High-precision size quality inspection device for circuit board production

Abstract

The invention relates to the technical field of electronic manufacturing quality detection, and discloses a high-precision size quality inspection device for circuit board production, which comprises an image acquisition and edge contour extraction module for acquiring an optical image of a circuit board, extracting an edge contour and converting the edge contour into a one-dimensional signal; the adaptive wavelet basis generation module constructs a function library and generates an adaptive wavelet basis by means of a dual-objective optimization algorithm; the wavelet scattering network processing module performs multi-layer wavelet transformation on the signal based on the adaptive wavelet basis to obtain high-order features; the local phase recovery module recovers the local phase of the high-order feature and enhances the weak spectrum feature; the abnormality identification module calculates a feature distance, determines an abnormality, and measures a dimensional deviation. The method has the advantages of high detection precision, strong adaptability, good anti-interference capability and low false alarm rate, and is particularly suitable for small defect detection and high-precision size measurement scenes.

Description

Technical Field

[0001] The present invention relates to the technical field of electronic manufacturing quality inspection, and more specifically, to a high-precision dimension quality inspection device for printed circuit board production. Background Art

[0002] With the development of miniaturization, thinness, and high integration of electronic products, the production and manufacturing of printed circuit boards have higher and higher requirements for dimensional accuracy. Traditional printed circuit board dimension quality inspection methods mainly include manual visual inspection, simple machine vision inspection, and automatic optical inspection (AOI) systems based on spatial domain analysis. Visual inspection has the disadvantages of strong subjectivity, low efficiency, and limited accuracy; simple machine vision inspection can improve efficiency, but the accuracy is not high when dealing with tiny defects and complex backgrounds; the AOI system based on spatial domain analysis has strict requirements for inspection conditions and is prone to missed inspections and false inspections in the case of uneven illumination and surface reflection.

[0003] In the prior art, there are also quality inspection methods using frequency domain analysis and wavelet transform, but these methods have the following deficiencies: using fixed wavelet basis functions, they cannot be optimized according to different printed circuit board characteristics; traditional wavelet analysis is sensitive to geometric transformations (such as rotation and scaling), resulting in unstable detection results; phase information is often ignored in the process of frequency domain feature extraction, reducing the accuracy of edge feature representation; the dimension measurement process relies on simple reference template matching and is difficult to adapt to complex backgrounds and diverse printed circuit board styles.

[0004] Therefore, there is an urgent need to propose a printed circuit board dimension quality inspection device that can adapt to different production conditions and has high precision and stability. Summary of the Invention

[0005] The present invention provides a high-precision dimension quality inspection device for printed circuit board production, which combines adaptive wavelet basis generation, wavelet scattering network processing, local phase recovery and feature enhancement, and anomaly recognition technology based on frequency domain features, and can effectively improve the detection accuracy and measurement stability of tiny defects on the edges of printed circuit boards.

[0006] The present invention provides a high-precision dimension quality inspection device for printed circuit board production, including: An image acquisition and edge contour extraction module, configured to acquire an optical image of a printed circuit board, extract an edge contour from the optical image, and convert it into a one-dimensional signal sequence; An adaptive wavelet basis generation module, configured to construct a wavelet basis function library and generate an adaptive wavelet basis function through a bi-objective optimization algorithm; A wavelet scattering network processing module, configured to perform multi-layer wavelet transform on the one-dimensional signal sequence based on the adaptive wavelet basis function to obtain a high-order feature representation; The local phase recovery and feature enhancement module is used to perform local phase recovery on high-order features, obtain the local phase consistency metric, and enhance weak spectral features based on the local phase consistency metric; The anomaly recognition and measurement module is used to calculate the feature distance metric and determine anomalies, and accurately measure the deviation between the actual size and the standard size.

[0007] Furthermore, the objective function of the bi-objective optimization algorithm is: ; Wherein, is the wavelet basis function to be optimized; is the trade-off parameter; is the reconstruction error term, and the smaller it is, the higher the reconstruction accuracy; represents the negative value of the energy concentration, making it the same minimization objective as the reconstruction error.

[0008] Furthermore, the steps for performing multi-layer wavelet transform on the one-dimensional signal sequence are as follows: Apply the Canny edge detection algorithm to process the original image and extract the edge contour of the circuit board; Convert it into a one-dimensional signal sequence using the chain code representation method, and record the direction changes of adjacent pixel points point by point along the edge contour; Normalize the one-dimensional signal sequence to eliminate the influence of the contour starting point selection and scale change.

[0009] Furthermore, in the wavelet scattering network processing module: perform the first-layer wavelet transform on the input contour signal to obtain the coefficients: ; Wherein, is the result of the first-layer wavelet transform, represents the input contour signal ; is the wavelet function adaptively generated in the first layer; ; represents the scale and the direction parameter pair; represents the convolution operation; represents the modulo operation, introducing non-linearity; Perform modulo operation and low-pass filtering on the output result of the first layer, and then perform the second-layer wavelet transform to obtain the coefficients: ; Wherein, is the result of the second-layer wavelet transform, is the second-layer wavelet function; ; represents the scale of the second layer and the direction Parameter pair; And so on, execute until the layer to form the final wavelet scattering feature.

[0010] Furthermore, in the local phase recovery and feature enhancement module, the key phase information is recovered by solving an optimization problem: ; Where is the recovered phase information, represents the scattering coefficient of the signal , represents the scattering coefficient of the original signal .

[0011] Furthermore, the local phase consistency metric is: ; Where represents the phase consistency metric at the frequency point , with a value range of [0, 1]; represents the frequency domain representation within the th local window; is the number of windows; represents the modulus operation of complex numbers.

[0012] Furthermore, the enhancement of weak spectral features is achieved in the spectral enhancement filter, and its mapping function is: ; Where is , representing the phase consistency metric at the frequency point , , respectively represent the horizontal and vertical frequency coordinates of the point in the frequency domain; and are parameters determined according to the circuit board material and the detection accuracy requirements, with a typical value of 1.5, with a typical value of 2.0.

[0013] Furthermore, the formula for calculating the feature distance metric is: ; Where is the feature vector of the test sample, with a dimension of ; is the feature vector of the reference model, with a dimension of ; is the feature vector of the th test sample; For the feature vector of the

[0014] Further, for determining the abnormal area, a sub-pixel precision edge positioning algorithm is adopted to calculate the deviation between the actual size and the standard size, and its precision can reach 1 / 10 of the pixel size, and a measurement precision of ±0.002 mm can be achieved in the printed circuit board detection.

[0015] The present invention provides a storage medium that stores non-temporary computer-readable instructions, which can execute the modules in a high-precision size quality inspection device for printed circuit board production as described above when executed by a computer.

[0016] The beneficial effects of the present invention are as follows: Through the optimized selection of the adaptive wavelet basis function and the combination of the wavelet scattering network, the present invention significantly improves the printed circuit board detection performance. In terms of detection accuracy, it can detect minute contour abnormalities of 0.01 mm, improving the size measurement accuracy; in terms of system adaptability, the adaptive algorithm can automatically adjust the detection parameters for different printed circuit boards, reducing the manual workload; the anti-interference ability is improved, and the geometric invariant features ensure that the measurement results are not affected by geometric transformations and shooting conditions; with the help of local phase recovery and spectrum enhancement technologies, the false alarm rate is reduced, and the tolerance for edge irregularities is increased. In addition, this technology can adapt to the changing environment on the production line, improving the quality inspection efficiency and accuracy. The present invention is applicable to high-precision size quality inspection of various printed circuit boards, can effectively identify subtle defects, and provides strong support for improving the manufacturing precision and reliability of electronic products. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a module diagram of a high-precision size quality inspection device for printed circuit board production according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0018] Now, the subject matter described herein will be discussed with reference to exemplary embodiments. It should be understood that discussing these embodiments is only to enable those skilled in the art to better understand and thus implement the subject matter described herein, and the functions and arrangements of the elements discussed can be changed without departing from the protection scope of the content of this specification. Each example can omit, substitute, or add various processes or components as needed. Additionally, the features described in some examples can also be combined in other examples.

[0019] In at least one embodiment of the present invention, a high-precision size quality inspection device for printed circuit board production is disclosed, as Figure 1 shown, including: An image acquisition and edge contour extraction module, configured to acquire an optical image of a printed circuit board, extract an edge contour from the optical image, and convert it into a one-dimensional signal sequence; Specifically including: Step 1.1: Collect the optical image of the circuit board to obtain high-resolution original image data, denoted as matrix , where represents the pixel coordinates.

[0020] Step 1.2: Apply the Canny edge detection algorithm to process the original image and extract the edge contour of the circuit board; The specific operation steps of the Canny algorithm are as follows: Apply a Gaussian filter to smooth the image, and its calculation formula is: ; Obtain the smoothed image : ; Among them, represents the convolution operation, The value is adaptively adjusted according to the image resolution : ; Calculate the gradient magnitude and direction of the image, and its calculation formula is: ; ; Among them, and are the gradients in the and directions respectively; Perform non-maximum suppression to retain the pixel points with the maximum local gradient; Apply double-threshold processing, where the low threshold , the high threshold , is the average gray value of the image; Connect the edges through the hysteresis threshold method to obtain the complete edge contour ; Step 1.3: Convert the extracted two-dimensional edge contour into a one-dimensional signal sequence using the chain code representation method; Select the starting point on the edge contour; Record the direction changes of adjacent pixel points point by point along the edge contour, and encode the direction using 8-connected chain codes; Form a one-dimensional direction sequence , where represents , , , respectively represent the , th, , Direction encoding of a point relative to the previous point, is the total number of points; Chain code calculation formula: , where The function maps the relative position to the corresponding direction encoding; , , , respectively represent the , -th point's horizontal and vertical coordinates.

[0021] Step 1.4, perform normalization processing on the one-dimensional signal sequence to eliminate the influence of contour starting point selection and scale change; Differential chain code Calculation: ; Shape factor Calculation: ; is used to characterize the complexity of the contour; After normalization processing, the one-dimensional contour signal of the -th point is obtained : ; Finally, the standardized one-dimensional contour signal is obtained, where , , respectively represent the one-dimensional contour signals of the , , -th points, is the total number of points, and this signal has invariance to contour starting point selection and scale change, providing stable input data for subsequent adaptive wavelet basis generation and wavelet scattering network processing.

[0022] Adaptive wavelet basis generation module, used to construct a wavelet basis function library and generate adaptive wavelet basis functions through a two-objective optimization algorithm; Specifically includes: Step 2.1, construct a wavelet basis function library , where , , respectively represent the , , -th wavelet basis functions, is the total number of wavelet basis functions.

[0023] Step 2.2: For each standard wavelet basis function, calculate its reconstruction error for the circuit board contour signal: ; where, represents the input circuit board contour signal, a vector with dimension ; represents the th wavelet basis function; represents the transpose of , used for the inverse wavelet transform; represents the convolution operation; represents the signal under the wavelet basis transformation coefficient; represents reconstructing the signal using the wavelet basis ; represents the square of the Euclidean norm, indicating the mean square error between the original signal and the reconstructed signal.

[0024] Step 2.3: Calculate the energy concentration of each wavelet basis function for the contour signal: ; where, represents the transformation coefficient vector of the signal under the wavelet basis ; represents the th element in the transformation coefficient vector; the first coefficients containing the main energy, usually taking ; represents the total number of coefficients after wavelet transform, equal to the length of the signal ; is the square of the coefficient amplitude, indicating energy; The larger this ratio, the more concentrated the energy is in a few coefficients, and the stronger the wavelet basis's ability to represent the signal. Step 2.4: Apply the bi-objective optimization algorithm to solve the objective function: ; where, is the wavelet basis function to be optimized; is the trade-off parameter, used to balance the reconstruction accuracy and energy concentration, with a value range of 0.4 to 0.6. When it is close to 0.4, more attention is paid to energy concentration, which is suitable for detecting edge mutation features. When it is close to 0.6, more attention is paid to reconstruction accuracy, which is suitable for detecting smooth edge features; is the reconstruction error term, the smaller it is, the higher the reconstruction accuracy; Represents the negative value of the energy concentration, making it the same minimization objective as the reconstruction error. Through this optimization process, an adaptive wavelet basis function most suitable for the current PCB characteristics is obtained. , which can maximize the energy concentration while ensuring the reconstruction accuracy, improving the expression ability of the PCB edge features.

[0025] The wavelet scattering network processing module performs multi-layer wavelet transform on the one-dimensional signal sequence based on the adaptive wavelet basis function to obtain high-order feature representations; Specifically including: Step 3.1, construct a wavelet scattering network with a network depth of layers, each layer containing rotation-equivariant wavelet filters. In the PCB detection scenario, the structure of the wavelet scattering network includes: the input layer receives the normalized one-dimensional contour signal; the intermediate layer performs multi-scale wavelet transform, non-linear activation, and downsampling operations; the output layer generates feature representations invariant to geometric transformations.

[0026] Step 3.2, perform the first-layer wavelet transform on the input contour signal to obtain the coefficients: ; where represents the input contour signal ; is the wavelet function adaptively generated in the first layer; represents the scale and the direction parameter pair; represents the convolution operation; represents the modulo operation, introducing non-linear characteristics.

[0027] This transform can capture the low-level features of the PCB edge (such as break points, continuous segments), and is particularly suitable for detecting edge breaks and small protrusions in the PCB; Step 3.3, perform the modulo operation and low-pass filtering on the output result of the first layer, and then perform the second-layer wavelet transform to obtain the coefficients: ; where is the wavelet function of the second layer; represents the scale and the direction parameter pair of the second layer.

[0028] This operation can extract more advanced morphological features, such as the curvature change and local geometric structure of the PCB edge; Step 3.4, and so on, execute until the Layer, forming the final wavelet scattering features, including scattering coefficients and wavelet modulation coefficients: ; ; Among them, is the -th order scattering coefficient; is the -th order wavelet modulation coefficient; represents the low-pass filter with scale and is defined as ; represents the scale and direction parameters of the -th layer.

[0029] These features are stable against geometric transformations (such as rotation, scale change) and can more accurately characterize the characteristics of the PCB edge. In practical applications, for PCB edge structures with different complexities, the network depth can be adaptively adjusted. For complex structures, can be taken, and for simple structures, can be taken to balance the computational complexity and feature extraction ability.

[0030] Local phase recovery and feature enhancement module, used to perform local phase recovery on high-order features, obtain the local phase consistency metric, and enhance weak spectral features based on the local phase consistency metric; Specifically includes: Step 4.1, perform local phase recovery on the wavelet scattering features, and recover the key phase information by solving an optimization problem: ; Among them, is the recovered phase information; represents the scattering coefficient of the signal ; represents the scattering coefficient of the original signal ; is the square of the Euclidean distance.

[0031] This step is crucial for the recovery of the fine structure of the PCB edge, especially for the detection of minute defects (such as line width variation, angle deviation).

[0032] Step 4.2, calculate the local phase consistency metric in the frequency domain representation: ; Among them, represents the phase consistency metric at the frequency point , and the value range is [0, 1]; represents the The frequency-domain representation within a local window; is the number of windows; represents the modulus operation of complex numbers.

[0033] Step 4.3, generate a spectral enhancement filter through a mapping function according to the phase consistency metric: ; where, represents the enhancement coefficient at the frequency point ; represents ; represents the mapping function that maps the phase consistency value to the enhancement coefficient; represents the enhancement intensity parameter that controls the overall enhancement amplitude; represents the non-linear exponent parameter that controls the non-linearity degree of the enhancement.

[0034] Spectral enhancement filter design steps: Select appropriate and values according to the material characteristics of the circuit board (take a smaller value for reflective materials and a larger value for matte materials); for each frequency point , calculate the enhancement coefficient ; normalize the enhancement coefficient to ensure that the maximum value does not exceed a preset upper limit (usually 3.0).

[0035] Step 4.4, apply the spectral enhancement filter to process the wavelet scattering features, ; ; where, is the frequency-domain representation of the original feature; is the enhanced frequency-domain representation; is the enhanced spatial-domain feature representation; represents the inverse Fourier transform.

[0036] Strengthen weak spectral features to form an enhanced feature representation. For circuit board edge detection, this step can particularly enhance the detection ability of small burrs, fractures, and irregular edges, and maintain the stability of the detection effect even under uneven illumination and surface reflection conditions.

[0037] Anomaly recognition and measurement module, used to calculate the feature distance metric and determine anomalies, and accurately measure the deviation between the actual size and the standard size; Specifically includes: Step 5.1, establish a standard circuit board reference model library, which contains the standard feature representations of different types of circuit boards. In practical applications, for different models of PCB boards, standard samples are collected under multiple angles and lighting conditions, and their feature representations are extracted and stored in the reference model library. For complex circuit boards, they can be divided into multiple regions for individual modeling to improve the detection accuracy; Step 5.2, calculate the distance metric between the enhanced features of the circuit board to be tested and the reference model: ; where, is the feature vector of the test sample, with a dimension of ; is the feature vector of the reference model, with a dimension of ; is the dimension of the feature vector; is the -th feature vector of the test sample; is the -th feature vector of the reference model.

[0038] This metric method is particularly suitable for detecting small size deviations of circuit boards because it equally emphasizes each component in the feature vector and can capture subtle shape differences; Step 5.3, set a threshold (determined according to the production accuracy requirements). When the distance metric exceeds the threshold, it is determined that the size is abnormal: ; where, is the abnormality mark (1 indicates abnormality, 0 indicates normal) at position ; is the distance metric at position ; is the threshold at position , which is dynamically set according to the importance of the region.

[0039] In the production application of circuit boards, different thresholds can be set for different types of abnormalities. For example, for the line width deviation of key connection parts, the threshold can be set to a more stringent value (such as 0.005 mm), while for non-critical areas, it can be appropriately relaxed (such as 0.01 mm); Step 5.4, for the detected abnormal regions, use a sub-pixel precision edge localization algorithm to accurately calculate the deviation between the actual size and the standard size, and output the measurement result: ; ; where, is the edge response function; is the gray - level gradient of the image at position ; is the weight coefficient, usually a Gaussian distribution ; is the initial edge - position estimate; is the exact edge position; and are respectively 's first - order and second - order derivatives.

[0040] This algorithm calculates the edge position by using the difference of the image gray - level gradient. The theoretical accuracy can reach 1 / 10 of the pixel size, and usually can achieve a measurement accuracy of ±0.002mm in printed circuit board detection. The system finally outputs the detailed information of the position, type, and deviation size of the abnormal area, which is convenient for the production line to adjust the process parameters in time.

[0041] This embodiment realizes multi - aspect technological breakthroughs by means of frequency - domain feature analysis, wavelet transform, adaptive wavelet - basis selection algorithm, wavelet scattering network, and spectrum enhancement technology. In terms of detection accuracy, it can sensitively capture contour abnormalities of 0.01mm, improving the size - measurement accuracy; the system adaptability is significantly enhanced, and it can adapt to the edge characteristics of different types of printed circuit boards; relying on the geometric - invariant features constructed by the wavelet scattering network, the anti - interference ability of the system is improved, and the measurement results are not affected by geometric transformations and shooting conditions; with the help of spectrum enhancement and anomaly - detection algorithms, the false - alarm rate is reduced, and the tolerance of edge irregularity is improved; in addition, this method can also meet the detection requirements of printed circuit boards with different materials and structures on the production line, improving the quality - inspection efficiency and accuracy, and meeting the quality - control requirements of high - precision printed - circuit - board production.

[0042] In an embodiment of the present invention, an example of the aforementioned high - precision size quality - inspection device for printed - circuit - board production is provided: Application - scenario description: Applied tests were carried out on the high - density interconnect (HDI) printed - circuit - board production line of an electronic manufacturing enterprise. This production line needs to detect more than 5000 printed - circuit boards of different models every day, with size tolerances reaching ±0.005mm, and has extremely high requirements for the accuracy and efficiency of quality inspection. The application scenario is the quality - inspection link of the HDI printed - circuit - board production line, and the main detection objects include: the line width and line spacing of high - precision multi - layer printed - circuit boards; the aperture and position of micro vias and buried vias; the precise measurement of pad size and shape; the regularity detection of edge contours and cutting boundaries. All these detection objects generally have small burrs, irregular edges, and slight deformations caused by process fluctuations, and traditional measurement methods are difficult to cope with these challenges.

[0043] The specific steps are as follows: Image acquisition and preprocessing: Use a 12-megapixel industrial camera and cooperate with a ring-shaped LED light source to collect circuit board images; set the image resolution to 4096×3072 pixels, corresponding to a physical resolution of 0.003 mm / pixel; use adaptive histogram equalization to improve image contrast; suppress noise through Gaussian filtering.

[0044] Edge contour extraction and signal conversion: Apply the Canny operator (low threshold 30, high threshold 90) to extract the initial edge; use morphological operations (opening operation, structural element is a 3×3 rectangle) to remove noise points; sample along the edge contour in a clockwise direction to generate a one-dimensional signal sequence containing 8000 - 12000 points; normalize and resample the signal sequence to ensure uniform sampling point spacing.

[0045] Adaptive wavelet basis generation: Select the initial wavelet basis from the preset wavelet library; set the optimization parameters: α = 0.5, number of iterations = 200, convergence threshold = 0.001; execute the bi-objective optimization algorithm to generate the optimal wavelet basis for different types of edge features; verify the optimization results: the reconstruction error is reduced by 47% - 56%, and the energy concentration is increased by 14% - 26%, as shown in Table 1: Table 1: Results of adaptive wavelet basis optimization for different types of circuit board features:

[0046] Wavelet scattering network construction and feature extraction: Configure the network parameters: depth P = 3, number of filters per layer J = 8, rotation angle interval = 22.5°; perform multi-layer wavelet transform on the one-dimensional signal sequence to generate scattering coefficients; extract the invariant feature representation to form a 512-dimensional feature vector; reduce the dimension to 128 dimensions through principal component analysis, retaining more than 95% of the information, as shown in Table 2: Table 2: Optimal parameter configuration of the wavelet scattering network for detection tasks with different accuracy requirements:

[0047] Local phase recovery and feature enhancement: Process the feature vector in segments, with each segment having a length of 64 points; execute the local phase recovery algorithm, with the number of iterations set to 50; calculate the local phase consistency metric, with the window size set to 16×16; construct a spectral enhancement filter: β = 1.5, γ = 2.0; apply the filter to enhance weak edge features, with the enhancement multiple reaching 2.5 - 3.8 times, as shown in Table 3: Table 3: Influence of spectral enhancement filter parameter adjustment on the detection performance of weak edge features:

[0048] Anomaly Recognition and Precise Measurement: Calculate the Euclidean distance between the computed enhanced features and the reference model; Set dynamic thresholds: τ = 0.005 mm for critical areas and τ = 0.008 mm for general areas; Apply the sub-pixel edge localization algorithm to the detected anomaly areas; Execute the Newton iteration method (number of iterations = 5) to calculate the precise edge positions; Measure the deviation between the actual size and the standard size, with an average accuracy of ±0.007 mm under different test conditions, as shown in Table 4: Table 4: Comparison of the detection accuracy and stability of this method and traditional methods under different challenge conditions:

[0049] Robustness Test for Geometric Transformations and Illumination Changes: Apply different degrees of rotation transformations (±10°) to the test samples; Apply different degrees of scaling transformations (±5%) to the test samples; Simulate image acquisition under different illumination conditions (illumination change ±30%); Measure using traditional methods and this method respectively; Calculate the deviation rate and the percentage increase in consistency of the measurement results; The results show that this method maintains high stability under various transformation conditions. Compared with traditional methods, the consistency of this method under rotation, scaling, and illumination change conditions has increased by 78.7%, 77.0%, and 77.2% respectively, as shown in Table 5: Table 5: Detection consistency of this method under different transformation conditions:

[0050] Verification of Technical Effects: The application benefits of this method on production lines of different types of circuit boards are significant, as shown in Table 6: Table 6: Application benefits of this method on production lines of different types of circuit boards:

[0051] Through the above implementation steps, this method has achieved high-precision and high-stability quality inspection of circuit board dimensions on actual production lines, significantly improving production efficiency and product quality. Especially for high-density interconnect circuit boards, the detection accuracy has increased by 36.8%, the false detection rate has decreased by 48.6%, the detection rate has increased by 51.2%, and the quality inspection efficiency has increased by 45.8%, meeting the strict requirements of modern electronic manufacturing for high-precision quality inspection.

[0052] The above describes the embodiments of the present invention, but these embodiments are not limited to the above specific implementation manners. The above specific implementation manners are merely illustrative and not restrictive. Under the inspiration of this embodiment, those of ordinary skill in the art can also make more equivalent embodiments in various forms, all of which fall within the protection scope of this embodiment.

Claims

1. A high-precision dimensional quality inspection device for circuit board production, characterized in that, including: An image acquisition and edge contour extraction module, configured to acquire an optical image of a circuit board, extract an edge contour from the optical image, and convert it into a one-dimensional signal sequence; An adaptive wavelet basis generation module, configured to construct a wavelet basis function library and generate an adaptive wavelet basis function through a bi-objective optimization algorithm; A wavelet scattering network processing module, configured to perform multi-layer wavelet transform on the one-dimensional signal sequence based on the adaptive wavelet basis function to obtain a high-order feature representation; A local phase recovery and feature enhancement module, configured to perform local phase recovery on the high-order feature to obtain a local phase consistency metric, and enhance weak spectral features based on the local phase consistency metric; An anomaly recognition and measurement module, configured to calculate a feature distance metric and determine anomalies, and accurately measure the deviation between the actual size and the standard size.

2. The high-precision dimension quality inspection device for circuit board production according to claim 1, characterized in that, The objective function of the bi-objective optimization algorithm is: ； Among them, is the wavelet basis function to be optimized; is the trade-off parameter; is the reconstruction error term, and the smaller it is, the higher the reconstruction accuracy; represents the negative value of the energy concentration, making it the same minimization objective as the reconstruction error.

3. A high-precision dimensional quality inspection device for circuit board production according to claim 1, characterized in that, The steps of performing multi-layer wavelet transform on the one-dimensional signal sequence are: Apply the Canny edge detection algorithm to process the original image and extract the edge contour of the circuit board; Convert it into a one-dimensional signal sequence using the chain code representation method, and record the direction change of adjacent pixel points point by point along the edge contour; Perform normalization processing on the one-dimensional signal sequence to eliminate the influence of the contour starting point selection and scale change.

4. A high-precision dimensional quality inspection device for circuit board production according to claim 1, characterized in that, In the wavelet scattering network processing module: perform the first-layer wavelet transform on the input contour signal to obtain coefficients: ； Among them, is the result of the first-layer wavelet transform, represents the input contour signal ; is the wavelet function adaptively generated in the first layer; represents the scale and the direction parameter pair; represents the convolution operation; represents the modulo operation, introducing non-linear characteristics; Perform modulus operation and low-pass filtering on the output result of the first layer, and then perform the second-layer wavelet transform to obtain coefficients ； Among them, is the result of the second-level wavelet transform, is the second-level wavelet function; represents the scale of the second level and direction parameter pair; And so on, until the layer, to form the final wavelet scattering features.

5. The high-precision dimension quality inspection device for circuit board production according to claim 1, characterized in that, In the local phase recovery and feature enhancement module, the key phase information is recovered by solving an optimization problem: ； Among them, is the restored phase information, represents the signal scattering coefficient, represents the original signal scattering coefficient.

6. The high-precision dimension quality inspection device for circuit board production according to claim 1, wherein, The local phase consistency metric is: ； Among them, represents the phase consistency metric at the frequency point with a value range of [0, 1]; represents the frequency domain representation within the th local window; is the number of windows; represents the modulus operation of complex numbers.

7. A high-precision dimensional quality inspection device for circuit board production according to claim 1, characterized in that, The enhancement of weak spectral features is implemented in a spectral enhancement filter, and its mapping function is: ； Among them, is , representing the phase consistency metric at the frequency point . and respectively represent the horizontal and vertical frequency coordinates of the point in the frequency domain; and are parameters determined according to the circuit board material and the detection accuracy requirements, with a typical value of 1.5, and a typical value of 2.

0.

8. The high-precision dimensional quality inspection device for circuit board production according to claim 1, wherein, The formula for calculating the feature distance metric is: ； Among them, is the feature vector of the test sample, with a dimension of ; is the feature vector of the reference model, with a dimension of ; is the feature vector of the th test sample; is the feature vector of the th reference model.

9. The high-precision dimension quality inspection device for circuit board production according to claim 1, wherein, For the determined anomaly region, a sub-pixel accuracy edge localization algorithm is adopted to calculate the deviation between the actual size and the standard size, and its accuracy can reach 1 / 10 of the pixel size, and a measurement accuracy of ±0.002 mm can be achieved in the circuit board detection.

10. A storage medium stores non-transitory computer-readable instructions, characterized in that, When the non-transitory computer-readable instructions are executed by a computer, they can execute the modules in a high-precision size quality inspection device for circuit board production as described in any one of claims 1-9.

Images