucial to the final evaluation of the prototype model.

[0015] "Image information" refers to two-dimensional or three-dimensional data that includes surface and / or internal features of a prototype model, captured by optical imaging equipment such as an industrial camera.

[0016] A "local region" refers to a sub-region of image information that is designated for specific analysis; it can be regular or irregular in shape.

[0017] "Background lighting variations" refer to non-defective changes in brightness, color, or texture in an image caused by internal optical effects such as light penetration, refraction, reflection, and scattering in transparent or translucent materials.

[0018] "Abnormal areas" refer to local areas in an image that differ significantly from the background lighting and shadow changes, as identified through comparative analysis. These areas may contain surface defects or features that require further analysis.

[0019] "Multi-faceted feature analysis" refers to a comprehensive evaluation of a local area, including but not limited to light response, shape features, and changes in image position, in order to fully determine whether it is a surface defect.

[0020] The detection method of this application first requires acquiring image information of the transparent or semi-transparent prototype model to be detected. Image acquisition can be achieved in various ways. For example, a fixed camera and light source can be used, and the pose of the prototype model can be changed by a robotic arm or turntable to capture images from different angles. Alternatively, images can be acquired by switching illumination sources at different positions, angles, or spectra while keeping the prototype model's pose unchanged. Furthermore, both methods can be combined, i.e., adjusting the illumination method while changing the prototype model's pose, to obtain more comprehensive image data. For example, a set of images can be acquired first under diffused light, and then another set under directional light, or an image can be captured at regular intervals during a 360-degree rotation of the prototype model.

[0021] After acquiring image information, it is necessary to compare and analyze local regions within the image to obtain comparative analysis results. Comparative analysis can be implemented using various algorithms. For example, image differencing methods can be used to perform pixel-level subtraction operations on images under different poses or lighting conditions to highlight areas of change. Statistical methods can also be used to calculate statistics on features such as brightness, contrast, and texture of local regions and compare them with the statistics of a pre-defined normal background region. Furthermore, machine learning methods can be used to train a classifier to identify local regions in the image that differ from changes in background lighting. For example, two images of the same prototype model under different lighting conditions can be compared pixel-by-pixel, or a local region of the current image can be compared with a pre-established image database of defect-free prototype models. Based on the comparative analysis results, local regions in the image that differ from changes in background lighting are identified as anomalous regions.

[0022] Once anomalous areas are identified, local regions for detailed analysis are determined based on these areas. Determining these local regions can be achieved through various strategies. For example, all identified anomalous areas can be included as local regions for detailed analysis. Alternatively, anomalous areas can be further filtered, for example, by identifying characteristics such as size, shape, or intensity to eliminate areas that are clearly not defects. Furthermore, adjacent anomalous areas can be merged to form larger local regions for detailed analysis. For instance, if multiple small anomalous areas are spatially close to each other, they can be merged into a larger region for subsequent analysis.

[0023] Subsequently, multi-faceted feature analysis is performed on the local area to be analyzed in detail to identify whether the local area is a surface defect. Multi-faceted feature analysis includes evaluating the light response of the local area to be analyzed in detail under different lighting conditions, analyzing the shape characteristics of the local area to be analyzed in detail, and determining the image position change of the local area to be analyzed in detail when the pose of the prototype model changes.

[0024] For evaluating light response, one can analyze the changes in brightness, color, and contrast of a local area under different lighting conditions (such as diffused light, directional light, polarized light, etc.). For example, surface defects typically exhibit stronger specular reflection or scattering under directional light, while they may be less noticeable under diffused light.

[0025] For shape feature analysis, geometric features of local areas can be extracted, such as area, perimeter, aspect ratio, roundness, and convexity. For example, scratches usually appear as elongated shapes, while bubbles may appear as circles or ellipses.

[0026] To determine changes in image position, one can track the positional shift of local areas in the image as the prototype model's posture changes. For example, surface defects typically shift in relative position as the prototype model rotates, while internal optical effects may remain relatively fixed or move in different ways.

[0027] Finally, based on the surface defect identification results, the surface defect detection information of the prototype model is output. The output information may include the location, type, size, and severity of the defects. For example, detected defects can be visually marked on the 3D model of the prototype model, or a detailed inspection report can be generated. Output methods include displaying on a screen, printing, or sending to other systems via a network.

[0028] The detection method presented in this application effectively solves the challenge of detecting surface defects in transparent or semi-transparent prototype models through a series of collaborative steps. First, by altering the prototype model's pose and / or lighting conditions, image information is acquired under multiple angles and lighting conditions. This allows the system to capture the unique optical responses of surface defects under different conditions, while also better separating internal optical effects. Next, local regions in the image information are compared and analyzed to identify anomalous regions that differ from changes in background lighting. This step utilizes the differences in image features between surface defects and internal optical effects to initially screen potential defect areas. Subsequently, based on the anomalous regions, local areas requiring detailed analysis are determined, avoiding intensive computation of the entire image and improving efficiency. Most importantly, multi-faceted feature analysis is performed on the local areas requiring detailed analysis, including light response, shape features, and image position changes. This multi-dimensional and comprehensive analysis method enables the system to more accurately distinguish between surface defects and internal optical effects, as surface defects typically exhibit patterns different from internal artifacts in these features. For example, surface defects may exhibit more pronounced light responses under different lighting conditions, their shape characteristics may better match specific defect types, and the image position changes of surface defects differ from internal optical effects when the prototype model's posture changes. Ultimately, based on the discrimination results from multi-faceted feature analysis, surface defect detection information for the prototype model is output, thus providing accurate and reliable detection results.

[0029] This application further proposes steps for identifying whether a local area to be analyzed in detail is a surface defect, including: A multi-frequency structured light pattern is projected onto the local area to be analyzed in detail, and a sequence of structured light images of the local area to be analyzed in detail is obtained. Obtain vibration data from the testing platform; Displacement compensation is performed on structured light image sequences using vibration data; Phase calculation is performed on the compensated structured light image sequence to obtain the phase calculation results; The phase calculation results are converted into three-dimensional surface topography data of the local area to be analyzed in detail; Local curvature and depth variation information are extracted from the three-dimensional surface topography data; Based on local curvature and depth variation information, determine whether the local area to be analyzed in detail is a surface defect.

[0030] Specifically, projecting multi-frequency structured light patterns onto a localized area to be analyzed involves using a projection device to project a series of stripe patterns with specific spatial frequencies and phases onto the target area, such as sinusoidal fringes or Gray code. By changing the frequency and phase of the structured light, multiple structured light images can be acquired, containing surface topography information of the target area. The aim is to achieve non-contact, high-precision measurement of the three-dimensional information of an object's surface by encoding the light field.

[0031] Acquiring vibration data from the detection platform can be understood as simultaneously monitoring and recording minute displacements or jitters that may occur on the platform during the structured light image sequence acquisition process. This can be achieved by installing vibration sensing units such as high-precision accelerometers, gyroscopes, or laser displacement sensors on the detection platform. The purpose is to quantify and record the impact of the external environment or the equipment's own operation on the stability of the measurement system, providing a basis for subsequent image correction.

[0032] In practical applications, vibration data is used to compensate for displacement in structured light image sequences. Specifically, this involves calculating the offset of each frame of the structured light image relative to its ideal position based on the acquired vibration data, and then performing corresponding geometric transformations (such as translation and rotation) on the image to eliminate image distortion and misalignment caused by vibration. For example, compensation can be performed using image registration algorithms or motion estimation models. The aim is to ensure the continuity and accuracy of the structured light image sequence, providing high-quality input data for subsequent phase calculation.

[0033] Furthermore, phase resolution is performed on the compensated structured light image sequence to obtain the phase resolution result. This involves extracting the wrapped phase information from the compensated structured light image sequence using specific algorithms (such as multi-frequency heterodyne method, phase shift method, etc.) and then unwrapping it to obtain a continuous phase distribution map. The purpose is to convert the deformation information of the structured light pattern on the object surface into continuous phase values, which are directly related to the three-dimensional morphology of the object surface.

[0034] Therefore, converting the phase calculation results into surface 3D topographic data of the local area to be analyzed in detail involves using calibration parameters and a geometric model to map the calculated phase information to 3D spatial coordinates, thereby reconstructing an accurate 3D point cloud or mesh model of the local area to be analyzed in detail. The purpose is to obtain realistic and quantified geometric topographic data of the prototype model surface, providing a foundation for subsequent defect analysis.

[0035] Extracting local curvature and depth variation information from surface 3D topography data specifically involves performing geometric analysis on the reconstructed 3D topography data to calculate the curvature (such as Gaussian curvature, mean curvature) of local regions and the local fluctuations or gradient changes in surface height. This can be achieved, for example, by fitting local surfaces or calculating changes in normal vectors. The aim is to extract geometric parameters that characterize surface defects from quantified 3D data, as many surface defects (such as pits, bumps, and scratches) can cause abnormal changes in local curvature or depth.

[0036] Ultimately, based on local curvature and depth variation information, determining whether a local area to be analyzed in detail is a surface defect involves comparing the extracted local curvature and depth variation information with preset defect discrimination criteria or models to determine whether the geometric features of the local area meet the definition of a surface defect. For example, curvature thresholds and depth variation gradient thresholds can be set; areas exceeding these thresholds are judged as defects. The aim is to achieve accurate and objective identification of surface defects based on precise three-dimensional geometric features.

[0037] Specifically, during the acquisition of structured light image sequences, minute vibrations of the detection platform can cause displacement and distortion in the image sequence, thus affecting the accuracy of 3D reconstruction. This application addresses this issue by synchronously acquiring vibration data of the detection platform and using this data to compensate for displacement in the structured light image sequence, effectively eliminating measurement errors caused by vibration and ensuring the accuracy of the image sequence. Subsequently, phase calculation is performed on the compensated structured light image sequence to convert the light field information into a continuous phase distribution. The phase calculation results are then converted into surface 3D topography data of the local area to be analyzed in detail. This process realizes the transformation from 2D images to high-precision 3D topography, laying the foundation for accurate defect identification.

[0038] Building upon this foundation, this application extracts local curvature and depth variation information from surface 3D topography data. Surface defects, such as pits, protrusions, and scratches, typically cause significant geometric changes in local areas, manifested as abnormal increases or decreases in curvature and abrupt changes in depth. By analyzing these quantified 3D geometric features, these minute topographical anomalies can be accurately captured, thus avoiding misjudgments caused by factors such as lighting and viewing angle in traditional methods. Ultimately, based on this precise local curvature and depth variation information, it is possible to reliably identify whether the local area to be analyzed in detail is a surface defect, significantly improving the accuracy and reliability of surface defect detection for transparent or semi-transparent prototype models.

[0039] Through the above technical solution, this application can effectively solve the problems of insufficient discrimination accuracy and robustness caused by complex optical effects and environmental interference in the detection of surface defects of transparent or semi-transparent prototype models using traditional two-dimensional image analysis methods. By introducing multi-frequency structured light measurement technology, high-precision reconstruction of the three-dimensional morphology of the defect area surface is achieved, overcoming the limitations of two-dimensional image information in terms of depth and geometric details.

[0040] Furthermore, by acquiring vibration data from the detection platform and performing displacement compensation on the structured light image sequence, the influence of external vibration on measurement accuracy is effectively eliminated, ensuring the accuracy of the three-dimensional topography data. Finally, by extracting local curvature and depth variation information based on accurate three-dimensional surface topography data, the geometric features of surface defects can be characterized more objectively and quantitatively. This significantly improves the ability to distinguish between small, complex, or optically inconspicuous surface defects, reduces false alarms and false negatives, and makes surface defect detection on transparent or semi-transparent prototype models more accurate, reliable, and automated.

[0041] In some preferred embodiments, it is assumed that it is necessary to detect minute scratches or dents on the surface of a transparent resin prototype model. First, the prototype model is placed on a detection platform. When the system identifies an abnormal area in the image and determines that it is a local area to be analyzed in detail, the detection system activates a structured light projection device to project a series of sinusoidal stripe structured light patterns of different frequencies and phases onto the local area. Simultaneously, a high-speed camera synchronously acquires a sequence of deformed image sequences formed by these structured light patterns on the surface of the prototype model.

[0042] During this process, a high-precision accelerometer installed on the detection platform collects the platform's vibration signals in real time. This vibration data is used to perform frame-by-frame displacement compensation on the acquired structured light image sequence to eliminate image misalignment caused by minor platform jitter. The compensated image sequence is then sent to the image processing unit, where phase calculation is performed using the multi-frequency heterodyne method to obtain the encapsulated phase map of the local region. Unencapsulation processing is then performed to obtain a continuous phase distribution.

[0043] Next, using pre-calibrated system parameters, the calculated phase information is converted into surface 3D topographic data of the local region, forming a high-precision 3D point cloud model. On this 3D model, the system calculates the local curvature of the local region (e.g., by fitting a local quadratic surface to calculate Gaussian curvature and mean curvature) and depth variation information (e.g., by calculating the local height gradient or the relative depth with the surrounding flat region).

[0044] For example, if a sudden increase in the average curvature of a local area is detected within a small range, and depth information shows a slight depression, and these features match a pre-defined geometric model of a scratch or pit, then the local area will be identified as a surface defect. In this way, even minute defects that are difficult to detect with the naked eye can be accurately identified through their unique geometric features in three-dimensional morphology, avoiding misjudgments that may be caused by complex optical phenomena such as light reflection and refraction in traditional methods.

[0045] The steps for obtaining vibration data from the testing platform include: A vibration sensing unit is installed on the testing platform; During the acquisition of structured light image sequences, vibration signals output by the vibration sensing unit are simultaneously acquired. Vibration data is obtained by time alignment and data preprocessing of the vibration signal.

[0046] Specifically, a vibration sensing unit is installed on the detection platform. This unit can be understood as a sensor used to monitor the platform's motion in real time; for example, an accelerometer, gyroscope, or a combination thereof can be used. Its purpose is to accurately capture minute vibrations of the detection platform during the acquisition of structured light image sequences. In practical applications, the vibration sensing unit is strategically mounted on the detection platform to ensure it can directly and accurately sense the platform's vibrations.

[0047] In the process of acquiring structured light image sequences, simultaneously acquiring vibration signals output by the vibration sensing unit ensures a high degree of temporal consistency between the acquisition of vibration signals and the acquisition of each frame of the structured light image sequence. This synchronous acquisition mechanism aims to provide a precise time reference for subsequent displacement compensation, thereby effectively eliminating image distortion caused by platform vibration.

[0048] Furthermore, the vibration signals undergo time alignment and data preprocessing to obtain vibration data. Time alignment refers to precisely matching the acquired vibration signals with the corresponding structured light image sequences on the time axis to ensure that each frame of the image can obtain its corresponding accurate vibration state data. Data preprocessing includes filtering, denoising, and baseline drift correction of the raw vibration signals to eliminate environmental interference and sensor errors, thereby obtaining high-quality, high-reliability vibration data to provide accurate input for subsequent displacement compensation.

[0049] The proposed solution utilizes a dedicated vibration sensing unit on the detection platform, ensuring synchronous acquisition of vibration signals during the structured light image sequence acquisition process. This allows for real-time and accurate capture of the platform's dynamic vibration information. This synchronous and precise acquisition method enables subsequent time alignment and data preprocessing of the vibration signals, effectively eliminating noise and timing deviations, thus yielding high-quality vibration data. This optimized vibration data can be used more precisely for displacement compensation in the structured light image sequence, significantly improving the accuracy of compensation and ensuring the precision of surface 3D topography reconstruction.

[0050] In some preferred embodiments, the detection platform can be equipped with a triaxial MEMS accelerometer as a vibration sensing unit. While a structured light projector projects a multi-frequency structured light pattern onto the prototype model under test and a camera acquires a sequence of structured light images, the accelerometer collects acceleration data in the X, Y, and Z directions at a frequency synchronized with the camera's frame rate (e.g., 100 frames per second). The acquired raw acceleration signals are first denoised using a low-pass filter to eliminate high-frequency environmental noise. Subsequently, the acceleration data is time-aligned using the camera's timestamp information to ensure that each frame corresponds to a precise vibration state. Next, the aligned acceleration data is integrated to obtain the platform's displacement information, and baseline drift correction is performed. Finally, this preprocessed and transformed displacement data is used as vibration data to accurately compensate for the displacement of the structured light image sequence, thereby eliminating image distortion caused by platform vibration and ensuring the accuracy of phase calculation and 3D topography reconstruction.

[0051] This application further proposes steps for extracting local curvature information from surface three-dimensional topography data, including: Multi-scale geometric feature analysis is performed on the local area to be analyzed in detail based on the surface three-dimensional topography data. The surface roughness parameters of the local area to be analyzed in detail are calculated at different spatial scales, and the local curvature parameters of the local area to be analyzed in detail are calculated at different spatial scales. The relative magnitudes and distribution patterns of surface roughness parameters and local curvature parameters at different spatial scales are compared to obtain the comparison results. Based on the comparison results, it is determined whether the local area to be analyzed in detail has the curvature characteristics of surface defects, and the local curvature information is extracted based on the determination results.

[0052] Specifically, multi-scale geometric feature analysis refers to processing the three-dimensional surface topography data of a local area using analysis windows or filters of different sizes to obtain the geometric features of that area at multiple spatial scales, including macroscopic, mesoscopic, and microscopic. For example, methods such as Gaussian filtering, wavelet transform, or morphological filtering can be used to smooth or extract surface features at different scales. Surface roughness parameters can be understood as indicators that measure the degree of microscopic unevenness of a surface, such as arithmetic mean roughness Ra and root mean square roughness Rq, which aim to quantify subtle changes in surface texture. Local curvature parameters are indicators that describe the degree of surface curvature, such as principal curvature, Gaussian curvature, or mean curvature, which aim to reflect the concavity and convexity characteristics of the surface's macroscopic or local shape. By calculating these parameters at different spatial scales, the geometric characteristics of a local area can be comprehensively captured, distinguishing morphological differences caused by the material's inherent properties, processing techniques, or actual defects.

[0053] Furthermore, the relative magnitudes and distribution patterns of surface roughness parameters and local curvature parameters at different spatial scales are compared to obtain comparative results. Relative magnitude refers to the numerical relationship between roughness and curvature parameters at the same or different scales; distribution pattern refers to the spatial arrangement characteristics or statistical distribution law of these parameters within a local region. For example, for a typical surface defect, its roughness parameter at the microscale may be significantly higher than that of a normal surface, while the local curvature parameter at a specific scale may also exhibit abnormal concentration or abrupt changes. By comprehensively comparing the relative magnitudes and distribution patterns of these parameters, abnormal features that do not conform to the normal surface morphology can be identified more accurately.

[0054] Based on the comparison results, it is determined whether the local area to be analyzed in detail possesses the curvature characteristics of surface defects, and local curvature information is extracted based on the determination results. Specifically, a discrimination model based on historical data or expert knowledge can be established. This model can determine whether the geometric features of a local area conform to the curvature characteristic pattern of known surface defects based on the comparison results of roughness parameters and curvature parameters at different scales. For example, when a local area exhibits high roughness at the microscale and sharp curvature changes at the mesoscale, this area may be identified as possessing the curvature characteristics of surface defects. Thus, local curvature information truly related to surface defects can be accurately extracted from complex three-dimensional surface topography data, providing a reliable basis for subsequent defect classification and evaluation.

[0055] This application's solution effectively addresses the problem of accurately identifying surface defects based solely on a single scale or simple curvature extraction by introducing multi-scale geometric feature analysis and comprehensively comparing surface roughness parameters and local curvature parameters at different spatial scales. Specifically, traditional methods may only focus on curvature changes at a specific scale, easily misjudging inherent micro-textures or measurement noise as defects, or ignoring defect features that only become apparent at different scales. This application, by simultaneously calculating surface roughness parameters and local curvature parameters at multiple spatial scales, can more comprehensively capture the geometric characteristics of local regions. For example, a tiny scratch may exhibit high roughness and sharp curvature changes at a microscale, but be inconspicuous at a macroscale; while a large pit may exhibit significant curvature changes at a macroscale, but not noticeable roughness changes at a microscale. By comparing the relative size and distribution patterns of these parameters at different scales, a more discernible defect feature fingerprint can be formed. This multi-dimensional analysis method enables the system to distinguish between morphological fluctuations caused by normal surface texture, processing marks, or measurement errors and genuine surface defects, thereby avoiding misjudgment and improving the accuracy and robustness of defect identification.

[0056] In some preferred embodiments, it is assumed that the surface of the transparent prototype model to be tested has a minor scratch and a slight dent.

[0057] First, multi-scale geometric feature analysis is performed on the surface three-dimensional topography data of the local area to be analyzed in detail. For example, three spatial scales can be set: micro scale (e.g., analysis window size of 0.1 mm), meso scale (e.g., analysis window size of 1 mm), and macro scale (e.g., analysis window size of 10 mm).

[0058] At the microscale, for scratched areas, the calculated surface roughness parameters (such as Ra value) will be significantly higher than those for normal surface areas. At the same time, local curvature parameters (such as maximum principal curvature) will exhibit sharp peaks or valleys, reflecting the elongation and depth of the scratch. For pitted areas, the microscale roughness parameters may be similar to those of normal surfaces, but the local curvature parameters may show slight pitting characteristics.

[0059] At the mesoscale, for scratched areas, the roughness parameter may still be slightly higher than that of normal surfaces, but the sharpness of the curvature parameter is reduced. For pitted areas, the local curvature parameter at the mesoscale will more clearly reflect the overall shape and depth of the pit, while the roughness parameter may still remain within the normal range.

[0060] At a macroscopic scale, the roughness parameters and local curvature parameters of scratches and pits may not differ significantly from those of normal surface areas because they are smoothed out macroscopically.

[0061] Next, the relative magnitudes and distribution patterns of surface roughness parameters and local curvature parameters at different spatial scales are compared. For example, it can be observed that in the scratched area at the microscale, both the roughness parameter and the local curvature parameter exhibit abnormally high values, and the curvature change is directional. In the pitted area at the mesoscale, the local curvature parameter shows a significant local trough, while the roughness parameter is close to normal levels at all scales. In the normal surface area, both the roughness parameter and the local curvature parameter are within the preset normal fluctuation range at all scales, and the distribution pattern is uniform.

[0062] Based on these comparisons, the system can identify surface defect curvature characteristics in scratched areas ("microscopic high roughness + microscopic sharp curvature") and surface defect curvature characteristics in pitted areas ("mesoscopic significant concave curvature"). Therefore, local curvature information associated with these defects can be accurately extracted and used as a clear indicator of surface defects. This multi-scale, multi-parameter integrated analysis method enables the accurate identification and location of even tiny or complex-shaped defects.

[0063] This application further proposes a method for comparing the relative magnitudes and distribution patterns of surface roughness parameters and local curvature parameters at different spatial scales, the method comprising the following steps: Establish a local geometric feature reference file for the local area to be analyzed in detail on the surface three-dimensional topography data; Update the local geometric feature baseline file using the exponentially weighted moving average method; Compare the expected fluctuation range of surface roughness parameters and local geometric feature reference files at different spatial scales, compare the expected fluctuation range of local curvature parameters and local geometric feature reference files at different spatial scales, and preliminarily identify whether the local area to be analyzed in detail is a surface defect based on the expected fluctuation range. Continuous verification is carried out on local areas that are initially identified as surface defects and require detailed analysis.

[0064] Specifically, establishing a local geometric feature baseline file for the local area to be analyzed in detail based on the surface three-dimensional topography data refers to collecting historical or reference data on the surface roughness parameters and local curvature parameters of each local area on the surface of the prototype model under normal and defect-free conditions, and integrating them into a baseline file. This baseline file may include statistical information such as the mean, standard deviation, and distribution range of these parameters, with the aim of providing a reliable reference standard for subsequent defect identification.

[0065] Updating the local geometric feature baseline file using the exponentially weighted moving average method can be understood as dynamically maintaining and adjusting the aforementioned baseline file. The exponentially weighted moving average method assigns higher weight to recent data, enabling the baseline file to reflect subtle changes or material property shifts during the prototype model production process in real time, avoiding misjudgments due to outdated baselines. For example, when production batches or environmental conditions change, this method allows the baseline file to smoothly adapt to the new normal state, ensuring accurate judgment.

[0066] In practical applications, comparing the expected fluctuation range of surface roughness parameters at different spatial scales with the local geometric feature reference file, and comparing the expected fluctuation range of local curvature parameters at different spatial scales with the local geometric feature reference file, involves comparing the surface roughness parameters and local curvature parameters of the currently detected local area with the normal fluctuation range defined in the reference file. The expected fluctuation range can be determined based on historical data or statistical methods, representing the reasonable range of variation that these parameters may exhibit under normal circumstances. If the current parameters exceed this expected fluctuation range, it is preliminarily determined that there may be surface defects in the local area.

[0067] Furthermore, continuous verification of local areas initially identified as surface defects for detailed analysis refers to not immediately making a final judgment on those areas initially identified as defects, but instead conducting continuous monitoring or repeated testing from multiple angles and under multiple conditions over a period of time. The purpose is to eliminate misjudgments caused by transient noise, measurement errors, or non-continuous characteristics, ensuring that the identified defects are stable and genuine surface anomalies.

[0068] This application's solution effectively addresses the robustness limitations that may arise from simple comparisons by introducing a local geometric feature baseline file and its dynamic update mechanism. First, establishing a baseline file provides a quantifiable "normal" state reference for defect identification, ensuring that the identification is not solely based on the relative magnitudes of parameters but rather on deviations from known normal states. Second, updating the baseline file using an exponentially weighted moving average method allows the system to adapt to gradual changes during production, avoiding accumulated errors caused by static baselines and thus improving the adaptability and accuracy of the identification. Furthermore, comparing the current parameters with the expected fluctuation range of the baseline file allows for more precise identification of anomalies exceeding the normal range, effectively filtering out inherent microstructure or measurement noise from the prototype model. Finally, continuous verification of areas initially identified as surface defects further enhances the reliability of the identification, avoiding false alarms caused by transient interference or accidental factors, and ensuring the accuracy and stability of the final defect detection results.

[0069] In some preferred embodiments, it is assumed that a transparent prototype model is being inspected on the production line. First, at the initial stage of inspection, or for a batch of known qualified prototype models, the system collects its surface 3D topography data and, for each local area to be analyzed in detail, calculates its surface roughness parameters and local curvature parameters at different spatial scales. This data is used to establish an initial local geometric feature baseline profile, which includes the mean and standard deviation of these parameters, defining the expected range of variation for the surface of a normal prototype model.

[0070] During subsequent testing, whenever a new prototype model passes through the testing platform, the system acquires real-time 3D surface topography data of its local area and calculates the corresponding surface roughness parameters and local curvature parameters. These newly calculated parameters are compared with the baseline profile. For example, if the surface roughness parameter of a certain local area exceeds two standard deviations of the expected fluctuation range defined in the baseline profile, the system will initially identify it as a surface defect.

[0071] Meanwhile, to adapt to minor changes in the production environment, the benchmark archive is continuously updated using an exponentially weighted moving average method. For example, by setting a decay factor (such as 0.1), each time new qualified prototype model data is collected, the mean and standard deviation of the benchmark archive are updated by weighted averaging based on the new data and the old benchmark values. This allows the benchmark archive to slowly drift towards the latest normal state, thereby maintaining its validity.

[0072] For localized areas initially identified as surface defects, the system does not immediately output a defect report. Instead, it initiates a continuous verification process. For example, over the next few seconds, the system might re-acquire images or 3D topographic data of the localized area from different lighting angles or by slightly altering the pose of the prototype model, and recalculate its geometric features. If the anomalous feature persists in multiple repeated inspections, and its parameter values ​​consistently exceed the expected fluctuation range of the updated baseline profile, then the localized area is finally confirmed as a surface defect, and the corresponding detection information is output. This continuous verification mechanism effectively avoids misjudgments caused by instantaneous measurement errors or environmental interference.

[0073] The steps for continuous verification of local areas initially identified as surface defects and requiring further detailed analysis include: Based on the initially identified defect characteristics, the defect type of the local area to be analyzed in detail is predicted; Based on the predicted defect type, obtain the duration range and stability parameters corresponding to the defect type; Set the verification duration based on the duration range; During the verification period, images or three-dimensional topographic data of the local area to be analyzed in detail are continuously collected. Based on image or 3D topography data, calculate in real time the fluctuation amplitude of geometric features and optical response features of the local area to be analyzed in detail; The fluctuation amplitude is compared with the stability parameter, thereby adjusting the judgment criteria; Based on the adjusted criteria, confirm whether the local area to be analyzed in detail is a persistent surface defect.

[0074] Specifically, based on the initially identified defect characteristics, the defect type of the local area to be analyzed in detail is preliminarily judged. This refers to using the geometric features (such as size, shape, and depth) and optical response features (such as brightness, contrast, and reflectivity) contained in the preliminary identification results, combined with a pre-set defect classification model, to make a preliminary judgment on the category of potential defects. For example, it can be preliminarily judged as a scratch, dent, bubble, or foreign object. The purpose is to provide targeted parameter basis for subsequent continuous verification. Based on the predicted defect type, the duration range and stability parameters corresponding to the defect type are obtained. This can be understood as the different temporal and spatial stability of different types of defects. For example, a tiny scratch may remain stable for a short time, while a bubble may maintain its shape for a longer period. The duration range refers to the minimum and maximum time span that this type of defect typically exists, while the stability parameter quantifies the tolerance of the defect's characteristic fluctuations under different environmental or observation conditions. These parameters can be obtained from historical data, expert experience, or material property databases. In practical applications, the verification duration is set according to the duration range. For example, a suitable verification duration can be selected based on the duration range corresponding to the predicted defect type. This duration should be sufficient to observe the stable performance of that type of defect while avoiding excessively long verification times that could affect detection efficiency. For instance, for defects predicted as transient contaminants, the verification duration can be set to several seconds; for defects predicted as structural defects, the verification duration can be set to tens of seconds or even several minutes. Within the verification duration, continuous acquisition of images or 3D topographic data of the local area to be analyzed refers to the continuous or periodic acquisition of data from the local area using imaging or 3D scanning equipment within the set verification duration. This ensures comprehensive monitoring of the dynamic changes in the area. Based on the image or 3D topographic data, real-time calculation of the fluctuation range of the geometric and optical response characteristics of the local area to be analyzed refers to analyzing the continuously acquired data to extract the changes in the geometric features (such as size, shape, curvature) and optical response characteristics (such as brightness, contrast, color) of the local area over time, and calculating the fluctuation range or standard deviation of these characteristics. Comparing the fluctuation amplitude with stability parameters to adjust the discrimination criteria involves comparing the real-time calculated characteristic fluctuation amplitude with preset stability parameters. If the fluctuation amplitude exceeds the allowable range of the stability parameters, it indicates that the region may not be a stable, real surface defect, or its type needs to be reassessed. Based on this comparison result, subsequent defect discrimination thresholds or criteria can be dynamically adjusted to better reflect the current observed situation. Confirming whether the local area to be analyzed in detail is a persistent surface defect based on the adjusted discrimination criteria involves reassessing the characteristics of that local area after the discrimination criteria have been dynamically adjusted.If the features of the area are still judged as defects under the adjusted criteria, and the fluctuation range is within an acceptable range, then it can be finally confirmed as a persistent surface defect.

[0075] This application's solution effectively addresses the potential for misjudgment during initial identification by introducing a continuous verification mechanism. Specifically, when a local area is initially identified as a surface defect, the system does not immediately confirm it as a defect. Instead, it first predicts the defect type based on its preliminary characteristics. Because different defect types have different durations and stability characteristics, the system can selectively acquire corresponding duration ranges and stability parameters, and set a reasonable verification duration accordingly. Within this verification duration, by continuously acquiring images or 3D topographic data, the system can monitor the dynamic changes in the geometric and optical response characteristics of the local area in real time. By comparing the fluctuation amplitude of these characteristics with preset stability parameters, the system can determine whether the anomaly is a transient disturbance or a stable, persistent real defect. If the fluctuation amplitude is too large, it indicates that the anomaly may not be a real surface defect, allowing the system to adjust the discrimination criteria and avoid false alarms. This ability to dynamically adjust the discrimination criteria enables the detection system to adapt to various transient disturbances that may exist in actual production environments, thereby improving the robustness and accuracy of defect detection.

[0076] In some preferred embodiments, assuming that during surface defect detection on a transparent prototype model, the system initially identifies a suspected micro-dimple in a local area. First, based on the pit's preliminary geometric features (e.g., diameter approximately 0.1 mm, depth approximately 0.02 mm) and optical response characteristics (e.g., local brightness slightly lower than the surrounding area), the system predicts the defect type might be a "micro-dimple." Next, the system queries a pre-defined defect database to obtain the duration range (e.g., typically stable within 5 to 60 seconds) and stability parameters corresponding to the "micro-dimple" type defect (e.g., geometric feature fluctuations should not exceed 5%, optical response fluctuations should not exceed 10%). Based on the duration range, the system sets the verification time to 10 seconds. During the next 10 seconds, the detection system continuously acquires high-resolution images of the local area at a rate of 5 frames per second. During the acquisition process, the system analyzes these images in real time, calculating the fluctuation range of features such as the suspected pit's diameter, depth (using 3D topographic data), and local brightness. For example, if, during the verification period, the diameter of a localized area fluctuates by 2%, the depth by 3%, and the local brightness by 4%, all within the preset stability parameters (5% and 10%), the system will maintain or fine-tune the discrimination criteria. Ultimately, based on the adjusted criteria, the system confirms the localized area as a persistent "micro-pit" surface defect and records it in the inspection report. Conversely, if the brightness fluctuation of the localized area suddenly increases to 20% during the verification period, far exceeding the 10% stability parameter, the system will determine that this may be a transient change in light and shadow or dust adhesion, rather than a true persistent pit, and thus adjust the discrimination criteria to exclude it from the defect list.

[0077] This application further proposes a step for predicting the defect type of a local area to be analyzed in detail based on the initially identified defect characteristics, including: Spectral and polarization characteristics were obtained by analyzing local areas that were initially identified as surface defects and required further detailed analysis. The spectral and polarization features are matched with a pre-defined database of surface defect spectral and polarization features and a database of internal optical effects spectral and polarization features to obtain the matching results. Based on the matching results, determine whether the spectral and polarization characteristics are closer to surface defect characteristics or internal optical effect characteristics, and combine them with the initially identified defect characteristics to predict the defect type of the local area to be analyzed in detail.

[0078] Specifically, analyzing the spectral and polarization characteristics of a localized area initially identified as a surface defect for further detailed analysis involves using spectral analysis equipment (e.g., a spectrometer) and polarization analysis equipment (e.g., a polarization camera or polarimeter) to collect information on the absorption, reflection, and transmission characteristics of light at different wavelengths in the localized area to be analyzed, as well as the changes in the polarization state of the light. Spectral characteristics can be understood as the material's response to light at different wavelengths, such as absorption peaks and transmittance curves, aiming to reveal the effect of the material's chemical composition or microstructure on light. Polarization characteristics can be understood as the changes in parameters such as the polarization direction and degree of polarization of light after interaction with the material, aiming to reflect the material's anisotropy, surface roughness, or internal stress.

[0079] The pre-defined surface defect spectral polarization feature library and internal optical effect spectral polarization feature library are databases established through experimental measurements and data analysis based on a large number of known defect samples and internal optical effect samples. These feature libraries store the standard spectral and polarization characteristics of various typical surface defects (such as scratches, pits, and cracks) and internal optical effects (such as bubbles, impurities, and stress lines) under different lighting and observation conditions. For example, surface scratches may lead to enhanced light scattering and reduced polarization in a specific direction, while internal bubbles may exhibit specific spectral absorption or scattering modes and affect the polarization state of transmitted light.

[0080] In practical applications, the acquired spectral and polarization features are matched with the aforementioned feature library. Various pattern recognition or machine learning algorithms, such as support vector machines, neural networks, and correlation coefficient matching, can be employed to quantify the similarity between the features of the local region to be analyzed and various defects or effects in the feature library, thus obtaining the matching results. Based on the matching results, it is determined whether the spectral and polarization features are closer to surface defect features or internal optical effect features. Combined with the initially identified defect features, the defect type of the local region to be analyzed is pre-judged. For example, if the matching results show that the spectral and polarization features highly match the features of an "internal bubble," and the initially identified defect features (such as circular outlines and internal uniformity) also support this judgment, then the local region can be pre-judged as an internal bubble.

[0081] This application's approach, by introducing spectral and polarization features, enables the analysis of local regions of transparent or semi-transparent prototype models from a deeper optical physics perspective. Surface defects typically alter light scattering and reflection characteristics, while internal optical effects more significantly influence light transmission, absorption, and polarization states. Spectral features reveal the material's selective response to different wavelengths of light, thus distinguishing between optical effects caused by inherent material properties or internal impurities and defects caused by surface morphology variations. Polarization features further provide information about surface microstructure, material anisotropy, or internal stress distribution, which is crucial for differentiating surface scratches and cracks from internal stress lines and birefringence effects. By matching these multi-dimensional optical features with a pre-defined feature library and combining them with previously obtained geometric features and conventional optical response features, the accuracy and reliability of defect type prediction can be significantly improved.

[0082] In some preferred embodiments, it is assumed that an anomaly is found in a local area during the initial inspection of a transparent prototype model. This anomaly is initially identified as a surface defect and requires further verification. To more accurately predict its defect type, this embodiment employs the following steps: First, the local area is scanned using a hyperspectral imaging system and a polarization camera to acquire its spectral reflectance curves, transmittance curves, and light intensity distribution and polarization degree information in the visible and near-infrared bands. For example, if the spectral reflectance of this area shows an abnormal peak within a specific wavelength range, and the polarization degree of the transmitted light is significantly reduced, this may indicate the presence of some kind of internal scatterer. Next, these acquired spectral and polarization features are matched with a pre-established feature library. This feature library contains spectral polarization fingerprints of various defects and effects such as known surface scratches, pits, internal bubbles, internal impurities, and stress patterns. Through a matching algorithm (e.g., calculating the Euclidean distance or cosine similarity of feature vectors), it is determined that the features of this anomaly area have a 95% similarity to the "internal bubble" feature and a 60% similarity to the "surface scratch" feature. Simultaneously, combining the geometric features of the abnormal region obtained in previous steps (e.g., exhibiting an approximately spherical outline with blurred edges), the local area was ultimately predicted to be an "internal bubble." This prediction based on multi-dimensional optical features significantly improves the ability to identify internal defects in transparent prototype models, avoiding misjudging them as surface defects, thereby guiding subsequent quality control processes to take targeted measures.

[0083] This application further proposes an optimization scheme, which includes the following steps: During the acquisition of spectral and polarization characteristics, the ambient temperature and humidity information of the local area to be analyzed in detail are acquired simultaneously. Based on ambient temperature and humidity information, the preset surface defect spectral polarization feature library and the internal optical effect spectral polarization feature library are corrected in real time to obtain the corrected feature library. The spectral and polarization features are matched with the corrected feature library to obtain the initial matching degree; Based on the initial matching degree, calculate the degree of difference between the spectral features and polarization features and the corrected feature library; The matching threshold is dynamically adjusted based on the degree of difference. Based on the dynamically adjusted matching threshold, it is determined whether the spectral features and polarization features are closer to surface defect features or internal optical effect features.

[0084] Specifically, during the acquisition of spectral and polarization features, the simultaneous acquisition of ambient temperature and humidity information for the local area to be analyzed in detail refers to using temperature and humidity sensors to collect real-time data on the ambient temperature and relative humidity of the local area while performing spectral and polarization measurements. Ambient temperature information can be understood as the temperature of the air surrounding the local area, and ambient humidity information as the relative humidity of the air surrounding the local area. The purpose is to provide necessary environmental parameters for subsequent feature library correction.

[0085] This process involves real-time correction of a pre-established spectral polarization feature library for surface defects and an internal optical effect spectral polarization feature library based on ambient temperature and humidity information. The resulting corrected feature library utilizes real-time ambient temperature and humidity data to modify a pre-established spectral polarization feature library of surface defects and internal optical effects acquired under standard environmental conditions. This modification can be based on a pre-established optical response model of the material under different temperatures and humidity levels. For example, it can be achieved through interpolation, regression analysis, or table lookup to adjust the shape, peak position, intensity, and polarization state parameters of the spectral curves in the feature library, making them more consistent with the current actual environmental conditions, thus obtaining a more representative corrected feature library.

[0086] In practical applications, matching spectral and polarization features with a corrected feature library to obtain the initial matching degree involves comparing the real-time spectral and polarization features of the local area to be analyzed with the environmentally corrected feature library and calculating the similarity or distance between them to quantify the degree of matching. For example, Euclidean distance, cosine similarity, or correlation coefficient can be used to represent the initial matching degree.

[0087] Furthermore, based on the initial matching degree, the degree of difference between the spectral and polarization features and the corrected feature library is calculated. This refers to assessing the deviation between the currently measured features and the features of each defect type in the corrected feature library, based on the initial matching degree. The degree of difference can be understood as follows: the lower the matching degree, the greater the difference, and vice versa. The purpose is to provide a quantitative basis for dynamically adjusting the matching threshold.

[0088] Based on this, the matching threshold is dynamically adjusted according to the degree of difference. This means that the matching threshold used to determine the defect type is adjusted in real time according to the calculated degree of difference. For example, when the degree of difference is large, the matching threshold can be appropriately relaxed to avoid missed detections; when the degree of difference is small, the matching threshold can be appropriately tightened to improve recognition accuracy and avoid false judgments. This dynamic adjustment mechanism enables the detection system to better adapt to actual working conditions.

[0089] Finally, based on the dynamically adjusted matching threshold, it is determined whether the spectral features and polarization features are closer to surface defect features or internal optical effect features. This means matching the features of the local area to be analyzed with the corrected feature library, and determining the degree of similarity between them and surface defect features or internal optical effect features based on the dynamically adjusted matching threshold, thereby making a more accurate prediction of the defect type.

[0090] In some preferred embodiments, it is assumed that during a certain detection process, the ambient temperature of the local area to be analyzed is 28°C and the relative humidity is 70%. The detection system acquires the spectral and polarization characteristics of this local area simultaneously with these environmental data. Based on a pre-established material optical response model, the system uses the 28°C and 70% temperature and humidity data to perform real-time correction on a pre-set database of surface defect spectral polarization characteristics and a database of internal optical effects, generating a corrected feature library suitable for the current environmental conditions. Subsequently, the acquired spectral and polarization characteristics are matched with the corrected feature library to obtain an initial matching degree. For example, if the initial matching degree shows a high similarity to a certain surface defect characteristic but a low similarity to an internal optical effect characteristic, the system calculates the degree of difference based on this initial matching degree. If the degree of difference indicates a certain degree of uncertainty in the matching result, the system dynamically adjusts the matching threshold, for example, by appropriately tightening or loosening the threshold, to ensure that under the current environmental conditions, it can more accurately determine whether the defect type in this local area is a surface defect or an internal optical effect. In this way, even if environmental conditions change, the detection system can maintain a high accuracy in defect prediction.

[0091] This application further proposes the following steps for obtaining the corrected feature library: During the acquisition of spectral and polarization features, the ambient temperature, ambient humidity, and production batch information of the local area to be analyzed are acquired simultaneously. Based on ambient temperature information, ambient humidity information, and the current production batch information of the prototype model, the calibration dataset that best matches the current conditions is queried from a pre-established database containing spectral polarization response data of multiple batches of materials under different temperatures and humidity. Using the calibration dataset, the spectral curves and polarization state parameters in the preset surface defect spectral polarization feature library and the internal optical effect spectral polarization feature library are adjusted point by point to obtain the calibrated feature library.

[0092] Specifically, in the process of acquiring spectral and polarization characteristics, in addition to simultaneously acquiring the ambient temperature and humidity information of the local area to be analyzed, the production batch information of the current prototype model is also acquired simultaneously. The production batch information can be understood as data that identifies the unique or batch-specific characteristics of the prototype model, such as its production source, production time, and the batch of materials used. Its purpose is to distinguish potential material or process differences between different batches of prototype models.

[0093] Furthermore, based on the acquired ambient temperature and humidity information, as well as the production batch information of the current prototype model, the system queries a pre-established database. This database contains spectral polarization response data of multiple batches of materials under different temperatures and humidity levels, aiming to provide a comprehensive calibration reference that considers material batches and environmental factors. Through this query, the system can find the calibration dataset that best matches the current testing conditions (including ambient temperature, humidity, and production batch).

[0094] Subsequently, using this calibration dataset, the spectral curves and polarization state parameters in the preset surface defect spectral polarization feature library and the internal optical effect spectral polarization feature library were adjusted point by point. Point-by-point adjustment refers to fine-tuning each wavelength or each polarization state parameter in the feature library to ensure that the calibrated feature library can more accurately reflect the actual optical response of the prototype model under the current detection conditions.

[0095] In some preferred embodiments, suppose a prototype factory produces two batches of prototype models, batch A and batch B. Although these two batches are similar in appearance, they have slight differences in spectral absorption characteristics and polarization response at specific wavelengths due to the use of raw materials from different suppliers. For example, under ambient temperature of 25°C and humidity of 60%, the spectral peak intensity of a certain defect feature in batch A at 550 nm is X, while the same defect feature in batch B under the same conditions may be X-ΔX.

[0096] If the system is corrected based solely on ambient temperature and humidity, it may misclassify defect features of batch B as non-defects, or misclassify non-defect areas as defects.

[0097] The proposed solution simultaneously acquires the production batch information of the current prototype model. When the prototype model is detected to belong to batch B, the system retrieves a calibration dataset for batch B under these temperature and humidity conditions from a pre-established database, based on the ambient temperature of 25°C, humidity of 60%, and the production batch B information. This calibration dataset contains correction parameters for the material properties of batch B; for example, it might instruct an adjustment of -ΔX to the spectral peak intensity at 550nm in a pre-defined feature library. Using this calibration dataset, the pre-defined surface defect spectral polarization feature library is adjusted point-by-point to accurately match the actual optical response of the batch B prototype model. Therefore, even if the defect features of batch B exhibit different spectral intensities than those of batch A, the batch-specific calibration feature library can accurately identify the defect, thus avoiding misjudgment and ensuring the accuracy and reliability of the detection.

[0098] refer to Figure 2 This application proposes an image analysis-based prototype model surface defect detection system, applied to the aforementioned image analysis-based prototype model surface defect detection method. The system includes: The acquisition module acquires image information of the transparent or semi-transparent hand model to be detected. The image information is obtained by changing the pose and / or lighting method of the hand model. The recognition module compares and analyzes local regions in the image information, obtains comparison and analysis results, and identifies local regions in the image that differ from the background lighting and shadow changes as abnormal regions. The module identifies the local area to be analyzed in detail based on the abnormal area. The analysis module performs multi-faceted feature analysis on the local area to be analyzed in detail in order to identify whether the local area to be analyzed in detail is a surface defect. The multi-faceted feature analysis includes: evaluating the light response of the local area to be analyzed in detail under different lighting conditions, analyzing the shape features of the local area to be analyzed in detail, and judging the image position change of the local area to be analyzed in detail when the pose of the prototype model changes. The output module outputs surface defect detection information of the prototype model based on the surface defect identification results.

[0099] Specifically, the acquisition module can be understood as the hardware and software components responsible for collecting visual data from the prototype model. For example, it may include one or more high-resolution industrial cameras, a six-axis robot or precision turntable for precisely controlling the prototype model's posture, and a programmable lighting system (such as a ring light source, backlight source, point light source, etc.). Its purpose is to maximize the capture of all potential defect information on the prototype model's surface through image acquisition under multiple angles and lighting conditions. Especially for transparent or translucent materials, changing the posture and lighting methods can effectively reveal subtle internal or surface anomalies.

[0100] The identification module can be understood as a unit that performs preliminary processing and analysis of the image information acquired by the acquisition module. This module can integrate image processing algorithms, such as anomaly detection algorithms based on background modeling, image differencing, texture analysis, or deep learning. Its purpose is to quickly filter out potentially defective anomaly areas by comparing the differences between the lighting, color, and texture features of local areas in the image and the expected background, providing targets for subsequent detailed analysis.

[0101] In practical applications, the determination module further filters and locates local areas requiring detailed analysis based on the abnormal regions output by the identification module. This module can employ image processing techniques such as region growing, connected component analysis, and morphological operations to aggregate, denoise, and optimize the boundaries of the identified abnormal regions, thereby accurately delineating the areas requiring detailed examination. Its purpose is to reduce the computational burden of subsequent complex analyses, improve detection efficiency, and ensure that the focus of analysis is concentrated on the most valuable areas.

[0102] Furthermore, the analysis module specifically performs in-depth feature extraction and defect identification on the local area to be analyzed in detail, as determined by the determination module. This module can contain multiple sub-modules, each responsible for: evaluating the light response of the local area to be analyzed in detail under different lighting conditions: for example, by analyzing the reflection, transmission, and scattering characteristics of the local area under different lighting conditions such as diffuse light, direct light, and polarized light, to distinguish different types of defects such as surface scratches, bubbles, and foreign objects. Analyzing the shape characteristics of the local area to be analyzed in detail: for example, calculating the geometric parameters of the local area such as area, perimeter, aspect ratio, roundness, and convexity, which are important for distinguishing point defects, line defects, and surface defects. Determining the image position change of the local area to be analyzed in detail when the pose of the prototype model changes: for example, by tracking the position and deformation of the local area in images of different poses, it is possible to effectively distinguish between real physical defects and pseudo-defects caused by lighting and viewing angle.

[0103] Furthermore, the output module is specifically responsible for presenting the analysis module's findings in a user-friendly manner. For example, this module can generate detailed inspection reports, including information such as defect type, location coordinates, size, and severity; it can also annotate and visualize defects on the digital model or real-time image of the prototype; or it can interact with other production management systems through audible and visual alarms, data interfaces, and other means. Its purpose is to provide quality control personnel with clear and intuitive defect information so that corrective actions can be taken promptly.

[0104] The above-disclosed content is only a preferred and feasible embodiment of the present invention, and is not intended to limit the scope of protection of the present invention. Therefore, all equivalent technical changes made based on the content of the present invention specification and drawings are included within the scope of protection of the present invention. Furthermore, the elements therein can be updated as technology develops.

Claims

1. A method for detecting surface defects in a prototype model based on image analysis, characterized in that, The method includes the following steps: Acquire image information of the transparent or semi-transparent hand model to be detected. The image information is obtained by changing the pose and / or lighting method of the hand model. By comparing and analyzing local regions in the image information, the results of the comparison and analysis are obtained. Based on the results of the comparison and analysis, local regions in the image that differ from the changes in light and shadow of the background are identified as abnormal regions. Based on the abnormal areas, determine the local areas that need to be analyzed in detail; Multi-faceted feature analysis is performed on the local area to be analyzed in detail to identify whether the local area to be analyzed in detail is a surface defect. The multi-faceted feature analysis includes: evaluating the light response of the local area to be analyzed in detail under different lighting conditions, analyzing the shape features of the local area to be analyzed in detail, and judging the image position change of the local area to be analyzed in detail when the pose of the prototype model changes. Based on the identification results of surface defects, the surface defect detection information of the prototype model is output.

2. The method for detecting surface defects in a prototype model based on image analysis as described in claim 1, characterized in that, The steps to determine whether a local area to be analyzed in detail is a surface defect include: A multi-frequency structured light pattern is projected onto the local area to be analyzed in detail, and a sequence of structured light images of the local area to be analyzed in detail is obtained. Obtain vibration data from the testing platform; Displacement compensation is performed on structured light image sequences using vibration data; Phase calculation is performed on the compensated structured light image sequence to obtain the phase calculation results; The phase calculation results are converted into three-dimensional surface topography data of the local area to be analyzed in detail; Local curvature and depth variation information are extracted from the three-dimensional surface topography data; Based on local curvature and depth variation information, determine whether the local area to be analyzed in detail is a surface defect.

3. The method for detecting surface defects in a prototype model based on image analysis as described in claim 2, characterized in that, The steps for obtaining vibration data from the testing platform include: A vibration sensing unit is installed on the testing platform; During the acquisition of structured light image sequences, vibration signals output by the vibration sensing unit are simultaneously acquired. Vibration data is obtained by time alignment and data preprocessing of the vibration signal.

4. The method for detecting surface defects in a prototype model based on image analysis as described in claim 2, characterized in that, The steps for extracting local curvature information from surface three-dimensional topography data include: Multi-scale geometric feature analysis is performed on the local area to be analyzed in detail based on the surface three-dimensional topography data. The surface roughness parameters of the local area to be analyzed in detail are calculated at different spatial scales, and the local curvature parameters of the local area to be analyzed in detail are calculated at different spatial scales. The relative magnitudes and distribution patterns of surface roughness parameters and local curvature parameters at different spatial scales are compared to obtain the comparison results. Based on the comparison results, it is determined whether the local area to be analyzed in detail has the curvature characteristics of surface defects, and the local curvature information is extracted based on the determination results.

5. The method for detecting surface defects in a prototype model based on image analysis as described in claim 4, characterized in that, The steps for comparing the relative magnitudes and distribution patterns of surface roughness parameters and local curvature parameters at different spatial scales include: Establish a local geometric feature reference file for the local area to be analyzed in detail on the surface three-dimensional topography data; Update the local geometric feature baseline file using the exponentially weighted moving average method; Compare the expected fluctuation range of surface roughness parameters and local geometric feature reference files at different spatial scales, compare the expected fluctuation range of local curvature parameters and local geometric feature reference files at different spatial scales, and preliminarily identify whether the local area to be analyzed in detail is a surface defect based on the expected fluctuation range. Continuous verification is carried out on local areas that are initially identified as surface defects and require detailed analysis.

6. The method for detecting surface defects in a prototype model based on image analysis as described in claim 5, characterized in that, The steps for continuous verification of local areas initially identified as surface defects and requiring further detailed analysis include: Based on the initially identified defect characteristics, the defect type of the local area to be analyzed in detail is predicted; Based on the predicted defect type, obtain the duration range and stability parameters corresponding to the defect type; Set the verification duration based on the duration range; During the verification period, images or three-dimensional topographic data of the local area to be analyzed in detail are continuously collected. Based on image or 3D topography data, calculate in real time the fluctuation amplitude of geometric features and optical response features of the local area to be analyzed in detail; The fluctuation amplitude is compared with the stability parameter, thereby adjusting the judgment criteria; Based on the adjusted criteria, confirm whether the local area to be analyzed in detail is a persistent surface defect.

7. The method for detecting surface defects in a prototype model based on image analysis as described in claim 6, characterized in that, Based on the initially identified defect characteristics, the steps for predicting the defect type of the local area to be analyzed in detail include: Spectral and polarization characteristics were obtained by analyzing local areas that were initially identified as surface defects and required further detailed analysis. The spectral and polarization features are matched with a pre-defined database of surface defect spectral and polarization features and a database of internal optical effects spectral and polarization features to obtain the matching results. Based on the matching results, determine whether the spectral and polarization characteristics are closer to surface defect characteristics or internal optical effect characteristics, and combine them with the initially identified defect characteristics to predict the defect type of the local area to be analyzed in detail.

8. The method for detecting surface defects in a prototype model based on image analysis as described in claim 7, characterized in that, The method also includes the following steps: During the acquisition of spectral and polarization characteristics, the ambient temperature and humidity information of the local area to be analyzed in detail are acquired simultaneously. Based on ambient temperature and humidity information, the preset surface defect spectral polarization feature library and the internal optical effect spectral polarization feature library are corrected in real time to obtain the corrected feature library. The spectral and polarization features are matched with the corrected feature library to obtain the initial matching degree; Based on the initial matching degree, calculate the degree of difference between the spectral features and polarization features and the corrected feature library; The matching threshold is dynamically adjusted based on the degree of difference. Based on the dynamically adjusted matching threshold, it is determined whether the spectral features and polarization features are closer to surface defect features or internal optical effect features.

9. The method for detecting surface defects in a prototype model based on image analysis as described in claim 8, characterized in that, The steps to obtain the corrected feature library include: During the acquisition of spectral and polarization features, the ambient temperature, ambient humidity, and production batch information of the local area to be analyzed are acquired simultaneously. Based on ambient temperature information, ambient humidity information, and the production batch information of the current prototype model, the calibration dataset that best matches the current conditions is queried from a pre-established database containing spectral polarization response data of multiple batches of materials under different temperatures and humidity. Using the calibration dataset, the spectral curves and polarization state parameters in the preset surface defect spectral polarization feature library and the internal optical effect spectral polarization feature library are adjusted point by point to obtain the calibrated feature library.

10. A prototype model surface defect detection system based on image analysis, applied to the prototype model surface defect detection method based on image analysis as described in claim 1, characterized in that, The system includes: The acquisition module acquires image information of the transparent or semi-transparent hand model to be detected. The image information is obtained by changing the pose and / or lighting method of the hand model. The recognition module compares and analyzes local regions in the image information, obtains comparison and analysis results, and identifies local regions in the image that differ from the background lighting and shadow changes as abnormal regions. The module identifies the local area to be analyzed in detail based on the abnormal area. The analysis module performs multi-faceted feature analysis on the local area to be analyzed in detail in order to identify whether the local area to be analyzed in detail is a surface defect. The multi-faceted feature analysis includes: evaluating the light response of the local area to be analyzed in detail under different lighting conditions, analyzing the shape features of the local area to be analyzed in detail, and judging the image position change of the local area to be analyzed in detail when the pose of the prototype model changes. The output module outputs surface defect detection information of the prototype model based on the surface defect identification results.

Images