A method and system for ceramic circuit line sub-pixel image measurement based on multi-spectral imaging

Abstract

The application discloses a kind of ceramic circuit line subpixel image measurement methods based on multispectral imaging, comprising: constructing multi-band composite illumination and imaging system, submicron level positioning and continuous scanning to the measured ceramic substrate;Multiple frame average acquisition is obtained to ceramic circuit line area multimodal image data;Establish the nonlinear calibration mapping model of image coordinate system and physical size coordinate system;Construct multispectral response coupling matrix and spectral domain enhancement model;Subpixel level spatial registration is carried out to multimodal image;Subpixel boundary positioning model is constructed, and the subpixel boundary positioning of ceramic circuit line is carried out;Construct line geometry parameter analytical model;Construct multispectral dynamic error compensation model, and the measurement result is adaptively corrected;Based on nonlinear calibration mapping model, the geometry parameter of subpixel level after error compensation is converted into physical size, and the final measurement result of ceramic circuit line is output. System, electronic equipment and computer readable storage medium are also disclosed.

Description

Technical Field

[0001] This invention relates to the field of precision inspection and multispectral visual measurement technology in electronic manufacturing, and in particular to a method and system for measuring subpixel images of ceramic circuit lines based on multispectral imaging. It belongs to a high-precision image measurement method and system for ceramic circuit lines that integrates spectral coupling modeling, subpixel curvature analysis and dynamic error compensation mechanism. Background Technology

[0002] Ceramic circuit boards, with their excellent thermal stability, electrical insulation properties, thermal conductivity, and mechanical strength, have become core components in high-end manufacturing fields such as power electronics, new energy vehicles, aerospace, and high-frequency communication systems. They mainly include copper-clad ceramic substrates (DCB), high-temperature co-fired ceramics (HTCC), and low-temperature co-fired ceramics (LTCC). As electronic devices develop towards higher power density, higher integration, and miniaturization, the line width and spacing of ceramic circuits are continuously shrinking. Currently, in high-end applications such as new energy vehicle power modules, the line width of ceramic circuits has been reduced to below 50μm. Micrometer-level or even submicrometer-level dimensional control has become a key indicator in ceramic circuit manufacturing processes. Even minute deviations in line dimensions can lead to problems such as thermal stress concentration, short circuits, and abnormal signal transmission, directly affecting product performance and reliability.

[0003] To ensure the manufacturing quality of ceramic circuits, high-precision inspection of geometric parameters such as line width, line spacing, line edge roughness, straightness, and pad size is required. Currently, the mainstream inspection method in the industry is automated optical inspection (AOI) systems. However, its core is based on a single visible light band imaging technology, which has many technical bottlenecks and can no longer meet the precision inspection requirements of current high-end ceramic circuits.

[0004] 1. Insufficient positioning accuracy: The line boundary positioning of traditional AOI systems usually stays at the pixel level accuracy. The error range of pixel level accuracy is usually 1-2 pixels. For high-resolution cameras, this error is equivalent to a physical size deviation of several micrometers, which cannot meet the sub-micrometer measurement requirements under nanometer-level technology.

[0005] 2. Low boundary contrast: The different materials such as copper layer, ceramic substrate, and oxide layer of ceramic circuit board do not have significant differences in reflectance characteristics under a single visible light band. For example, copper has high reflectivity under visible light, while the ceramic substrate has more significant reflectance characteristics in the near-infrared band. Single-band imaging is difficult to form a stable and high-contrast line boundary, which easily leads to blurred boundaries.

[0006] 3. Large edge positioning deviation: The edges of ceramic circuits have natural transition areas and gray-level gradient areas. Traditional threshold segmentation and first-order gradient detection methods (such as Canny edge detection and Sobel operator) only consider the first-order derivative features of the image, which cannot effectively handle sub-pixel gray-level gradients and are prone to large edge positioning deviations.

[0007] 4. Measurement errors are significantly affected by the environment: Factors such as changes in the incident angle of the light source, light source aging, ambient light pollution, equipment vibration, differences in the micro-texture of the ceramic substrate surface, and drift in material reflectivity in the industrial production environment can all introduce significant systematic and random errors into traditional measurement methods. In some scenarios, the deviation can reach 5-10%, which seriously affects the repeatability and stability of the measurement.

[0008] 5. Insufficient application of multispectral technology: The few existing detection methods that use multispectral imaging are mostly limited to simple image fusion. They lack dedicated multispectral coupling modeling for the optical properties of ceramic materials and have not built an effective dynamic error compensation mechanism. As a result, they are not suitable for mass production testing of ceramic circuits and have poor measurement consistency.

[0009] Meanwhile, while traditional contact measurement methods can achieve high accuracy, they suffer from low measurement efficiency, are prone to physical damage to the delicate circuitry of ceramic circuits, and are unsuitable for online continuous inspection. They have been gradually replaced by non-contact vision measurement methods. Therefore, there is an urgent need to develop a high-precision image measurement method and system that integrates multispectral enhancement and sub-pixel fine resolution to specifically address the aforementioned technical pain points in ceramic circuit line inspection. This would simultaneously improve measurement accuracy, repeatability, and stability, support online integration, meet the inspection needs of mass production of ceramic circuits, reduce human intervention, and adapt to various ceramic substrate types. Summary of the Invention

[0010] To address the problems existing in the prior art, this invention provides a method and system for subpixel image measurement of ceramic circuit lines based on multispectral imaging. This method combines multidisciplinary knowledge from computer vision, signal processing, and optical engineering, aiming to overcome the bottlenecks of traditional detection methods and provide an efficient, non-destructive solution. Specifically, it integrates multispectral coupling modeling, spectral domain adaptive enhancement, and subpixel continuous boundary resolution techniques to achieve high-precision non-contact measurement of ceramic circuit lines. The method includes constructing a multi-band composite illumination and imaging system, acquiring multimodal image data, establishing a calibration mapping model, constructing a multispectral response coupling matrix and spectral domain enhancement model, subpixel-level multi-band spatial registration, subpixel boundary localization, geometric parameter analysis, dynamic error compensation, and outputting measurement results. The system correspondingly implements each functional module of the above method. This invention enhances the contrast of circuit boundaries by leveraging multi-band spectral reflectivity differences. Combined with grayscale curvature continuity and gradient phase consistency, it achieves boundary positioning accuracy better than 0.05 pixels. A dynamic error compensation model reduces environmental interference, enabling sub-micron level measurements of geometric parameters such as line width, line spacing, and line edge roughness in ceramic circuits. It is suitable for online precision inspection of products such as copper-clad ceramic substrates, HTCC, and LTCC, with a processing speed of up to 50 frames / second, repeatability error less than 1%, and measurement deviation controlled within 0.5%. This provides reliable technical support for the intelligent upgrading of the electronics manufacturing industry. This invention focuses on the application of multispectral imaging technology to ceramic materials, utilizing the differences in reflectivity of different wavelengths to enhance image features. This method is not only applicable to standard ceramic substrates but can also be extended to the inspection of composite materials or multilayer circuit boards. Through sub-pixel-level resolution, this invention can capture micron-level defects, improve product quality control, address the pain points of traditional methods, and provide a comprehensive and robust inspection solution suitable for high-precision electronics manufacturing.

[0011] The embodiments of the present invention address the problems of low positioning accuracy, poor boundary contrast, weak environmental interference resistance, and inability to adapt to online precision testing in existing ceramic circuit measurement methods. They provide a sub-pixel image measurement method and system for ceramic circuits based on multispectral imaging. Through full-process technological innovation including multispectral coupling modeling, spectral domain adaptive enhancement, sub-pixel continuous boundary resolution, and multispectral dynamic error compensation, the following objectives are achieved:

[0012] 1. By utilizing the difference in spectral reflectance between ceramic and metal materials at different wavelengths, the optical contrast of the line boundary region is enhanced, improving the image signal-to-noise ratio by at least 20%. Experiments have verified that the boundary contrast is improved by more than 30%.

[0013] 2. Construct a sub-pixel boundary localization model based on grayscale curvature continuity and gradient phase consistency to achieve sub-pixel level precise boundary localization of ceramic circuit lines. The localization accuracy is better than 0.05 pixels, and can reach 0.03 pixels in actual tests. The standard deviation is 0.02 pixels in Monte Carlo simulation verification.

[0014] 3. Establish a high-precision pixel-physical size nonlinear calibration mapping model to achieve sub-micron-level size conversion with a conversion error of less than 0.1μm;

[0015] 4. Construct a multispectral dynamic error compensation model to adaptively compensate for interference factors such as spectral response deviation, incident angle change, and material reflectivity drift, reduce the measurement impact caused by environmental interference, and control the measurement deviation within 0.5%.

[0016] 5. Achieve stable measurement of micron-level geometric parameters such as ceramic circuit line width, line spacing, line edge roughness, straightness, and pad size, with a measurement repeatability error of less than 1%;

[0017] 6. Design a hardware system and software module adapted to the above measurement methods, supporting online continuous scanning measurement, with a data processing speed of up to 50 frames / second. It can be integrated into automated production lines, supporting a data throughput of hundreds of frames per second, meeting the testing needs of mass production of ceramic circuits.

[0018] This invention provides a method for subpixel image measurement of ceramic circuit lines based on multispectral imaging. It integrates technologies from multiple disciplines, including optical engineering, computer vision, signal processing, and numerical analysis, and constructs a closed-loop measurement system encompassing image acquisition, spectral enhancement, subpixel registration, boundary localization, parameter analysis, error compensation, and result output. This method is applicable to online continuous scanning measurement of alumina, aluminum nitride, and silicon nitride ceramic substrates, copper-clad ceramic substrates, HTCC, and LTCC ceramic circuit boards. It supports a data throughput of hundreds of frames per second, with a measurement repeatability error of less than 1% and a measurement deviation controlled within 0.5%. The method includes:

[0019] S1 constructs a multi-band composite illumination and imaging system covering the 400-1000nm band, supports different band switching and 0-45 degree incident angle adjustment, and is equipped with a precision motion platform to realize submicron-level positioning and continuous scanning of the ceramic substrate under test;

[0020] In a preferred embodiment, the multi-band composite illumination and imaging system includes a multi-band composite light source, a high-resolution camera, a precision motion platform, a calibration module, and an image acquisition and processing unit.

[0021] S2, under different wavebands and different incident angles, performs multi-frame averaging acquisition of the ceramic circuit line area to obtain multimodal image data, and suppresses noise through multi-frame fusion to ensure image quality;

[0022] In a preferred embodiment, S2 includes:

[0023] S21, the ceramic circuit board to be tested is placed on a precision motion platform;

[0024] S22 performs multi-frame averaging acquisition of ceramic circuit lines under different sub-bands within the 400-1000nm band and different incident angles of 0-45 degrees, obtains multimodal image data, and stores it in RAW or TIFF format.

[0025] S3 employs a laser-etched 10μm pitch precision grid calibration plate, collects at least 100 control points, and establishes a nonlinear calibration mapping model between the image coordinate system and the physical size coordinate system to achieve high-precision conversion from pixel size to physical size; wherein, the nonlinear calibration mapping model includes radial and tangential distortion correction, and the conversion error of the high-precision conversion is less than 0.1μm;

[0026] S4. Construct a multispectral response coupling matrix and a spectral domain enhancement model to maximize the contrast enhancement of the line boundary region;

[0027] In a preferred embodiment, S4 includes:

[0028] S41, Based on the difference in spectral reflectance between the ceramic substrate material and the metal conductor in different wavelength bands, a multispectral response coupling matrix is ​​constructed; the multispectral response coupling matrix is ​​defined as: ,in For the first The matrix represents the response intensity of the corresponding material in the specified band. It is initialized using the reflectivity data of the ceramic substrate and the metal conductor obtained experimentally. The matrix elements can be fine-tuned according to the specific material type and metal conductor type of the ceramic circuit board being tested.

[0029] S42, a spectral domain enhancement model is constructed by combining the band energy ratio function, spectral gradient tensor, and spectral domain stability factor, generating an adaptive fusion factor to maximize the contrast enhancement of the line boundary region; the enhancement weight function of the spectral domain enhancement model... ,in: The band energy ratio function is calculated using the following formula: ,in For the first Energy value of band image The sum of the energy values ​​of all acquired band images. It is the spectral gradient tensor, obtained through convolution operations; The spectral stability factor is calculated based on the variance of historical data.

[0030] S5, with spectral domain consistency as a constraint, performs sub-pixel-level spatial registration on multimodal images of different bands, eliminates geometric deviations in cross-band imaging, and achieves cross-band sub-pixel-level geometric alignment.

[0031] S6, Construct a sub-pixel boundary positioning model, and realize the sub-pixel boundary positioning of ceramic circuit lines based on the sub-pixel boundary positioning model;

[0032] In a preferred embodiment, S6 includes:

[0033] S61 extracts initial pixel-level edges using the Canny adaptive thresholding algorithm;

[0034] S62, Based on the initial pixel-level edge, construct a sub-pixel boundary localization model based on grayscale curvature continuity and gradient phase consistency constraints;

[0035] S63, sub-pixel boundary localization of ceramic circuit lines is achieved through second-order differential extremum analysis, local Gaussian-parabolic joint fitting, and boundary energy function optimization; wherein the sub-pixel boundary localization model is achieved by minimizing the boundary energy function, which is: ,in The actual grayscale value of the image. To fit the grayscale values ​​of the curve, the gradient descent algorithm is used in the solution process, combined with constraints on grayscale curvature continuity and gradient phase consistency to avoid local optima. Sub-pixel level boundary solutions are achieved by performing three-point parabolic fitting or second-derivative extremum analysis on the edge grayscale curves. The formula for calculating sub-pixel coordinates using the three-point parabolic fitting is as follows: ,in pixel-level edge coordinates , , The gray values ​​of three adjacent sampling points. These are the sub-pixel boundary coordinates;

[0036] S7, construct the analytical model of line geometric parameters, and determine the line width based on the analytical model of line geometric parameters, and analyze the line spacing, line edge roughness, straightness and pad size;

[0037] In a preferred embodiment, S7 includes:

[0038] S71, Construct an analytical model of the line's geometric parameters;

[0039] S72, Solving for the line width based on the minimum distance algorithm of normal projection;

[0040] S73, based on the processing of the curved path, performs precise analysis of line spacing, line edge roughness, conductor straightness, and pad size; wherein, the formula for calculating the line edge roughness is: Calculate the line edge roughness, where The number of sampling points. The deviation value between the boundary point and the fitted curve; the straightness of the traverse is obtained by calculating the maximum deviation value after removing outliers through least squares straight line fitting combined with the RANSAC algorithm.

[0041] S8. A multispectral dynamic error compensation model is constructed. Through real-time spectral monitoring, sensor feedback, and Kalman filtering estimation, spectral response deviation, incident angle variation, and material reflectivity drift are obtained, respectively. Compensation coefficients are generated to adaptively correct the measurement results. The multispectral dynamic error compensation model supports online updates and is defined as follows: ,in This is the comprehensive error compensation coefficient. The spectral response deviation is calculated using Fourier analysis based on real-time spectral monitoring. The change in incident angle is obtained based on sensor feedback from the tilt sensor; Kalman filtering is used to estimate the material reflectivity drift.

[0042] S9, based on a nonlinear calibration mapping model, converts the sub-pixel level geometric parameters after error compensation into physical dimensions and outputs the final measurement results of the ceramic circuit.

[0043] A second aspect of the present invention provides a subpixel image measurement system for ceramic circuit lines based on multispectral imaging, which realizes fully automated processing from image acquisition to measurement result output. The hardware part supports online integration, and the software part supports GPU parallel acceleration, significantly improving measurement efficiency, including:

[0044] Multi-band composite illumination module is used to build a multi-band composite illumination and imaging system covering the 400-1000nm band. It supports switching between different bands and adjusting the incident angle from 0 to 45 degrees. When paired with a precision motion support platform, it can basically support the ceramic circuit under test and realize sub-micron level positioning and continuous scanning of the ceramic substrate under test.

[0045] The high-precision imaging acquisition module is used to perform multi-frame averaging acquisition of ceramic circuit lines under different wavelengths and different incident angles to obtain multimodal image data. Noise is suppressed through multi-frame fusion to ensure image quality.

[0046] The calibration module uses a laser-etched 10μm pitch precision grid calibration plate to collect at least 100 control points and establish a nonlinear calibration mapping model between the image coordinate system and the physical size coordinate system to achieve high-precision conversion from pixel size to physical size. The nonlinear calibration mapping model includes radial and tangential distortion correction, and the conversion error of the high-precision conversion is less than 0.1μm.

[0047] The multispectral enhancement module constructs a multispectral response coupling matrix based on the difference in spectral reflectance between ceramic substrate materials and metal conductors in different bands. At the same time, it combines the band energy ratio function, spectral gradient tensor and spectral domain stability factor to construct a spectral domain enhancement model and generate an adaptive fusion factor to maximize the contrast enhancement of the line boundary region.

[0048] The subpixel registration module is used to perform subpixel-level spatial registration of multimodal images in different bands with spectral domain consistency as a constraint, to eliminate geometric deviations in cross-band imaging and achieve cross-band subpixel-level geometric alignment.

[0049] A subpixel positioning module is used to construct a subpixel boundary positioning model and realize the subpixel boundary positioning of ceramic circuit lines based on the subpixel boundary positioning model.

[0050] The geometric parameter analysis module is used to construct a geometric parameter analysis model of the line, and determine the line width based on the geometric parameter analysis model of the line, and analyze the line spacing, line edge roughness, straightness and pad size;

[0051] The dynamic error compensation module is used to construct a multispectral dynamic error compensation model. Through real-time spectral monitoring, sensor feedback, and Kalman filter estimation, it obtains spectral response deviation, incident angle change, and material reflectivity drift, respectively, and generates compensation coefficients to adaptively correct the measurement results. The multispectral dynamic error compensation model supports online updates.

[0052] The results output module is used to convert the sub-pixel level geometric parameters after error compensation into physical dimensions based on the nonlinear calibration mapping model, and output the final measurement results of the ceramic circuit.

[0053] A third aspect of the present invention provides an electronic device including a processor and a memory, the memory storing a plurality of instructions, the processor being configured to read the instructions and execute the method as described in the first aspect.

[0054] A fourth aspect of the present invention provides a computer-readable storage medium storing a plurality of instructions which can be read by a processor and executed as described in the first aspect.

[0055] The method, system, and electronic device provided by this invention have the following beneficial effects:

[0056] Specialized modeling of the optical properties of ceramic materials was developed, breaking through the accuracy bottleneck of traditional single-band imaging. By combining multispectral enhancement and sub-pixel resolution, sub-micron level measurement of ceramic circuits was achieved. At the same time, the introduced dynamic error compensation mechanism effectively reduced the impact of industrial environmental interference, ensuring the repeatability and stability of the measurement. Furthermore, the system is adapted to online continuous scanning measurement scenarios and can be seamlessly integrated into the automated production line of ceramic circuits, realizing intelligent and unmanned testing processes. Attached Figure Description

[0057] Figure 1 This is a schematic diagram of the overall structure of the multispectral measurement system of the present invention. The diagram shows the hardware composition and optical path of the system, including a multi-band composite light source 1, a light source control module 2, an illumination optical system 3, a ceramic circuit board under test 4, a precision motion platform 5, a high-resolution multispectral camera 6, a filter wheel or beam splitting system 7, an image acquisition and processing computer 8, a calibration plate 9, and an ambient light shield 10. The diagram also shows the positional relationship of each hardware component and the optical imaging path.

[0058] Figure 2 This is a flowchart of the multi-band image acquisition process of the present invention. Rectangles represent operation steps and diamonds represent decision-making. It shows the entire process of system initialization, light source band switching and incident angle setting, adaptive adjustment of exposure parameters, multi-frame average acquisition, image preview and quality check, data storage, and band cycle judgment. Key parameters such as exposure time 50ms, band switching delay <100ms, and acquisition of ≥5 frames per band are marked.

[0059] Figure 3 This is a flowchart of the spectral response enhancement algorithm of the present invention, showing the core modules and data flow of the algorithm. Starting from the input multi-band original image, it goes through normalization processing, band energy ratio calculation, local spectral gradient calculation, spectral domain stability factor calculation, adaptive weight fusion, and finally outputs an enhanced fused image. Key formulas and technical parameters are marked, such as normalization formula, band energy ratio formula, GPU parallel acceleration, single frame processing <10ms, contrast improvement ≥30%, etc. Dashed boxes indicate optional temperature / noise compensation modules.

[0060] Figure 4 This is a schematic diagram of the sub-pixel edge localization model of the present invention, divided into three parts: (a) pixel-level initial edge extraction, (b) grayscale curve sampling along the normal direction, and (c) sub-pixel localization result. It shows the localization process from pixel-level edge to sub-pixel boundary and marks key features such as Canny adaptive threshold, grayscale gradient area, and sub-pixel offset. (Better than 0.05 pixels), curvature continuity constraints, gradient phase consistency constraints, etc., with the boundary energy function shown at the bottom. The minimization iteration process was marked with an actual positioning accuracy of 0.03 pixels and a Monte Carlo simulation standard deviation of 0.02 pixels.

[0061] Figure 5 This is a schematic diagram of the geometric parameter analysis of the circuit of the present invention, showing the geometric parameter analysis process of a typical ceramic circuit circuit cross section, including (a) boundary point cloud extraction, (b) width calculation using the minimum distance algorithm of normal projection, (c) line spacing calculation, (d) line edge roughness Ra calculation, and (e) conductor straightness fitting. The calculation formulas and solution methods of each parameter are marked, and it is clearly supported for curve path processing and pad size measurement.

[0062] Figure 6 This is a schematic diagram of the pixel coordinate and physical size calibration mapping model of the present invention, including (a) calibration plate (laser-etched 10μm grid), (b) control point acquisition (at least 100 points), (c) polynomial mapping function, and (d) transformation error distribution heatmap. It shows the construction process and accuracy indicators of the calibration mapping model, and marks key information such as polynomial mapping formula, radial / tangential distortion correction, and transformation error <0.1μm. It clearly supports neural network-assisted nonlinear fitting.

[0063] Figure 7 This is a flowchart of the dynamic error correction mechanism of the present invention, illustrating the closed-loop error correction process of "real-time acquisition - error source monitoring - compensation model calculation - self-learning update - result output". The monitoring methods for each error source are detailed (spectral response deviation). Fourier analysis and incident angle variation were used. Using sensor feedback and material reflectivity drift Kalman filtering estimation is used, along with compensation model formulas, self-learning update mechanisms, and accuracy metrics (deviation control <0.5%).

[0064] Figure 8 This is a schematic diagram of the electronic device structure described in this invention. Detailed Implementation

[0065] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0066] To make the objectives, technical solutions, and advantages of this invention clearer, the embodiments of this invention will be described in great detail and comprehensively below with reference to the accompanying drawings. This invention not only provides the core method flow but also explores various implementation variations, exception handling mechanisms, resource optimization strategies, and adaptation schemes for different database platforms to ensure the robustness, versatility, and industrial applicability of the technical solution.

[0067] The core execution steps of the sub-pixel image measurement method for ceramic circuit lines based on multispectral imaging of the present invention are as follows: constructing a multi-band composite illumination and imaging system → acquiring multimodal image data → establishing a nonlinear calibration mapping model → constructing a multispectral response coupling matrix and spectral domain enhancement model → sub-pixel-level multi-band spatial registration → sub-pixel boundary localization → geometric parameter analysis → multispectral dynamic error compensation → outputting physical dimension measurement results; the corresponding measurement system provides hardware support and software algorithm modules to realize the above steps. The overall system structure is shown in Figure 1, and the specific implementation details of each step are as follows:

[0068] Example 1: Construction of a Multi-band Composite Illumination and Imaging System

[0069] The multi-band composite illumination and imaging system constructed in this invention forms the hardware foundation of the entire measurement method. It is used to achieve stable and high-precision acquisition of multimodal images of the ceramic circuit area. The overall system structure is shown in Figure 1, and mainly includes a multi-band composite light source 1, a light source control module 2, an illumination optical system 3, the ceramic circuit board under test 4, a precision motion platform 5, a high-resolution multispectral camera 6, a filter wheel or beam splitting system 7, an image acquisition and processing computer 8, a calibration plate 9, and an ambient light shield 10. The technical parameters and functional requirements of each component are as follows:

[0070] 1. Multi-band composite light source 1: It adopts a programmable LED array light source, covering the visible light to near-infrared band of 400-1000nm, supports programmable switching of continuous band or discrete sub-band, the band switching delay is less than 100ms, and the light source brightness is adjustable to meet the reflection characteristics requirements of different ceramic materials.

[0071] 2. Light source control module 2: It is matched with a multi-band composite light source to realize functions such as band switching, brightness adjustment, and incident angle control of the light source. It supports automated script control to ensure the stability and repeatability of light source parameters.

[0072] 3. Illumination Optical System 3: Includes a collimating lens and a diffuse reflector, used to collimate and diffuse the light emitted by the multi-band composite light source to ensure that the light is uniformly incident on the surface of the ceramic circuit board under test, avoiding local overexposure or underexposure. It also supports an incident angle adjustment of 0-45 degrees to adapt to the differences in the reflection characteristics of different ceramic substrates and metal conductors.

[0073] 4. Precision Motion Platform 5: Adopts a high-precision electric displacement stage with a positioning accuracy of 0.1μm and a motion repeatability better than 0.05μm. It supports continuous scanning mode and fixed-point acquisition mode. It is used to carry the ceramic circuit board under test, realize the precise positioning of the test area and online continuous scanning, and meet the batch testing requirements.

[0074] 5. High-resolution multi-spectral camera 6: Select an area array or line array camera with a pixel size ≤3.45μm and a resolution of at least 2048x2048. Support simultaneous acquisition of visible light (450-700nm) and near-infrared (800-950nm) channels. Paired with a high-transmittance optical lens with extremely low lens distortion coefficient to ensure the geometric accuracy of the image.

[0075] 6. Filter wheel or beam splitting system 7: Compatible with high-resolution multi-spectral cameras, with built-in filters for different wavelengths, supporting fast switching to achieve monochromatic light imaging in different wavelengths and ensuring the band specificity of multimodal image data;

[0076] 7. Image Acquisition and Processing Computer 8: Equipped with a high-performance CPU, GPU, and large-capacity storage device, and equipped with an image acquisition card and dedicated measurement software, it realizes real-time acquisition, storage, processing, and analysis of multimodal image data, supports GPU parallel acceleration, and greatly improves the processing speed of image algorithms;

[0077] 8. Calibration plate 9: A precision grid calibration plate with laser etching and a grid spacing of 10μm. The flatness and dimensional accuracy of the calibration plate are both at the submicron level. It is used to establish a calibration mapping model between the image coordinate system and the physical dimension coordinate system.

[0078] 9. Ambient light shield 10: Made of light-shielding material, it seals off the system's lighting and imaging areas, effectively shielding against ambient light pollution in industrial production environments and preventing ambient light from interfering with image acquisition quality.

[0079] 10. Auxiliary components: including data cables, power cables, and mounting brackets, used to physically connect and fix the various components of the system, ensure the mechanical stability of the system, and reduce measurement errors caused by equipment vibration.

[0080] After the system is built, overall debugging and calibration are required, including calibration of light source brightness uniformity, camera exposure parameters, motion platform positioning accuracy, and optical system coaxiality. This ensures that the parameters of each component are matched to meet the accuracy requirements of multimodal image acquisition. Simultaneously, noise testing is performed to ensure that the signal-to-noise ratio of the acquired images meets the requirements of subsequent processing. The system features a high dynamic range (HDR) mode, capable of handling imaging scenarios with significant differences in reflectivity between ceramic substrates and metal conductors, effectively avoiding overexposure or underexposure, and providing a high-quality image data foundation for subsequent spectral enhancement and sub-pixel resolution.

[0081] Example 2: Acquisition of Multimodal Image Data

[0082] After completing the construction and debugging of the multi-band composite illumination and imaging system, multimodal image data acquisition of the ceramic circuit area was performed. The acquisition process is as follows: Figure 2 As shown, rectangular boxes represent operation steps and diamond boxes represent decision-making judgments. The core is to perform multi-frame averaging acquisition of the tested ceramic circuit area under different spectral bands and different incident angles to obtain high signal-to-noise ratio, multi-band, multi-modal image data. The specific acquisition steps are as follows:

[0083] 1. S1 System Initialization: Start all equipment of the multi-band composite illumination and imaging system, including multi-band composite light source, high-resolution multi-spectral camera, precision motion platform, image acquisition and processing computer, etc., complete the self-test and parameter initialization of the equipment, restore the parameters of the measurement software to the default values, and ensure that the equipment is in normal working condition.

[0084] 2. S2 Light Source Band Switching and Incident Angle Setting: Based on the material type of the ceramic circuit board under test (such as alumina, aluminum nitride, silicon nitride ceramic substrate, copper layer thickness, etc.), the working band of the multi-band composite light source is set through the light source control module. The visible light band (450-700nm) and the near-infrared band (800-950nm) are usually selected as the core acquisition bands. At the same time, the incident angle of the illumination optical system is adjusted. The incident angle adjustment range is 0-45 degrees. The optimal incident angle can be selected according to the reflection spectrum characteristics of the material. The band switching delay is less than 100ms to ensure the accuracy of the band and incident angle parameter settings.

[0085] 3. S3 Exposure Parameter Adaptive Adjustment: Based on the brightness of the ceramic circuit area under test, the exposure parameters of the high-resolution multi-spectral camera are adaptively adjusted. The core exposure time is set to 50ms, and the camera gain parameters are adjusted at the same time to ensure that the gray value of the acquired image is within a reasonable range and to avoid overexposure or underexposure. The exposure parameters can be adjusted through the automatic exposure function of the measurement software or manually fine-tuned.

[0086] 4. S4 Multi-frame Averaging Acquisition: The ceramic circuit board under test is fixed on a precision motion platform. The test circuit area is moved to the center of the camera's field of view by the precision motion platform. The camera's acquisition function is activated to acquire multiple frames of the test area under the current band and incident angle. At least 5 frames of images are acquired for each band. Multi-frame averaging fusion is used to suppress image noise and improve the signal-to-noise ratio of the image, providing high-quality image data for subsequent spectral enhancement and sub-pixel resolution.

[0087] 5. S5 Image Preview and Quality Check: After acquisition, the fused image is previewed using measurement software. The image quality is checked in terms of grayscale, contrast, noise level, and sharpness to determine whether the image meets the requirements for subsequent processing. If the image quality is not up to standard (e.g., overexposure, underexposure, blur, excessive noise, etc.), return to step S3 to readjust the exposure parameters, or return to step S2 to reset the light source band and incident angle.

[0088] 6. S6 Data Storage: If the image quality check meets the standards, the fused multimodal image data will be stored in RAW or TIFF format in the storage device of the image acquisition and processing computer. At the same time, the relevant parameters of image acquisition (such as band, incident angle, exposure time, gain, etc.) will be recorded to facilitate subsequent traceability and processing.

[0089] 7. S7 Band Cyclic Judgment: Determine whether image acquisition of all preset bands has been completed. If not, return to step S2 to switch to the next preset band and repeat steps S2-S6 until image acquisition of all preset bands is completed. If completed, end the image acquisition process.

[0090] The entire image acquisition process supports automated script control. Acquisition scripts written in the measurement software enable automatic switching and setting of light source bands, incident angles, and exposure parameters, as well as fully automated execution of multi-frame acquisition, image fusion, quality checks, and data storage. This ensures the consistency and repeatability of the acquisition process and avoids errors caused by human operation. The acquired multimodal image data contains optical characteristic information of ceramic circuit lines under different bands, providing a data foundation for subsequent multispectral enhancement to improve the contrast of circuit boundaries.

[0091] Example 3: Establishment of a nonlinear calibration mapping model between the image coordinate system and the physical size coordinate system

[0092] Due to lens distortion (radial and tangential distortion) in camera imaging, and the proportional relationship between image pixel size and actual physical size, a nonlinear calibration mapping model between the image coordinate system and the physical size coordinate system needs to be established to achieve high-precision conversion from sub-pixel image coordinates to physical size. A schematic diagram of the calibration mapping model is shown in Figure 6. The specific implementation steps are as follows:

[0093] 1. Calibration plate placement and acquisition: Fix the laser-etched 10μm pitch precision grid calibration plate on the precision motion platform. Adjust the position of the calibration plate through the precision motion platform so that the calibration plate is in the center of the camera's field of view and the plane of the calibration plate is parallel to the imaging plane of the camera. Then, acquire the image of the calibration plate through a high-resolution multi-spectral camera. During the acquisition process, ensure that the image is clear, without blur, and without noise, and that the grid features of the calibration plate are intact.

[0094] 2. Control point extraction: The collected calibration board image is processed by measurement software to extract the corner points of the calibration board grid as control points. At least 100 control points are required. The distribution of control points should evenly cover the entire camera field of view to avoid the control points being concentrated in local areas, so as to ensure the accuracy and robustness of the calibration model.

[0095] 3. Distortion correction parameter calculation: Analyze the extracted control point coordinates and calculate the radial and tangential distortion parameters of the camera lens. Radial distortion mainly includes barrel distortion and pincushion distortion. Tangential distortion is caused by the misalignment of the optical center and mechanical center of the camera lens. Through distortion correction parameters, geometric distortion correction can be performed on the subsequently acquired ceramic circuit image to eliminate the image geometric error caused by lens distortion.

[0096] 4. Calibration mapping model construction: based on the image coordinates of the extracted control points ( , ) and the corresponding physical dimension coordinates ( , A nonlinear calibration mapping model is constructed, which adopts either a polynomial mapping model or a perspective projection model. The calculation formula for the core polynomial mapping model is as follows:

[0097] ,

[0098] ,

[0099] in, , , …, , , … are the polynomial fitting coefficients, which can be obtained by fitting the coordinate data of the control points using the least squares method;

[0100] 5. Model accuracy verification: Select some control points that were not involved in the fitting as verification points, substitute the image coordinates of the verification points into the constructed calibration mapping model, calculate the corresponding physical size coordinates, and compare them with the actual physical size coordinates of the verification points to calculate the model's transformation error. The transformation error of the model is required to be less than 0.1μm. If the transformation error does not meet the requirements, increase the number of control points and re-fit and solve until the transformation error meets the accuracy requirements.

[0101] 6. Model Optimization (Optional Step): In complex industrial production environments, if the ceramic circuit board has slight warping or deformation, a neural network can be integrated to assist in calibration. The neural network performs nonlinear fitting on the coordinate data of the control points to further improve the nonlinear adaptability of the calibration mapping model and ensure the accuracy of pixel size to physical size conversion.

[0102] The established nonlinear calibration mapping model will be stored in the measurement software. In the subsequent geometric parameter measurement process, the subpixel-level image coordinates of the ceramic circuit can be directly substituted into the model to achieve high-precision conversion to physical dimensions. This model is one of the key links to achieve submicron-level measurement, effectively eliminating geometric errors caused by lens distortion and ensuring the spatial accuracy of the measurement.

[0103] Example 4: Construction of Multispectral Response Coupling Matrix and Spectral Domain Enhancement Model

[0104] The spectral reflectance of ceramic substrate materials and metallic conductors differs significantly across different wavelength bands. This embodiment constructs a multispectral response coupling matrix based on this difference, and simultaneously builds a spectral enhancement model by combining the band energy ratio function, spectral gradient tensor, and spectral stability factor to maximize contrast enhancement in the line boundary region. The flowchart of the spectral enhancement algorithm is shown in Figure 3. The specific implementation steps are as follows:

[0105] 1. Image Preprocessing: The acquired multimodal image data is normalized to eliminate grayscale differences caused by factors such as light source brightness and camera exposure parameters under different wavelengths. The normalization calculation formula is as follows:

[0106] ,in The grayscale values ​​of the original image. The minimum grayscale value of the original image. The maximum grayscale value of the original image. The normalized grayscale value is [0,1].

[0107] 2. Construction of Multispectral Response Coupling Matrix: Based on the difference in spectral reflectance between the ceramic substrate material and the metallic conductor at different wavelengths, a multispectral response coupling matrix is ​​constructed. The matrix is ​​defined as follows: ,in For the first The matrix is ​​initialized with the reflectivity data of the ceramic substrate and the metal conductor obtained by experiment. In practical applications, the matrix elements can be fine-tuned according to the specific material type of the ceramic circuit board under test (such as alumina, aluminum nitride, silicon nitride) and the type of metal conductor (such as copper, silver) to optimize the coupling effect between different bands and maximize the difference in reflection characteristics between the circuit and the substrate.

[0108] 3. Band Energy Ratio Calculation: Calculate the energy values ​​of images under different bands and construct a band energy ratio function. ,in For the first Energy value of band image The sum of energy values ​​of all acquired band images is represented by the band energy ratio function, which reflects the weight of different bands in image enhancement. The higher the energy value of a band, the greater its contribution to the enhancement of the line boundary.

[0109] 4. Spectral gradient tensor calculation: Perform convolution operations on the normalized multi-band image to calculate the spectral gradient tensor. Convolution operations can be performed using the improved Prewitt operator or the Sobel operator, employing a 3x3 convolution kernel to compute... direction and The gradient value of the direction, the spectral gradient tensor reflects the rate of change of gray values ​​in the image. The gradient value of the line boundary region is significantly higher than that of other regions. The gradient tensor can be used to accurately locate the candidate region of the line boundary.

[0110] 5. Calculation of spectral domain stability factor: Based on the variance of historically acquired multi-band image data, the spectral domain stability factor is calculated. The smaller the variance, the more stable the imaging of that band, and the larger the spectral stability factor; conversely, the larger the variance, the smaller the spectral stability factor. The spectral stability factor can reduce the impact of unstable imaging bands on image enhancement and improve the robustness of the enhancement model.

[0111] 6. Adaptive Enhancement Weight Generation: Combining the band energy ratio function, spectral gradient tensor, and spectral domain stability factor, an enhancement weight function for the spectral domain enhancement model is constructed. In practical applications, adjustment factors can be introduced to perform weighted optimization on each part. The optimized enhancement weight formula is as follows: ,in , , This is a stability adjustment factor that can be fine-tuned based on the actual imaging conditions. For band energy ratio, For local spectral gradients;

[0112] 7. Image Fusion Generation: The generated adaptive enhancement weights are weighted and fused with the normalized multi-band image to generate an enhanced fused image. During the fusion process, the contrast of the line boundary area is maximized, while the noise in the background area is effectively suppressed.

[0113] 8. Accelerated Processing: This spectral domain enhancement process supports GPU parallel acceleration. The fusion algorithm is performed in parallel by the GPU of the image acquisition and processing computer. The processing time for a single frame image is less than 10ms, which meets the efficiency requirements of online detection.

[0114] After the spectral domain enhancement processing in this embodiment, the contrast of the ceramic circuit line boundary area is improved by more than 30%, the image signal-to-noise ratio is improved by at least 20%, and the grayscale difference between the line boundary and the ceramic substrate is significantly increased, providing a clear and high-contrast image basis for subsequent sub-pixel boundary positioning, effectively solving the problem of blurred line boundaries in traditional single-band imaging.

[0115] Example 5: Sub-pixel-level multi-band spatial registration under spectral domain consistency constraints

[0116] Due to factors such as optical deviations in filters of different wavelengths and geometric deviations in camera imaging, there are slight geometric misalignments between multimodal images acquired in different wavelengths. Direct fusion and boundary localization would introduce geometric errors. Therefore, this embodiment uses spectral domain consistency as a constraint to perform sub-pixel-level spatial registration on multimodal images of different wavelengths, achieving cross-wavelength sub-pixel-level geometric alignment. The specific implementation steps are as follows:

[0117] 1. Feature point extraction: Feature points are extracted from the fused images of each band after spectral domain enhancement. Feature points are selected from corners, inflection points and pads of the line boundary. The feature points are required to have high recognition and stability and to exist in all band images. The Harris corner detection algorithm or SIFT feature extraction algorithm is used to accurately extract the feature points.

[0118] 2. Construction of Spectral Domain Consistency Constraint: Using the consistency of spectral response characteristics of the same feature point in images of different bands as a constraint condition, a spectral domain consistency constraint function is constructed. That is, the deviation of the output value of the spectral response coupling matrix of the same feature point in different bands must be less than a preset threshold. This constraint condition can effectively eliminate false feature points and improve the accuracy of feature point matching.

[0119] 3. Feature point matching: Based on the spectral domain consistency constraint function, feature points in images of different bands are precisely matched. The nearest neighbor matching algorithm is combined with distance threshold screening to eliminate incorrect matching pairs and ensure the accuracy of feature point matching.

[0120] 4. Subpixel-level registration parameter solution: The coordinate data of the matched feature point pairs are interpolated to obtain subpixel-level feature point coordinates. Then, based on the subpixel-level feature point coordinates, the least squares method is used to solve the registration parameters such as translation, rotation, and scaling of the image. The accuracy of the registration parameter solution reaches the subpixel level.

[0121] 5. Image Transformation and Resampling: Based on the obtained sub-pixel level registration parameters, affine or projection transformations are performed on the images of each band to achieve geometric alignment of the images. During the transformation process, bilinear interpolation or cubic spline interpolation resampling methods are used to ensure the continuity and clarity of the gray values ​​of the transformed images and avoid image distortion caused by resampling.

[0122] 6. Registration accuracy verification: Select some feature points that were not involved in the registration parameter solution as verification points to verify the geometric deviation after cross-band image registration. The geometric deviation after registration is required to be less than 0.05 pixels, which meets the accuracy requirements of sub-pixel level boundary positioning.

[0123] After subpixel-level multi-band spatial registration in this embodiment, multimodal images of different bands achieve precise geometric alignment, eliminating geometric deviations in cross-band imaging, ensuring the spatial accuracy of subsequent subpixel boundary positioning, and providing a geometrically consistent image foundation for the fusion and analysis of multi-band image features.

[0124] Example 6: Sub-pixel boundary localization based on grayscale curvature continuity and gradient phase consistency

[0125] Subpixel boundary localization is a core step in achieving high-precision measurement of ceramic circuit lines. This embodiment constructs a subpixel boundary localization model based on constraints of grayscale curvature continuity and gradient phase consistency. Through second-order differential extremum analysis, local Gaussian-parabolic joint fitting, and boundary energy function optimization, a boundary localization accuracy better than 0.05 pixels is achieved. A schematic diagram of the subpixel edge localization model is shown in Figure 4. The specific implementation steps are as follows:

[0126] 1. Initial pixel-level edge extraction: For the fused image after spectral domain enhancement and sub-pixel-level spatial registration, the Canny adaptive thresholding algorithm is used to extract the initial pixel-level edges. The Canny algorithm achieves accurate pixel-level edge extraction through steps such as Gaussian filtering for noise reduction, gradient value calculation, non-maximum suppression, double threshold detection and edge connection. The adaptive threshold can be automatically adjusted according to the gray-level distribution of the image, avoiding the problems of incomplete edge extraction or too many false edges caused by fixed thresholds.

[0127] 2. Edge region grayscale sampling: Grayscale sampling is performed along the normal direction of the initial pixel-level edge to obtain the grayscale distribution curve of the edge region. The sampling step size in the normal direction is sub-pixel level to ensure that the sampling points cover the grayscale gradient area of ​​the line edge and fully capture the grayscale change characteristics of the edge.

[0128] 3. Construction of grayscale curvature continuity and gradient phase consistency constraints: Construct grayscale curvature continuity constraints and gradient phase consistency constraints. The grayscale curvature continuity constraint requires that the grayscale curvature of the line boundary changes continuously in space without abrupt changes; the gradient phase consistency constraint requires that the gradient phase of the line boundary remains consistent in space. These two constraints can effectively constrain the shape of the sub-pixel boundary and avoid local deviations in boundary positioning.

[0129] 4. Initial Solution of Sub-pixel Boundaries: The gray-level distribution curve of the edge region is processed, and the initial solution of the sub-pixel boundaries is achieved through second-order differential extremum analysis or local Gaussian-parabolic joint fitting.

[0130] (1) Second-order differential extreme value analysis: Perform second-order differential operation on the gray-scale distribution curve. The extreme point of the second-order differential corresponds to the inflection point of the gray-scale distribution curve, that is, the sub-pixel position of the line boundary. By solving the extreme point of the second-order differential, the initial coordinates of the sub-pixel boundary are obtained.

[0131] (2) Three-point parabolic fitting: Parabolic fitting is performed on three adjacent sampling points of the gray-level distribution curve. The fitting formula is as follows: Then, the extreme points of the fitted parabola are solved to obtain the sub-pixel level boundary coordinates. The calculation formula is as follows: ,in pixel-level edge coordinates , , The gray values ​​of three adjacent sampling points. These are the sub-pixel boundary coordinates;

[0132] 5. Optimization of Boundary Energy Function: Constructing the boundary energy function ,in The actual grayscale value of the image. To fit the grayscale value of the curve, the initial coordinates of the sub-pixel boundary are optimized by minimizing the boundary energy function. The solution process uses the gradient descent algorithm, and combines the constraints of grayscale curvature continuity and gradient phase consistency to avoid the algorithm getting trapped in local optima. Through iterative calculation, the boundary energy function converges to the minimum value.

[0133] 6. Subpixel boundary coordinate output: After optimizing the boundary energy function, the subpixel boundary coordinates of the ceramic circuit are obtained. The positioning accuracy is better than 0.05 pixels, and can reach 0.03 pixels in actual tests. The standard deviation of the subpixel boundary positioning is 0.02 pixels, which is verified by Monte Carlo simulation.

[0134] The subpixel boundary localization method in this embodiment effectively solves the problem that the traditional first-order gradient method cannot handle the gray-level gradient area at the edge of the circuit. By combining the dual constraints of gray-level curvature continuity and gradient phase consistency with second-order differential extremum analysis and Gaussian-parabolic joint fitting, subpixel-level accurate localization of the ceramic circuit line boundary is achieved, providing accurate boundary coordinate data for subsequent high-precision analysis of geometric parameters.

[0135] Example 7: Analysis of Geometric Parameters of Ceramic Circuits

[0136] After obtaining the sub-pixel boundary coordinates of the ceramic circuit lines, this embodiment constructs an analytical model of the circuit geometry parameters to achieve accurate analysis of geometric parameters such as line width, line spacing, line edge roughness, straightness, and pad size. A schematic diagram of the circuit geometry parameter analysis is shown in Figure 5. It supports the processing of curved paths. The specific implementation steps are as follows:

[0137] 1. Boundary point cloud extraction: The sub-pixel boundary coordinates obtained from sub-pixel boundary localization are integrated to extract the boundary point cloud of the ceramic circuit line. The boundary point cloud contains all the sub-pixel coordinate data of the left and right boundaries of the line, providing a data foundation for geometric parameter analysis.

[0138] 2. Line Width Calculation: The line width is calculated using the minimum distance algorithm based on normal projection. This algorithm first projects the point clouds of the left and right boundaries along the normal direction of the line boundary. Then, it calculates the minimum distance between the points on the left and right boundaries after projection, which is the line width at that location. The calculation formula is as follows: ,in The sub-pixel coordinates of the right boundary. The sub-pixel coordinates of the left boundary. To calibrate the pixel-to-physical size conversion ratio in the mapping model, this algorithm can effectively avoid width measurement errors caused by line tilt and supports width measurement of curved paths.

[0139] 3. Line Spacing Calculation: For two adjacent lines, extract the boundary point clouds of the two lines, project them along the common normal direction of the two lines, and calculate the minimum distance between the boundary points of the two lines after projection. This distance is the line spacing, and the calculation formula is as follows: ,in The boundary sub-pixel coordinates of the second line. The boundary sub-pixel coordinates of the first line support line spacing measurement for multiple parallel lines;

[0140] 4. Solving for line edge roughness: The line edge roughness is calculated using the arithmetic mean deviation. To characterize the data, firstly, least-squares line fitting or curve fitting is performed on the sub-pixel boundary point cloud of the line to obtain the fitted curve of the boundary. Then, the deviation value between each boundary point and the fitted curve is calculated. Finally, according to the formula Calculate the line edge roughness, where The number of sampling points is the maximum number of sampling points; the more sampling points, the higher the accuracy of the roughness calculation.

[0141] 5. Straightness calculation: For ceramic circuit lines with straight paths, least squares straight line fitting is performed on its sub-pixel boundary point cloud. At the same time, the RANSAC algorithm is used to remove outliers (such as noise points and line defect points) in the boundary point cloud. Then, the maximum deviation between the boundary point cloud and the fitted straight line is calculated, which is the straightness of the line. The smaller the maximum deviation, the better the straightness of the line.

[0142] 6. Solder pad size calculation: For solder pads of ceramic circuits, extract the sub-pixel boundary point cloud of the solder pad. According to the shape of the solder pad (circle, square, rectangle, etc.), use the corresponding fitting algorithm (such as circle fitting, rectangle fitting) to fit the boundary point cloud, and then solve the dimensional parameters such as the diameter, side length, and area of ​​the solder pad.

[0143] 7. Geometric Parameter Integration: The obtained geometric parameters, such as line width, line spacing, line edge roughness, straightness, and pad size, are integrated and stored in the measurement software's database for subsequent error compensation and result output.

[0144] The geometric parameter analysis model in this embodiment is based on sub-pixel level boundary point cloud data and uses targeted algorithms to achieve accurate analysis of various geometric parameters. It effectively avoids the size errors caused by traditional pixel-level analysis, supports the measurement of curved paths and complex-shaped pads, and meets the diverse geometric parameter detection needs of ceramic circuits.

[0145] Example 8: Construction and Error Compensation of a Multispectral Dynamic Error Compensation Model

[0146] Factors such as spectral response deviation, incident angle variation, material reflectance drift, and equipment vibration in industrial production environments can all introduce errors into the measurement of geometric parameters of ceramic circuits. This embodiment constructs a multispectral dynamic error compensation model. Through real-time spectral monitoring, sensor feedback, and filtering estimation, adaptive error compensation is performed on the measurement results. The flowchart of the dynamic error correction mechanism is shown in Figure 7. The specific implementation steps are as follows:

[0147] 1. Real-time error source monitoring: Through various sensors and a real-time spectral monitoring module, the main error sources affecting measurement accuracy are monitored in real time.

[0148] Spectral response deviation The spectral energy distribution of multi-band light sources is acquired through a real-time spectral monitoring module. Fourier analysis is used to analyze the changes in the spectral energy distribution, and the spectral response deviation is calculated. This reflects the changes in spectral response caused by light source aging and fluctuations in light source brightness.

[0149] Change in incident angle The incident angle of the illumination optical system is monitored in real time using an tilt sensor to obtain the change value of the incident angle. This reflects the deviation in the angle of incidence caused by equipment vibration and light source position shift;

[0150] Material reflectivity drift Based on historically collected reflectivity data of ceramic circuit board materials, Kalman filtering is used to estimate the changes in material reflectivity, thus obtaining the material reflectivity drift. This reflects the changes in reflectivity caused by differences in the microstructure of the ceramic substrate surface and variations in the oxide layer.

[0151] 2. Construction of Multispectral Dynamic Error Compensation Model: Based on the error source parameters obtained from real-time monitoring, a multispectral dynamic error compensation model is constructed. The model is defined as follows: ,in To establish the comprehensive error compensation coefficients, a correction formula for the geometric parameters is also constructed: ,in These are the measured values ​​of geometric parameters after error compensation. These are the original geometric parameter measurements. , , These are the weighting coefficients for each error source, which can be fine-tuned according to the actual measurement scenario;

[0152] 3. Online update of error compensation coefficients: This multispectral dynamic error compensation model supports online updates. By continuously collecting error source parameters and measurement results in the industrial production environment, the model's weight coefficients are optimized through self-learning, continuously improving the accuracy of error compensation and adapting to the dynamic changes in the industrial environment.

[0153] 4. Geometric parameter error compensation: Substitute the original geometric parameter measurement values ​​into the error correction formula, and combine them with the comprehensive error compensation coefficient calculated in real time to perform adaptive error compensation for all geometric parameters such as line width, line spacing, and line edge roughness, thereby eliminating measurement deviations caused by error sources.

[0154] 5. Accuracy verification of compensation: The measurement results after error compensation are verified by using a standard ceramic circuit template. The geometric parameters of the standard template are known values. The measurement deviation after error compensation is required to be controlled within 0.5%, and the repeatability error is less than 1%, which meets the accuracy requirements of industrial precision testing.

[0155] The multispectral dynamic error compensation model in this embodiment constructs a closed-loop error correction mechanism of "real-time monitoring - model calculation - online compensation - self-learning optimization", which effectively reduces the measurement error caused by various interference factors in the industrial production environment, and ensures the repeatability and stability of the measurement results. Even in industrial environments with noise levels as high as 10%, it can still achieve high-precision measurement.

[0156] Example 9: Output of physical dimension measurement results

[0157] After completing the analysis of the geometric parameters of the ceramic circuit and multispectral dynamic error compensation, this embodiment, based on the previously established nonlinear calibration mapping model between the image coordinate system and the physical size coordinate system, converts the sub-pixel level geometric parameter measurements after error compensation into actual physical dimensions and outputs the final measurement results. The specific implementation steps are as follows:

[0158] 1. Transformation from subpixel coordinates to physical dimensions: Substitute the error-compensated subpixel coordinates of the line boundary into the nonlinear calibration mapping model. , and , It achieves high-precision conversion from sub-pixel coordinates to physical size coordinates, with a conversion error of less than 0.1μm;

[0159] 2. Calculation of physical dimension parameters: Based on the converted physical dimension coordinates, the actual physical dimensions of the ceramic circuit lines, such as width, line spacing, line edge roughness, straightness, and pad size, are recalculated. The measurement unit for all parameters is unified to micrometers (μm) to meet the dimensioning requirements of industrial production.

[0160] 3. Measurement Result Integration and Display: The calculated actual physical size parameters are integrated and displayed through the visualization interface of the measurement software. The display content includes the number of the ceramic circuit board under test, the measurement area, the measured values ​​of each geometric parameter, the measurement accuracy, the error range, etc. At the same time, measurement curves (such as the distribution curve of line width and the distribution curve of line edge roughness) can be generated to intuitively reflect the size distribution characteristics of the ceramic circuit line.

[0161] 4. Measurement result storage and traceability: The final physical dimension measurement results are stored in the database of the image acquisition and processing computer in the form of reports (such as Excel and PDF). At the same time, the original data of image acquisition, intermediate data of processing, and relevant parameters of error compensation are linked to realize the full-process traceability of measurement results and meet the quality control requirements of industrial production.

[0162] 5. Online detection data interaction (optional): If the measurement system of the present invention is integrated into the automated production line of ceramic circuits, the measurement results can be exchanged with the manufacturing execution system (MES) of the production line through industrial communication protocols (such as Modbus, Profinet) to realize the real-time uploading and analysis of measurement data. If the measurement results exceed the preset dimensional tolerance range, the system will automatically issue an alarm signal to realize real-time control of the manufacturing quality of ceramic circuits.

[0163] The measurement result output stage of this embodiment achieves accurate conversion from sub-pixel level image data to actual physical dimensions, and also completes the visualization, storage, and traceability of measurement results, meeting the quality control requirements of precision inspection in industrial production. In the case of online continuous scanning measurement scenarios, the processing speed of this system reaches 50 frames / second, supporting a data throughput of hundreds of frames per second, which can meet the online inspection efficiency requirements of mass production of ceramic circuits.

[0164] The present invention also provides a memory that stores multiple instructions for implementing the method as described in Embodiment 1.

[0165] like Figure 8 As shown, the present invention also provides an electronic device, including a processor 301 and a memory 302 connected to the processor 301. The memory 302 stores a plurality of instructions, which can be loaded and executed by the processor to enable the processor to perform the method as described in Embodiment 1.

[0166] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention. Clearly, those skilled in the art can make various alterations and modifications to the invention without departing from its spirit and scope. Thus, if these modifications and modifications of the invention fall within the scope of the claims and their equivalents, the invention is also intended to include these modifications and modifications.

Claims

1. A method for measuring subpixel images of ceramic circuit lines based on multispectral imaging, characterized in that, include: S1 constructs a multi-band composite illumination and imaging system covering the 400-1000nm band, supports different band switching and 0-45 degree incident angle adjustment, and is equipped with a precision motion platform to realize submicron-level positioning and continuous scanning of the ceramic substrate under test; S2, under different wavebands and different incident angles, performs multi-frame averaging acquisition of the ceramic circuit line area to obtain multimodal image data, and suppresses noise through multi-frame fusion to ensure image quality; S3 employs a laser-etched 10μm pitch precision grid calibration plate, collects at least 100 control points, and establishes a nonlinear calibration mapping model between the image coordinate system and the physical size coordinate system to achieve high-precision conversion from pixel size to physical size; wherein, the nonlinear calibration mapping model includes radial and tangential distortion correction, and the conversion error of the high-precision conversion is less than 0.1μm; S4. Construct a multispectral response coupling matrix and a spectral domain enhancement model to maximize the contrast enhancement of the line boundary region; S5, with spectral domain consistency as a constraint, performs sub-pixel-level spatial registration on multimodal images of different bands, eliminates geometric deviations in cross-band imaging, and achieves cross-band sub-pixel-level geometric alignment. S6, Construct a sub-pixel boundary positioning model, and realize the sub-pixel boundary positioning of ceramic circuit lines based on the sub-pixel boundary positioning model; S7, construct the analytical model of line geometric parameters, and determine the line width based on the analytical model of line geometric parameters, and analyze the line spacing, line edge roughness, straightness and pad size; S8. Construct a multispectral dynamic error compensation model. Through real-time spectral monitoring, sensor feedback, and Kalman filtering estimation, obtain spectral response deviation, incident angle change, and material reflectivity drift, respectively, and generate compensation coefficients to adaptively correct the measurement results. The multispectral dynamic error compensation model supports online updates. S9, based on a nonlinear calibration mapping model, converts the sub-pixel level geometric parameters after error compensation into physical dimensions and outputs the final measurement results of the ceramic circuit.

2. The method for measuring sub-pixel images of ceramic circuit lines based on multispectral imaging according to claim 1, characterized in that, The multi-band composite illumination and imaging system includes a multi-band composite light source, a high-resolution camera, a precision motion platform, a calibration module, and an image acquisition and processing unit.

3. The method for measuring sub-pixel images of ceramic circuit lines based on multispectral imaging according to claim 2, characterized in that, S2 includes: S21, the ceramic circuit board to be tested is placed on a precision motion platform; S22 performs multi-frame averaging acquisition of ceramic circuit lines under different sub-bands within the 400-1000nm band and different incident angles of 0-45 degrees, obtains multimodal image data, and stores it in RAW or TIFF format.

4. The method for measuring sub-pixel images of ceramic circuit lines based on multispectral imaging according to claim 3, characterized in that, S4 includes: S41, Based on the difference in spectral reflectance between the ceramic substrate material and the metal conductor in different wavelength bands, a multispectral response coupling matrix is ​​constructed; the multispectral response coupling matrix is ​​defined as: ,in For the first The matrix represents the response intensity of the corresponding material in the specified band. It is initialized using the reflectivity data of the ceramic substrate and the metal conductor obtained experimentally. The matrix elements can be fine-tuned according to the specific material type and metal conductor type of the ceramic circuit board being tested. S42, a spectral domain enhancement model is constructed by combining the band energy ratio function, spectral gradient tensor, and spectral domain stability factor, generating an adaptive fusion factor to maximize the contrast enhancement of the line boundary region; the enhancement weight function of the spectral domain enhancement model... ,in: The band energy ratio function is calculated using the following formula: ,in For the first Energy value of band image The sum of the energy values ​​of all acquired band images. It is the spectral gradient tensor, obtained through convolution operations; The spectral stability factor is calculated based on the variance of historical data.

5. The method for measuring sub-pixel images of ceramic circuit lines based on multispectral imaging according to claim 4, characterized in that, S6 includes: S61 extracts initial pixel-level edges using the Canny adaptive thresholding algorithm; S62, Based on the initial pixel-level edge, construct a sub-pixel boundary localization model based on grayscale curvature continuity and gradient phase consistency constraints; S63, sub-pixel boundary localization of ceramic circuit lines is achieved through second-order differential extremum analysis, local Gaussian-parabolic joint fitting, and boundary energy function optimization; wherein the sub-pixel boundary localization model is achieved by minimizing the boundary energy function, which is: ,in The actual grayscale value of the image. To fit the grayscale values ​​of the curve, the gradient descent algorithm is used in the solution process, combined with constraints on grayscale curvature continuity and gradient phase consistency to avoid local optima. Sub-pixel level boundary solutions are achieved by performing three-point parabolic fitting or second-derivative extremum analysis on the edge grayscale curves. The formula for calculating sub-pixel coordinates using the three-point parabolic fitting is as follows: ,in pixel-level edge coordinates , , The gray values ​​of three adjacent sampling points. These are the sub-pixel boundary coordinates.

6. The method for measuring sub-pixel images of ceramic circuit lines based on multispectral imaging according to claim 5, characterized in that, S7 includes: S71, Construct an analytical model of the line's geometric parameters; S72, Solving for the line width based on the minimum distance algorithm of normal projection; S73, based on the processing of the curved path, performs precise analysis of line spacing, line edge roughness, conductor straightness, and pad size; wherein, the formula for calculating the line edge roughness is: Calculate the line edge roughness, where The number of sampling points. The deviation value between the boundary point and the fitted curve; the straightness of the traverse is obtained by calculating the maximum deviation value after removing outliers through least squares straight line fitting combined with the RANSAC algorithm.

7. The method for measuring sub-pixel images of ceramic circuit lines based on multispectral imaging according to claim 6, characterized in that, The multispectral dynamic error compensation model is defined as follows: ,in This is the comprehensive error compensation coefficient. The spectral response deviation is calculated using Fourier analysis based on real-time spectral monitoring. The change in incident angle is obtained based on sensor feedback from the tilt sensor; Kalman filtering is used to estimate the material reflectivity drift.

8. A subpixel image measurement system for ceramic circuit lines based on multispectral imaging, used to implement the method according to any one of claims 1-7, characterized in that, include: Multi-band composite illumination module is used to build a multi-band composite illumination and imaging system covering the 400-1000nm band. It supports switching between different bands and adjusting the incident angle from 0 to 45 degrees. When paired with a precision motion support platform, it can basically support the ceramic circuit under test and realize sub-micron level positioning and continuous scanning of the ceramic substrate under test. The high-precision imaging acquisition module is used to perform multi-frame averaging acquisition of ceramic circuit lines under different wavelengths and different incident angles to obtain multimodal image data. Noise is suppressed through multi-frame fusion to ensure image quality. The calibration module uses a laser-etched 10μm pitch precision grid calibration plate to collect at least 100 control points and establish a nonlinear calibration mapping model between the image coordinate system and the physical size coordinate system to achieve high-precision conversion from pixel size to physical size. The nonlinear calibration mapping model includes radial and tangential distortion correction, and the conversion error of the high-precision conversion is less than 0.1μm. The multispectral enhancement module constructs a multispectral response coupling matrix based on the difference in spectral reflectance between ceramic substrate materials and metal conductors in different bands. At the same time, it combines the band energy ratio function, spectral gradient tensor and spectral domain stability factor to construct a spectral domain enhancement model and generate an adaptive fusion factor to maximize the contrast enhancement of the line boundary region. The subpixel registration module is used to perform subpixel-level spatial registration of multimodal images in different bands with spectral domain consistency as a constraint, to eliminate geometric deviations in cross-band imaging and achieve cross-band subpixel-level geometric alignment. A subpixel positioning module is used to construct a subpixel boundary positioning model and realize the subpixel boundary positioning of ceramic circuit lines based on the subpixel boundary positioning model. The geometric parameter analysis module is used to construct a geometric parameter analysis model of the line, and determine the line width based on the geometric parameter analysis model of the line, and analyze the line spacing, line edge roughness, straightness and pad size; The dynamic error compensation module is used to construct a multispectral dynamic error compensation model. Through real-time spectral monitoring, sensor feedback, and Kalman filter estimation, it obtains spectral response deviation, incident angle change, and material reflectivity drift, respectively, and generates compensation coefficients to adaptively correct the measurement results. The multispectral dynamic error compensation model supports online updates. The results output module is used to convert the sub-pixel level geometric parameters after error compensation into physical dimensions based on the nonlinear calibration mapping model, and output the final measurement results of the ceramic circuit.

9. An electronic device, characterized in that, It includes a processor and a memory, the memory storing multiple instructions, and the processor being used to read the instructions and execute the method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a plurality of instructions, which can be read by a processor and executed as described in any one of claims 1-7.

Images