A multi-angle automatic detection system and method for mobile phone appearance based on multi-faceted reflection structure

Abstract

This invention relates to the field of physical property testing technology, and discloses an automatic multi-angle inspection system and method for mobile phone appearance based on a multi-faceted reflection structure. The system includes: a panoramic acquisition unit for acquiring the original panoramic image formed by the mobile phone under test through a multi-faceted panoramic reflection mechanism; a reference data storage unit for storing standardized three-dimensional model data of the mobile phone under test; a central processing unit for establishing the three-dimensional model data as a spatial measurement reference, constructing an optical path mapping matrix based on distribution parameters, mapping the surface mesh of the three-dimensional model data to the camera coordinate system to determine geometric distortion sampling coordinates, and then extracting pixels to complete spatial topology reconstruction and generate a standardized surface unfolded map; and a defect detection module for analyzing the features in the surface unfolded map. This invention eliminates nonlinear perspective distortion in reflective imaging, achieves tight alignment between pixel sites and physical topology, and provides a physical standard with an objective scale for defect quantitative analysis.

Description

Technical Field

[0001] This invention relates to the field of physical property testing technology, specifically to an automatic multi-angle inspection system and method for mobile phone appearance based on a multi-faceted reflective structure. Background Technology

[0002] Currently, multi-angle imaging technology is a mainstream approach for determining detection schemes for scratches and dents on the surface of used mobile phones. In this approach, the detection system typically employs a multi-faceted reflective optical structure to acquire a panoramic image of all surfaces of the phone in a single pass, improving detection efficiency and ensuring the integrity of the detection coverage. Optical reflection imaging follows the principle of perspective projection, where surface features are projected onto the image sensor array via reflective surfaces. The image quality depends on the consistency of the optical path and the linearity of the object-image relationship. However, as detection precision evolves towards the micrometer level, complex morphological features such as the curved edges and four beveled surfaces of mobile phones, when projected onto the two-dimensional image sensor via reflective surfaces at specific angles, produce nonlinear topological distortion and resolution gradient changes. The pursuit of efficiency advantages in panoramic imaging is accompanied by information distortion during the object-image spatial transformation process, making it impossible for the photoelectric signals collected by the sensor to directly reflect the true physical scale of the measured target, thus forming an inherent constraint on high-precision physical measurement.

[0003] Existing improvement approaches often focus on increasing the number of image acquisition devices or increasing the pixel scale of sensors. While these approaches can alleviate clarity issues in some areas, they cannot eliminate physical scale distortion caused by optical path transformations at the physical mechanism level. This design approach simplifies the measurement problem in three-dimensional physical space to a recognition task at the two-dimensional image level, stripping away the true geometric spatial attributes of the target. This results in the detection system being unable to maintain a constant physical measurement benchmark in distorted areas. For example, Chinese invention patent CN112634173B discloses a curved screen image correction method and system. It extracts feature points by displaying a checkerboard pattern on the screen and uses coordinate mapping to generate a correction matrix to complete image stitching. However, the solution is based on two-dimensional image feature matching and presupposes linear projection relationships between objects. When the detection system evolves towards micron-level precision and undergoes complex optical path transitions through multi-faceted precision reflection structures, two-dimensional visual calibration is difficult to compensate for the nonlinear topological deformation of the curved surface, causing the physical scale to shift in distorted areas. The solution has strict requirements for the consistency of the measured target's posture and lacks a measurement benchmark coupled with a physical entity CAD model, making it unable to maintain a constant physical standard under complex working conditions.

[0004] Therefore, the technical problem to be solved by this invention is how to construct a measurement benchmark that can eliminate interference from complex reflected light paths and restore the true physical topology of the mobile phone, so as to achieve objective physical property characterization of minute defects. Summary of the Invention

[0005] This invention proposes an automatic multi-angle inspection system for mobile phone appearance based on a multi-faceted reflection structure, comprising:

[0006] The panoramic acquisition unit is used to acquire the original panoramic image formed by the mobile phone under test through the multi-faceted panoramic reflection mechanism;

[0007] The baseline data storage unit is used to store the standardized three-dimensional geometric model data of the mobile phone under test;

[0008] The central processing unit includes: a spatial mapping module, which is used to establish standardized three-dimensional geometric model data as a spatial measurement benchmark, and construct an optical path mapping matrix between each sampling pixel site and a physical topological site based on the geometric distribution parameters of the reflective surface of the multi-faceted panoramic reflection mechanism and the optical center position parameters of the panoramic acquisition unit;

[0009] The distortion correction module is used to inversely project the surface mesh of the standardized 3D geometric model data to the camera coordinate system through the optical path mapping matrix, and calculate the geometric distortion sampling coordinates of each node of the surface mesh;

[0010] The image reconstruction module is used to extract pixels from the original panoramic image and perform spatial topology reconstruction based on the geometric distortion sampling coordinates, and output a standardized surface unfolding map that eliminates nonlinear perspective distortion and is closely aligned with the physical three-dimensional structure of the mobile phone under test.

[0011] The defect detection module is used to extract defect features from the standardized surface unfolded diagram and determine the surface quality level of the mobile phone under test based on the physical size and distribution density of the defect features.

[0012] Preferably, the central control unit is also used to perform illuminance normalization correction on the standardized surface unfolded map: using the spatial optical path mapping matrix, the cosine value of the angle between the normal vector corresponding to each sampled pixel and the optical axis is calculated; the brightness gain coefficient of each sampled pixel is determined according to the cosine value of the angle, and the brightness correction mapping is performed on the pixels extracted from the original panoramic image to eliminate the local illuminance difference inside the multi-faceted panoramic reflection device, so that the pixel gray value in the output standardized surface unfolded map represents the physical reflectivity characteristics of the surface of the mobile phone under test.

[0013] Preferably, the multi-faceted panoramic reflection device includes an eight-faceted precision reflection structure; the eight-faceted precision reflection structure is made of optical-grade float glass, and the included angle between adjacent reflection surfaces is in the range of 78° to 90°.

[0014] Preferably, the system also includes a uniform illumination module; the uniform illumination module is used to provide 360° surround illumination for the mobile phone under test, so as to cooperate with the multi-faceted panoramic reflection device to realize all-round imaging of the front, back, four edges and four bevels of the mobile phone under test. When identifying defect features, the central control unit performs regional cropping of the original panoramic image through the image segmentation and reconstruction module, and feeds the cropped feature tensor into the surface feature extraction module to identify five types of physical defects: scratches, bumps, cracks, wear and discoloration.

[0015] Preferably, the system also includes a multi-interface data reading module; the multi-interface data reading module is used to read the hardware parameters and security verification information of the mobile phone under test; the central control unit associates the hardware parameters, security verification information and surface quality grading data to generate a comprehensive test report.

[0016] Preferably, the central control unit also integrates a data synchronization interface, which is used to send the comprehensive test report to the background management system in real time via API interface, and generate recycling pricing data according to the preset residual value calculation rules.

[0017] Preferably, when constructing the spatial optical path mapping matrix, the central control unit extracts the coordinates of the edge calibration points of each reflective surface in the original panoramic image to complete the geometric calibration based on spatial homography transformation, so as to correct the physical position deviation of the multi-faceted panoramic reflective device.

[0018] Preferably, the reference data storage unit stores a standardized three-dimensional topology library of different mobile phone models; the central control unit reads the device identifier of the mobile phone under test and automatically matches the corresponding model data from the standardized three-dimensional topology library to change the measurement reference system in real time.

[0019] A method for automatic multi-angle detection of mobile phone appearance based on multi-faceted reflective structure includes the following steps:

[0020] The panoramic acquisition unit acquires the original panoramic image formed by the mobile phone under test through the multi-faceted panoramic reflection mechanism; the standardized three-dimensional geometric model data of the mobile phone under test is retrieved from the reference data storage unit, and the standardized three-dimensional geometric model data is established as the spatial measurement reference.

[0021] The central processing unit calculates the optical path mapping matrix between each sampled pixel in the original panoramic image and the physical topological position in the spatial measurement reference, based on the geometric distribution parameters of the reflective surface of the multi-faceted panoramic reflection mechanism and the optical center position parameters of the panoramic acquisition unit, using the law of light reflection.

[0022] The central processing unit inversely projects the surface mesh of the standardized 3D geometric model data to the camera coordinate system through the optical path mapping matrix, and determines the geometric distortion sampling coordinates of each node of the surface mesh;

[0023] Based on the geometric distortion sampling coordinates, the central processing unit extracts pixels from the original panoramic image and completes spatial topology reconstruction to generate a standardized surface unfolding map that eliminates perspective distortion and is aligned with the physical three-dimensional structure of the mobile phone under test.

[0024] The central processing unit extracts defect features from the standardized surface unfolded diagram and determines the surface quality level of the mobile phone under test based on the physical size and distribution density of the defect features.

[0025] The beneficial effects of this invention are:

[0026] 1. In the multi-angle automatic inspection system for mobile phone appearance, the influence of nonlinear topological distortion under complex optical paths on the accuracy of physical measurement is eliminated. Through a pre-set inverse mapping mechanism based on CAD prior topology, the nonlinear projection generated by the eight-sided precision reflection structure is restored to the real physical coordinate system. Since the pixel group in the panoramic image is inversely mapped back to the physical space according to the standard three-dimensional mesh model of the target being measured, the detection resolution of the arc-shaped area at the edge of the target being measured is no longer constrained by the sensor projection angle. This eliminates the physical scale distortion caused by perspective deformation in the conventional two-dimensional image cropping mode, ensuring the objectivity of the system in measuring microscopic physical defects at the 0.05mm level.

[0027] 2. To achieve low-level data coupling and attitude adaptation between the detection system and the physical hardware space, this invention deeply integrates the assembly angle parameters of the reflector with the front-end sampling logic through a reverse perspective projection matrix, constructing a monotonic deterministic correlation between the hardware physical space and the digital measurement space. When the target under test produces a small attitude shift within the detection area, the system calculates the real-time mapping of each sampling point of the 3D mesh model to the camera coordinate system, performs in-situ attitude compensation and topology unfolding, and ensures that every pixel sent to the recognition stage accurately corresponds to its actual physical size, avoiding sampling deviations caused by target pose changes in traditional solutions and improving the consistency of physical property evaluation under complex working conditions.

[0028] 3. A physical property decoupling mechanism is constructed to improve the feature representation quality of minor defects. The cosine value of the angle between the normal vectors corresponding to the mapping matrix is ​​used to perform brightness compensation on the sampled pixels. The light intensity attenuation law in the photoelectric acquisition process is separated from the real reflectivity characteristics of the material surface. This mechanism cancels the edge dark corners and uneven surface reflection common in multi-faceted reflective structures. The output is a topologically normalized expansion tensor with a constant 1:1 mapping relationship with the real physical scale and the elimination of environmental noise interference. This physical-level data reconstruction enables the subsequent analysis process to be directly executed based on the uniform parameters with objective scaling, avoiding the systematic risk of feature fitting of distorted images in the recognition process. Attached Figure Description

[0029] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0030] Figure 1 This is a flowchart illustrating the module operation and workflow of the multi-angle automatic detection system for mobile phone appearance with a multi-faceted reflective structure according to the present invention.

[0031] Figure 2 This is a functional interaction and business application logic diagram of the multi-angle automatic detection system for mobile phone appearance with multi-faceted reflective structure according to the present invention. Detailed Implementation

[0032] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0033] A multi-angle automatic inspection system for mobile phone appearance based on a multi-faceted reflective structure includes:

[0034] The panoramic acquisition unit is used to acquire the original panoramic image formed by the mobile phone under test through the multi-faceted panoramic reflection mechanism;

[0035] The baseline data storage unit is used to store the standardized three-dimensional geometric model data of the mobile phone under test;

[0036] The central processing unit includes: a spatial mapping module, which is used to establish standardized three-dimensional geometric model data as a spatial measurement benchmark, and construct an optical path mapping matrix between each sampling pixel site and a physical topological site based on the geometric distribution parameters of the reflective surface of the multi-faceted panoramic reflection mechanism and the optical center position parameters of the panoramic acquisition unit;

[0037] The distortion correction module is used to inversely project the surface mesh of the standardized 3D geometric model data to the camera coordinate system through the optical path mapping matrix, and calculate the geometric distortion sampling coordinates of each node of the surface mesh;

[0038] The image reconstruction module is used to extract pixels from the original panoramic image and perform spatial topology reconstruction based on the geometric distortion sampling coordinates, and output a standardized surface unfolding map that eliminates nonlinear perspective distortion and is closely aligned with the physical three-dimensional structure of the mobile phone under test.

[0039] The defect detection module is used to extract defect features from the standardized surface unfolded diagram and determine the surface quality level of the mobile phone under test based on the physical size and distribution density of the defect features.

[0040] Preferably, the central control unit is also used to perform illuminance normalization correction on the standardized surface unfolded map: using the spatial optical path mapping matrix, the cosine value of the angle between the normal vector corresponding to each sampled pixel and the optical axis is calculated; the brightness gain coefficient of each sampled pixel is determined according to the cosine value of the angle, and the brightness correction mapping is performed on the pixels extracted from the original panoramic image to eliminate the local illuminance difference inside the multi-faceted panoramic reflection device, so that the pixel gray value in the output standardized surface unfolded map represents the physical reflectivity characteristics of the surface of the mobile phone under test.

[0041] Preferably, the multi-faceted panoramic reflection device includes an eight-faceted precision reflection structure; the eight-faceted precision reflection structure is made of optical-grade float glass, and the included angle between adjacent reflection surfaces is in the range of 78° to 90°.

[0042] Preferably, the system also includes a uniform illumination module; the uniform illumination module is used to provide 360° surround illumination for the mobile phone under test, so as to cooperate with the multi-faceted panoramic reflection device to realize all-round imaging of the front, back, four edges and four bevels of the mobile phone under test. When identifying defect features, the central control unit performs regional cropping of the original panoramic image through the image segmentation and reconstruction module, and feeds the cropped feature tensor into the surface feature extraction module to identify five types of physical defects: scratches, bumps, cracks, wear and discoloration.

[0043] Preferably, the system also includes a multi-interface data reading module; the multi-interface data reading module is used to read the hardware parameters and security verification information of the mobile phone under test; the central control unit associates the hardware parameters, security verification information and surface quality grading data to generate a comprehensive test report.

[0044] Preferably, the central control unit also integrates a data synchronization interface, which is used to send the comprehensive test report to the background management system in real time via API interface, and generate recycling pricing data according to the preset residual value calculation rules.

[0045] Preferably, when constructing the spatial optical path mapping matrix, the central control unit extracts the coordinates of the edge calibration points of each reflective surface in the original panoramic image to complete the geometric calibration based on spatial homography transformation, so as to correct the physical position deviation of the multi-faceted panoramic reflective device.

[0046] Preferably, the calculation logic for the brightness gain coefficient G when the central control unit performs brightness correction mapping is as follows: , where θ is the angle between the normal vector corresponding to the sampled pixel and the optical axis of the panoramic photoelectric acquisition unit.

[0047] Preferably, the reference data storage unit stores a standardized three-dimensional topology library of different mobile phone models; the central control unit reads the device identifier of the mobile phone under test and automatically matches the corresponding model data from the standardized three-dimensional topology library to change the measurement reference system in real time.

[0048] A method for automatic multi-angle detection of mobile phone appearance based on multi-faceted reflective structure includes the following steps:

[0049] The panoramic acquisition unit acquires the original panoramic image formed by the mobile phone under test through the multi-faceted panoramic reflection mechanism; the standardized three-dimensional geometric model data of the mobile phone under test is retrieved from the reference data storage unit, and the standardized three-dimensional geometric model data is established as the spatial measurement reference.

[0050] The central processing unit calculates the optical path mapping matrix between each sampled pixel in the original panoramic image and the physical topological position in the spatial measurement reference, based on the geometric distribution parameters of the reflective surface of the multi-faceted panoramic reflection mechanism and the optical center position parameters of the panoramic acquisition unit, using the law of light reflection.

[0051] The central processing unit inversely projects the surface mesh of the standardized 3D geometric model data to the camera coordinate system through the optical path mapping matrix, and determines the geometric distortion sampling coordinates of each node of the surface mesh;

[0052] Based on the geometric distortion sampling coordinates, the central processing unit extracts pixels from the original panoramic image and completes spatial topology reconstruction to generate a standardized surface unfolding map that eliminates perspective distortion and is aligned with the physical three-dimensional structure of the mobile phone under test.

[0053] The central processing unit extracts defect features from the standardized surface unfolded diagram and determines the surface quality level of the mobile phone under test based on the physical size and distribution density of the defect features.

[0054] Example 1: When the system faces the objective condition of evaluating high-precision appearance defects in batches of used mobile phones, especially when dealing with terminal devices with large curvature four-sided bezels and four beveled surfaces, the panoramic imaging mode based on the multi-faceted reflection optical structure faces physical constraints of nonlinear perspective distortion and edge illumination attenuation. This constraint causes microscopic physical defects located in the corner areas of the mobile phone with a size close to 0.05mm to experience resolution compression and loss of brightness features on the two-dimensional sensor array. This makes the photoelectric signals collected by the sensor unable to directly reflect the true physical scale of the mobile phone under test, thus affecting the objectivity of the surface quality grade determination and the consistency rate of the color assessment. This invention provides a multi-angle automatic detection system and method for mobile phone appearance based on a multi-faceted reflection structure. It uses a preset three-dimensional geometric benchmark to reconstruct the distortion correction problem into a physical space topology restoration task. The spatial mapping module in the central processing unit establishes the standardized three-dimensional geometric model data of the mobile phone under test as the spatial measurement benchmark, and based on the reflection of the multi-faceted panoramic reflection mechanism composed of eight optical-grade float glass with included angles between 78° and 90°, it performs a multi-faceted panoramic reflection. The geometric distribution parameters of the surface and the optical center position parameters of the panoramic acquisition unit are used to construct an optical path mapping matrix between each sampling pixel point and the physical topological point using the law of light reflection. At the junction of adjacent panoramic views generated by the 8-surface precision reflection structure, an image overlap buffer with a width of 40 pixels is set. The image reconstruction module calculates the gradient difference of gray values ​​of adjacent views in the buffer and executes a linear weighted fusion algorithm based on distance ratio. The fusion weight coefficient gradually increases from 0 to 1 with the horizontal offset of the pixel point from the center line in steps of 0.025 per pixel, thereby eliminating physical scale jumps and pixel breakage distortion at the view seams. The distortion correction module reverses the surface mesh of the standardized 3D geometric model data to the camera coordinate system through the optical path mapping matrix to determine the geometric distortion sampling coordinates of each node of the surface mesh. The image reconstruction module extracts pixels from the acquired original panoramic image based on the geometric distortion sampling coordinates and completes spatial topological reconstruction. This reverse perspective unfolding mechanism based on CAD 3D prior topology solves the technical contradiction between single panoramic imaging efficiency and edge microscale fidelity.

[0055] In the spatial topology reconstruction operation sequence, for the microscopic sampling point mesh on the surface of the standardized three-dimensional geometric model data... The central processing unit is based on the calculation formula Determine its distortion mapping coordinates on the 2D panoramic image. ,in, The distortion mapping coordinates on the 2D panoramic image. This is the inverse perspective projection matrix from the three-dimensional world coordinate system to the two-dimensional image pixel coordinate system of the camera. This represents the coordinates of the microscopic sampling point grid nodes. Simultaneously, the central processing unit uses this optical path mapping matrix to calculate the cosine of the angle between the normal vector corresponding to each sampling pixel and the optical axis, and then calculates it according to the formula... The brightness gain coefficient G of each sampled pixel is determined, where G is the brightness gain coefficient and θ is the angle between the normal vector of the sampled pixel and the optical axis of the panoramic photoelectric acquisition unit. The central processing unit corrects the brightness features of the extracted pixels accordingly, outputting a standardized surface unfolded map that eliminates nonlinear perspective distortion and local illumination differences and is closely aligned with the physical three-dimensional structure of the phone under test. This standardized surface unfolded map is fed into the defect detection module as an input benchmark, guiding the Qwen3.5-Plus multimodal model embedded within the defect detection module to cluster defect feature vectors and determine physical boundaries on an objective data plane with a constant physical scale. The model employs a 3-layer deep convolutional processing architecture at its bottom layer. The first feature extraction layer is configured with 32 convolutional kernels of size 3 x 3 pixels, with a fixed stride of 2 pixels. The input image tensor is divided into 64-pixel x 64-pixel local sampling windows, with each adjacent window maintaining a 50° interval. The overlap rate is % to ensure defect connectivity; when the gradient change amplitude of a local pixel exceeds 40 units (corresponding to gray levels 0 to 255) and the continuous distribution length exceeds 5 pixels, the system determines that there are physical morphological abrupt changes such as scratches or bumps in the area. The system establishes a system-level synergistic benchmark for optical topology reconstruction and multimodal model boundary recognition. The defect detection module determines the surface quality level of the mobile phone under test based on the physical size and distribution density of the extracted defect features. The multi-interface data reading module synchronously reads the hardware parameters and security verification information of the mobile phone under test. The central control unit associates the hardware parameters, security verification information and surface quality grading data to generate a comprehensive test report. The comprehensive test report is sent to the background management system in real time through the integrated data synchronization interface using the API interface. The system finally outputs a standardized physical inspection file containing five-level color rating and recycling pricing data generated based on preset residual value calculation rules.

[0056] Example 2: Batch testing of used mobile phones in industrial settings faces physical constraints such as alternating ambient light interference and illuminance attenuation at curved edges. To determine the distribution pattern of surface physical defects, a measurement and testing platform is established, integrating a multi-faceted panoramic reflection mechanism composed of eight optical-grade float glass and an image sensor with a resolution of 20 million pixels. The test benchmark data comes from the standard three-dimensional geometric model data file published by the mobile phone manufacturer. The test environment actively injects flickering disturbances from a 50Hz power frequency light source with a light intensity fluctuation amplitude of 15%. Simultaneously, simulated obstructions with a transmittance fluctuation range of 85% to 95% are randomly applied to the four edges and four beveled areas of the mobile phone under test. The geometric distribution parameters of the reflective surfaces of the multi-faceted panoramic reflection mechanism are constrained by both the field of view overlap rate and the optical path depth. When the angle between adjacent reflective surfaces decreases, the... As the field of view increases, multiple reflections lead to a decrease in light flux. When the angle increases, the edge optical path exceeds the lens depth of field boundary, resulting in defocus blur. The system sets the angle of the reflective surface in the range of 78° to 90°. Under the operating parameters of 85°, the panoramic acquisition unit controls the edge feature pixel distortion rate to within 12%. The measurement and testing platform sets three detection comparison groups. The control group uses a fixed two-dimensional coordinate system to directly crop the original panoramic image and input it into the Qwen3.5-Plus multimodal model. The out-of-range control group sets the angle of the reflective surface to 70° and 100° respectively. The sample group of this invention uses an 85° angle of the reflective surface. The central processing unit extracts standardized three-dimensional geometric model data as a spatial measurement benchmark and constructs an optical path mapping matrix based on the geometric distribution parameters of the reflective surface and the optical center position parameters of the panoramic acquisition unit.

[0057] A gradient test sequence was designed based on the physical defect size. Standard physical notches ranging from 0.02mm to 0.1mm were implanted on the surface of the test benchmark prototype. The panoramic acquisition unit captured the original panoramic image under a 50Hz power frequency flicker environment. In the control group without the optical path mapping matrix enabled, a notch pixel with a size of 0.05mm located in a high-curvature side region experienced nonlinear perspective distortion compression. Its geometric width attenuated to 1.2 consecutive pixels on the sensor array, and the local brightness contrast of this image region decreased to 12. The defect detection rate output by the multimodal model was on the order of 34.2%. In the prototype of this invention, the central processing unit inversely projects the surface mesh of the standardized three-dimensional geometric model data to the camera coordinate system through the optical path mapping matrix. For the high-curvature side region, the central control unit calculated that the angle between the normal vector corresponding to the sampled pixel and the optical axis reached 60°. The central control unit then calculated the angle based on the formula... The brightness gain coefficient of the region is calculated, where G is a dimensionless brightness gain coefficient and θ is the angle between the normal vector corresponding to the sampled pixel and the optical axis of the panoramic photoelectric acquisition unit. The central control unit multiplies the original pixel gray value of 45 by the brightness gain coefficient to obtain a corrected pixel with a gray value of 90. The image reconstruction module extracts pixels from the original panoramic image based on the geometric distortion sampling coordinates and completes spatial topology reconstruction, outputting a standardized surface unfolded map.

[0058] The standardized surface unfolded map physically restores the pixel geometric width of a 0.05mm scribe line to 4.5 consecutive pixels, restoring the local brightness and contrast to 45. After receiving this standardized surface unfolded map, the defect detection module achieves a feature detection rate of 98.7%. For gradient size test data, the detection rate of the sample group maintained a stable range of 98.5% to 99.1% within the defect size range from 0.1mm to 0.05mm. When the defect size decreased to 0.03mm, the detection rate dropped to 61.4%. This data discontinuity indicates that 0.05mm reaches the physical diffraction limit of a 20-megapixel optical sensor at a specific depth of field, establishing 0.05mm as the physical lower limit boundary for the geometric topology reconstruction measurement of this system. Beyond this range... In the data comparison of the control group, the multiple reflections under the 70° angle configuration caused the light intensity signal-to-noise ratio in the edge area to drop to 8dB, and the detection rate to decrease to 55.3%. Under the 100° angle configuration, the optical path difference exceeded the depth of field, and high-frequency edge details produced optical blur spots, and the detection rate dropped to 42.8%. This result indicates that the 78° to 90° range is the technical parameter boundary for ensuring the coordination of illuminance normalization correction and spatial topology reconstruction. This physical spatial topology restoration mechanism unfolds a three-dimensional mesh through inverse perspective using the optical path mapping matrix, compensating for environmental alternating disturbances and curvature-induced illuminance distortion. The system maintains an objective data plane with a constant physical scale under alternating photoelectric fluctuations and perspective distortion conditions, and outputs defect feature distribution density and surface quality grading data with physical size alignment.

[0059] Example 3: Industrial-grade quality inspection lines for used terminal equipment face physical constraints, including the quantitative judgment of multi-dimensional physical defects and the joint evaluation of internal hardware conditions. Conventional inspection modes independently interpret appearance images and hardware parameters. Multi-source heterogeneous data lacks a standardized mapping space, and the output comprehensive condition rating data deviates from the physical attributes of the tested mobile phone. A multi-angle automatic inspection system for mobile phone appearance based on a multi-faceted reflective structure utilizes an image segmentation and reconstruction module to analyze the original panoramic image. The image segmentation and reconstruction module extracts target pixels based on geometric distortion sampling coordinates and outputs a standardized surface unfolded map aligned with the physical three-dimensional topology. The defect detection module receives this standardized surface unfolded map and resamples it into an input tensor with a fixed resolution and color channel dimensions. The defect detection module then maps this input tensor to... The system incorporates a convolutional feature extraction layer of the Qwen3.5-Plus multimodal model. Within this layer, a visual coding network separates spatial physical features based on the principle of high-frequency feature mutations at image edges. A two-dimensional sliding sampling window is used to calculate the grayscale difference between the center pixel and its neighboring pixels along both the horizontal and vertical directions. This extracts the pixel gradient magnitude matrix representing physical morphological mutations, transforming the original input tensor into a geometric topological feature map. The multimodal model calculates the spatial cosine similarity between the local pixel feature gradient and the preloaded standard physical scratch map. Based on this similarity, the boundary contours of five types of physical defects—scratches, bumps, cracks, wear, and discoloration—are determined. The defect detection module accumulates the area of ​​connected components for each type of physical defect, calculates the percentage of defective pixels per unit physical area, and outputs an optical defect density index.

[0060] The multi-interface data reading module reads the hardware parameters and security verification information of the mobile phone under test via the physical bus. It uses a one-way hash algorithm to calculate the device identification code in the security verification information and outputs a desensitized feature string. The central control unit reads the desensitized feature string, hardware parameters, and optical defect density index. The central control unit then uses a formula... Calculate the comprehensive decay index ,in, The comprehensive decay index, The dimensionless optical attenuation weight is set based on the reciprocal of the variance of historical optical quality inspection data. It is the optical defect density index. The dimensionless hardware degradation weight is set based on the historical hardware failure probability distribution. Based on the dimensionless hardware loss factor converted from hardware parameters, the central control unit compares the comprehensive attenuation index with the five preset discrete quantization grading threshold boundaries and outputs five-level color rating data. The central control unit multiplies the five-level color rating data with the pre-loaded residual value conversion factor and outputs recycling pricing reference data. The multi-angle automatic detection system for mobile phone appearance based on multi-faceted reflection structure outputs a standardized physical inspection file containing desensitized feature strings, optical defect density index and five-level color rating data. Through the integrated data synchronization interface, the standardized physical inspection file is sent to the background management system using the application programming interface. The system outputs unified rating data by combining external appearance optical features and internal hardware physical parameters.

[0061] Example 4: When the system faces the situation of cross-batch hardware assembly and initial deployment of multi-faceted panoramic reflection mechanisms, the central processing unit starts an offline spatial calibration program to build the underlying measurement benchmark before routine testing. The panoramic acquisition unit acquires the initial calibration image of the built-in standard three-dimensional target under constant light source illumination. The spatial mapping module extracts the actual two-dimensional pixel coordinates of each feature point in the initial calibration image. By placing a standard three-dimensional calibration target with a 10 mm x 10 mm black and white square array at the center of the multi-faceted panoramic reflection device, the panoramic acquisition unit acquires the feature corner pixel values ​​containing 16 known physical spatial coordinates; the spatial mapping module... Using the camera's optical center as the origin, the system searches for the optimal spatial transformation matrix parameters that minimize the root mean square value of pixel deviation to less than 0.5 pixels through 0.1 mm-level translational steps and 0.5-degree-interval rotational scans. This completes the initial geometric coupling between the multi-faceted reflector and the camera coordinate system. The actual two-dimensional pixel coordinates and the known three-dimensional feature coordinate array of the standard three-dimensional target are then input into a direct linear transformation model to solve for the actual optical path deflection parameters of the multi-faceted panoramic reflector mechanism in its current assembly state. Based on these actual optical path deflection parameters, the spatial mapping module calculates the inverse perspective projection matrix from the three-dimensional world coordinate system to the camera's two-dimensional image pixel coordinate system. The central processing unit will use the reverse perspective projection matrix Write it to the baseline data storage unit for subsequent distortion correction module to use.

[0062] In the pre-initialization steps of defect detection module deployment and multi-dimensional benchmark rating calculation, the benchmark data storage unit imports an offline reference dataset containing known optical defect morphologies and precise color rating labels. The defect detection module uses this offline reference dataset to input the backpropagation error signal into the convolutional feature extraction layer of the multimodal model. Based on this, it updates the convolutional kernel parameters and establishes the extraction anchor points for local pixel feature gradients. Simultaneously, the central control unit statistically analyzes the distribution variance of various optical defect features and the expected probability distribution of various hardware failures in the offline reference dataset, and calculates the reciprocal of the above distribution variance to generate dimensionless optical attenuation weights. The expected value of the above probability distribution is extracted and assigned to the dimensionless hardware attenuation weight. Through the above steps, the central control unit constructs an initialization system instance containing a reverse perspective projection matrix and quantized attenuation weights, establishing a numerical benchmark for subsequent routine measurements.

[0063] Example 5: When the system encounters the situation of importing a new model of terminal equipment into the testing benchmark library for the first time, the central processing unit starts the equipment pre-alignment program. The spatial mapping module controls the panoramic acquisition unit to acquire the original panoramic image of the physical calibration component with standard positioning marks. The pixel coordinates of the corner extreme points of the physical calibration component in the original panoramic image are extracted, and the pixel coordinates of the corner extreme points are registered with the theoretical edge feature node set in the standardized three-dimensional geometric model data in the benchmark data storage unit to calculate the three-dimensional translation vector and spatial rotation matrix. The central processing unit compensates for the physical assembly tolerance of the multi-faceted panoramic reflection mechanism according to the three-dimensional translation vector and spatial rotation matrix, and recalculates the reverse perspective projection matrix. ,in, The central processing unit updates the inverse perspective projection matrix from the 3D world coordinate system to the camera's 2D image pixel coordinate system. Write it into the reference data storage unit as a measurement reference.

[0064] After establishing the spatial measurement benchmark, the central control unit imports a priori dataset containing both defect-free good products and samples with known edge defects. The Qwen3.5-Plus multimodal model within the defect detection module calculates the spatial cosine similarity of the local pixel feature gradients of various samples in the priori dataset. The central control unit then calculates the maximum spatial cosine similarity of the basic environmental interference generated by defect-free good products and applies the formula... Calculate the baseline confidence threshold, where, The baseline confidence threshold, The maximum spatial cosine similarity of the baseline environment interference is given, and 1.5 is a dimensionless tolerance compensation constant. When the spatial cosine similarity of the output in routine detection exceeds this baseline confidence threshold... At that time, the central control unit determines that a physical defect exists in the target physical area, and accordingly generates a matrix containing the aforementioned reverse perspective projection matrix. Compared with baseline confidence threshold The static parameter sequence is obtained, and the system initialization adaptation settings are completed.

[0065] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A multi-angle automatic detection system for mobile phone appearance based on multi-faceted reflective structure, characterized in that, include: The panoramic acquisition unit is used to acquire the original panoramic image formed by the mobile phone under test through the multi-faceted panoramic reflection mechanism; The baseline data storage unit is used to store the standardized three-dimensional geometric model data of the mobile phone under test; The central processing unit includes: a spatial mapping module, which is used to establish standardized three-dimensional geometric model data as a spatial measurement benchmark, and construct an optical path mapping matrix between each sampling pixel site and a physical topological site based on the geometric distribution parameters of the reflective surface of the multi-faceted panoramic reflection mechanism and the optical center position parameters of the panoramic acquisition unit; The distortion correction module is used to inversely project the surface mesh of the standardized 3D geometric model data to the camera coordinate system through the optical path mapping matrix, and calculate the geometric distortion sampling coordinates of each node of the surface mesh; The image reconstruction module is used to extract pixels from the original panoramic image and perform spatial topology reconstruction based on the geometric distortion sampling coordinates, and output a standardized surface unfolding map that eliminates nonlinear perspective distortion and is closely aligned with the physical three-dimensional structure of the mobile phone under test. The defect detection module is used to extract defect features from the standardized surface unfolded diagram and determine the surface quality level of the mobile phone under test based on the physical size and distribution density of the defect features. And, the central control unit is used to perform illuminance normalization correction on the standardized surface unfolded map: using the spatial optical path mapping matrix, it calculates the cosine value of the angle between the normal vector corresponding to each sampled pixel and the optical axis; based on the cosine value of the angle and the calculation formula... The brightness gain coefficient G of each sampled pixel is determined, where G is the brightness gain coefficient and θ is the angle between the normal vector corresponding to the sampled pixel and the optical axis of the panoramic photoelectric acquisition unit. The pixels extracted from the original panoramic image are subjected to brightness correction mapping to eliminate the local illuminance difference inside the multi-faceted panoramic reflection device, so that the pixel gray values ​​in the output standardized surface unfolding map characterize the physical reflectivity characteristics of the surface of the mobile phone under test. The multi-faceted panoramic reflective device includes an eight-faceted precision reflective structure; the eight-faceted precision reflective structure is made of optical-grade float glass, and the included angle between adjacent reflective surfaces is in the range of 78° to 90°. In this process, at the junction of adjacent panoramic views generated by the 8-sided precision reflection structure, an image overlap buffer with a width of 40 pixels is set. The image reconstruction module calculates the gradient difference of gray values ​​of adjacent views in the buffer and executes a linear weighted fusion algorithm based on distance ratio. The fusion weight coefficient gradually increases from 0 to 1 with the horizontal offset of the pixel from the center line in steps of 0.025 per pixel, thereby eliminating physical scale jumps and pixel breaks at the view seams.

2. The automatic multi-angle detection system for mobile phone appearance based on a multi-faceted reflection structure according to claim 1, characterized in that, The system also includes a uniform illumination module; the uniform illumination module is used to provide 360° surround illumination for the mobile phone under test, so as to cooperate with the multi-faceted panoramic reflection device to realize all-round imaging of the front, back, four bezels and four bevels of the mobile phone under test. When identifying defect features, the central control unit performs regional cropping of the original panoramic image through the image segmentation and reconstruction module, and feeds the cropped feature tensor into the surface feature extraction module to identify five types of physical defects: scratches, bumps, cracks, wear and discoloration.

3. The automatic multi-angle detection system for mobile phone appearance based on a multi-faceted reflection structure according to claim 1, characterized in that, The system also includes a multi-interface data reading module; the multi-interface data reading module is used to read the hardware parameters and security verification information of the mobile phone under test; the central control unit associates the hardware parameters, security verification information and surface quality grading data to generate a comprehensive test report.

4. The automatic multi-angle detection system for mobile phone appearance based on a multi-faceted reflection structure according to claim 3, characterized in that, The central control unit also integrates a data synchronization interface, which is used to send the comprehensive test report to the background management system in real time via API interface, and generate recycling pricing data according to the preset residual value calculation rules.

5. The automatic multi-angle detection system for mobile phone appearance based on a multi-faceted reflection structure according to claim 1, characterized in that, When constructing the spatial optical path mapping matrix, the central control unit extracts the coordinates of the edge calibration points of each reflective surface in the original panoramic image to complete the geometric calibration based on spatial homography transformation, so as to correct the physical position deviation of the multi-faceted panoramic reflective device.

6. The automatic multi-angle detection system for mobile phone appearance based on a multi-faceted reflection structure according to claim 1, characterized in that, The reference data storage unit stores a standardized three-dimensional topology library for different mobile phone models; the central control unit reads the device identifier of the mobile phone under test and automatically matches the corresponding model data from the standardized three-dimensional topology library to change the measurement reference system in real time.

7. A method for automatic multi-angle detection of mobile phone appearance based on a multi-faceted reflection structure, used to implement the automatic multi-angle detection system for mobile phone appearance based on a multi-faceted reflection structure as described in claim 1, characterized in that, The multi-faceted panoramic reflective device includes an eight-sided precision reflective structure; the eight-sided precision reflective structure is made of optical-grade float glass, and the included angle between adjacent reflective surfaces is in the range of 78° to 90°; it includes the following steps: The panoramic acquisition unit acquires the original panoramic image formed by the mobile phone under test through the multi-faceted panoramic reflection mechanism; the standardized three-dimensional geometric model data of the mobile phone under test is retrieved from the reference data storage unit, and the standardized three-dimensional geometric model data is established as the spatial measurement reference. The central processing unit calculates the optical path mapping matrix between each sampled pixel in the original panoramic image and the physical topological position in the spatial measurement reference, based on the geometric distribution parameters of the reflective surface of the multi-faceted panoramic reflection mechanism and the optical center position parameters of the panoramic acquisition unit, using the law of light reflection. The central control unit uses the spatial optical path mapping matrix to calculate the cosine value of the angle between the normal vector corresponding to each sampled pixel and the optical axis; based on the cosine value of the angle and the calculation formula... The brightness gain coefficient G of each sampled pixel is determined, where G is the brightness gain coefficient and θ is the angle between the normal vector corresponding to the sampled pixel and the optical axis of the panoramic photoelectric acquisition unit; brightness correction mapping is performed on the pixels extracted from the original panoramic image to eliminate local illuminance differences inside the multi-faceted panoramic reflection device. The central processing unit inversely projects the surface mesh of the standardized 3D geometric model data onto the camera coordinate system through the optical path mapping matrix, and determines the geometric distortion sampling coordinates of each node of the surface mesh; The central processing unit extracts pixels from the original panoramic image based on the geometric distortion sampling coordinates and completes spatial topology reconstruction to generate a standardized surface unfolding map that eliminates perspective distortion and is aligned with the physical three-dimensional structure of the mobile phone under test. At the junction of adjacent panoramic views generated by the 8-face precision reflection structure, an image overlap buffer with a width of 40 pixels is set. By calculating the gradient difference of gray values ​​of adjacent views within the buffer, a linear weighted fusion algorithm based on distance ratio is executed. The fusion weight coefficient gradually increases from 0 to 1 with the horizontal offset of the pixel from the center line in steps of 0.025 per pixel, thereby eliminating physical scale jumps and pixel breaks at the view seams. The central processing unit extracts defect features from the standardized surface unfolded diagram and determines the surface quality level of the mobile phone under test based on the physical size and distribution density of the defect features.

Images