An LED circuit board defect recognition method based on optical imaging

Abstract

The application discloses an LED circuit board defect identification method based on optical imaging, relates to the technical field of defect identification, and captures an image of an LED circuit board through an optical imaging device, and then acquires an optical image of the LED circuit board, performs image processing on the optical image, generates a corresponding standard style optical image, performs feature extraction on the standard style optical image, obtains corresponding image optical feature information, constructs a defect identification model, inputs the image optical feature information into the defect identification model, and then classifies and identifies a defect type corresponding to the LED circuit board, feeds back a classification and identification result of the defect identification model on the LED circuit board to an operation terminal corresponding to a production line, controls a mechanical arm arranged on the production line by the operation terminal, and then picks out abnormal LED circuit boards.

Description

Technical Field

[0001] This invention relates to the field of defect identification technology, specifically a method for identifying defects in LED circuit boards based on optical imaging. Background Technology

[0002] With the rapid development of LED technology, LED circuit boards are being used more and more widely in lighting, display screens and other fields. However, during the production process, LED circuit boards may have various defects, such as solder joint defects, circuit breaks, electrical connection defects, etc. These defects seriously affect the quality and service life of the products.

[0003] Furthermore, traditional defect detection methods mainly rely on manual visual inspection after LED circuit board production, which is inefficient and inaccurate, and cannot promptly detect and screen out abnormal LED circuit boards on the production line. Therefore, it is of great significance to study an efficient and accurate LED circuit board defect identification method to detect defective LED circuit boards on the production line in a timely manner and ensure the production quality of LED circuit boards. Summary of the Invention

[0004] To address the aforementioned problems, the present invention aims to provide a method for identifying defects in LED circuit boards based on optical imaging.

[0005] The objective of this invention can be achieved through the following technical solution: a method for identifying defects in LED circuit boards based on optical imaging, comprising the following steps:

[0006] Step S1: The LED circuit board is imaged and captured using an optical imaging device to obtain an optical image of the LED circuit board;

[0007] Step S2: Perform image processing on the optical image to generate a corresponding standard style optical image, extract features from the standard style optical image, and obtain the corresponding image optical feature information;

[0008] Step S3: Construct a defect recognition model and input the image optical feature information into the defect recognition model to classify and identify the defect type corresponding to the LED circuit board;

[0009] Step S4: The defect identification model's classification and identification results for the LED circuit boards are fed back to the corresponding operation terminal on the production line. The operation terminal then controls the robotic arms deployed on the production line to pick out the abnormal LED circuit boards.

[0010] Furthermore, the process of capturing an optical image of the LED circuit board using an optical imaging device includes:

[0011] Set the device parameters for the optical imaging equipment;

[0012] For each processing stage of LED circuit boards, there is a corresponding circuit board storage area on the production line. Optical imaging equipment is arranged in the circuit board storage area on the production line. The circuit board storage area is used to store LED circuit boards.

[0013] The optical imaging equipment is activated to capture images of the LED circuit board, obtaining optical images of the LED circuit board from several angles. When the imaging area captured cannot cover the surface area of ​​the LED circuit board, the layout coordinates of the optical imaging equipment on the production line are adjusted to take pictures in different areas. The regional optical images obtained from the separate area shooting are then stitched together to obtain a complete optical image of the LED circuit board.

[0014] Furthermore, the process of image processing the optical image to generate a corresponding standard style optical image includes:

[0015] Determine whether there is distortion in an optical image, and what type of distortion it is;

[0016] Perform corresponding distortion correction processing on the optical image according to the type of distortion;

[0017] Set up a pixel cell window to divide the optical image into several image sub-regions, and obtain the regional image brightness and regional image contrast of each image sub-region;

[0018] Determine whether the brightness and contrast of a sub-region of an image meet the requirements, and process sub-regions that do not meet the requirements.

[0019] Each optical image is cropped to a preset standard size, thereby generating a standard style optical image.

[0020] Furthermore, the process of determining whether the regional image brightness and regional image contrast of an image sub-region meet the requirements, and processing image sub-regions that do not meet the requirements, includes:

[0021] Image sub-regions are labeled 1, 2, 3, ..., k, where k is a natural number greater than 0. tS[k] represents the brightness of the image sub-region numbered k, and Db[k] represents the contrast of the image sub-region numbered k. Brightness and contrast thresholds are set and denoted as D... 亮度 and D 对比度 ;

[0022] When tS[k]≥D 亮度 At that time, the corresponding image sub-region is marked as a brightness imbalance region, and brightness adjustment is performed. When tS[k] < D 亮度 At that time, no operation is performed;

[0023] When Db[k]≥D 对比度 When the corresponding image sub-region is marked as a contrast-disproportioned region, contrast correction is performed. When Db[k] < D 对比度 At that time, no operation is performed.

[0024] Furthermore, the process of extracting features from the standard optical image to obtain the corresponding image optical feature information includes:

[0025] Key points in a standard style optical map are detected using the SIFT algorithm.

[0026] The edge contours of the LED circuit board and the edge contours of the circuits distributed on the LED circuit board in the standard style optical diagram are identified by the Canny edge detection algorithm and combined as the edge contour information of the standard style optical diagram.

[0027] Feature extraction of standard optical patterns includes shape feature extraction, texture feature extraction, color feature extraction and frequency domain feature extraction, thereby obtaining the geometric attributes, texture information, color distribution information and frequency domain features of the LED circuit board and the circuits distributed on the LED circuit board corresponding to the current standard optical pattern;

[0028] The key points, edge contour information, geometric attributes, texture information, color distribution information, and frequency domain features of the image corresponding to the standard style optical map are integrated and then used as the image optical feature information of the corresponding standard style optical map.

[0029] Furthermore, the process of constructing a defect identification model includes:

[0030] The initial CNN model is constructed using convolutional neural network technology;

[0031] Select LED circuit boards that have been classified into defect types and LED circuit boards without defects, and obtain the optical feature information corresponding to their respective standard style optical images after processing, and construct the information pooling layer and data output layer of the CNN model.

[0032] Optical feature information is input into the information pooling layer to obtain the corresponding main optical features and the pooling coefficients corresponding to the main optical features. The part with the highest data proportion in the optical feature information is taken as the main optical feature, and the defect judgment interval is set.

[0033] Based on the relationship between the pooling coefficient of the main optical features and the defect judgment interval, it is determined whether there is a defect in the current LED circuit board and the type of defect identified. The data output layer outputs the identification result, compares the identification result with the actual result, and decides whether to train the CNN model based on the comparison result, thereby constructing the final defect identification model.

[0034] Furthermore, the process of inputting the image optical feature information into the defect recognition model, and then classifying and identifying the corresponding defect type of the LED circuit board, includes:

[0035] Once the defect identification model is successfully built, the optical feature information of the image corresponding to the currently unclassified LED circuit board is input into the defect identification model to classify and identify the defect types of the LED circuit board. The defect types include solder joint defects, electrical connection defects, circuit defects, and interlayer defects.

[0036] Furthermore, the defect identification model's classification and identification results for LED circuit boards are fed back to the corresponding operation terminal on the production line. The operation terminal then controls the robotic arms deployed on the production line to pick out the abnormal LED circuit boards. This process includes:

[0037] The defect identification model feeds back the identification results in real time to the corresponding operation terminal on the production line, and the operation terminal locates the defective LED circuit board and the corresponding circuit board storage area on the production line.

[0038] Each circuit board storage area is equipped with a corresponding robotic arm. When the identification result indicates that the LED circuit board has a defect, the corresponding LED circuit board is marked as an abnormal LED circuit board. The robotic arm is then controlled to pick out the abnormal LED circuit boards in each circuit board storage area and classify and store them according to the defect type.

[0039] Compared with the prior art, the beneficial effects of the present invention are as follows: optical images of LED circuit boards are acquired through optical imaging equipment, corresponding standard optical images are generated after image processing, and corresponding image optical feature information is obtained after feature extraction. A defect identification model is constructed by inputting the image optical feature information to classify and identify the defect types corresponding to the LED circuit boards. The classification and identification results are fed back to the corresponding operation terminal on the production line in real time. The operation terminal controls the robotic arms arranged on the production line to pick out abnormal LED circuit boards, which effectively improves the efficiency and accuracy of defect identification of LED circuit boards, so as to promptly screen out LED circuit boards with defects, ensure the processing quality of LED circuit boards, and maintain the efficient operation of the LED circuit board production line. Attached Figure Description

[0040] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0041] like Figure 1 As shown, a method for identifying defects in LED circuit boards based on optical imaging includes the following steps:

[0042] Step S1: The LED circuit board is imaged and captured using an optical imaging device to obtain an optical image of the LED circuit board;

[0043] Step S2: Perform image processing on the optical image to generate a corresponding standard style optical image, extract features from the standard style optical image, and obtain the corresponding image optical feature information;

[0044] Step S3: Construct a defect recognition model and input the image optical feature information into the defect recognition model to classify and identify the defect type corresponding to the LED circuit board;

[0045] Step S4: The defect identification model's classification and identification results for the LED circuit boards are fed back to the corresponding operation terminal on the production line. The operation terminal then controls the robotic arms deployed on the production line to pick out the abnormal LED circuit boards.

[0046] It should be further explained that, in the specific implementation process, the process of capturing an image of the LED circuit board using an optical imaging device to obtain an optical image of the LED circuit board includes:

[0047] Set the equipment parameters of the optical imaging device, including lens focal length, exposure time and light source illuminance;

[0048] The optical imaging equipment is arranged in the circuit board storage area on the production line. The production line is used to process LED circuit boards. Each processing stage of the production line that processes LED circuit boards has a corresponding circuit board storage area on its production line.

[0049] The circuit board storage area is used to store LED circuit boards;

[0050] Each LED circuit board stored in the circuit board storage area is labeled with a corresponding circuit board code and storage area code, which serve as a unique authentication identifier for the LED circuit board in its current location.

[0051] When the LED circuit boards on the production line enter the circuit board storage area, the optical imaging equipment is activated and the LED circuit boards are imaged and captured according to the set equipment parameters, thereby obtaining optical images of the LED circuit boards from several angles to cover the entire board area.

[0052] When the imaging area captured by the optical imaging equipment cannot cover the surface area of ​​the LED circuit board, the layout coordinates of the optical imaging equipment on the production line are adjusted to take pictures of the LED circuit board in sections, and the regional optical images obtained by the section pictures are stitched together to obtain a complete optical image of the LED circuit board.

[0053] It should be further explained that, in the specific implementation process, the process of image processing of optical images to generate corresponding standard style optical images includes:

[0054] Determine whether there is distortion in an optical image, and what type of distortion it is;

[0055] And perform corresponding distortion correction processing on the optical image according to the type of distortion;

[0056] If the optical image has distortion, and the distortion type is radial distortion, a distortion model is constructed;

[0057] Obtain distortion parameters, including barrel distortion parameters and pincushion distortion parameters;

[0058] The barrel distortion parameter corresponds to barrel distortion in radial distortion, which manifests as straight lines bulging outward in the optical image. The barrel distortion parameter is a positive value and is denoted as p1, p2, p3, ..., pn, where n is an integer greater than 1.

[0059] Among them, the order of p1-pn gradually increases, p1 is the main control parameter, and p2-pn is used to describe the distortion details in barrel distortion.

[0060] The pincushion distortion parameter corresponds to the pincushion distortion in radial distortion, which manifests as the inward concavity of straight lines in the optical image. The pincushion distortion parameter is negative and is denoted as q1, q2, q3, ..., qm, where m is an integer greater than 1.

[0061] Among them, the order of q1-qm gradually increases, q1 is the main control parameter, and q2-qm is used to describe the distortion details in pincushion distortion.

[0062] Taking barrel distortion in radial distortion as an example, it can be expressed by the following formula:

[0063] X-distorted = X(1 + p1 × r) 2 +p2×r 2 +p3×r 2 +……pn×r 2 );

[0064] Y-distorted = Y(1 + p1 × r) 2 +p2×r 2 +p3×r 2 +……pn×r 2 );

[0065] Where (X, Y) are the coordinates of a point in an ideal, distortion-free optical image, (X-distorted, Y-distorted) are the coordinates of a point in a distorted optical image, and r 2=X 2 +Y 2 It is the square of the distance from the point to the center of the optical image;

[0066] The distortion parameters of the optical image are input into the distortion model, which calculates the ideal position of each pixel in the optical image, thereby correcting the distortion of the optical image, and using bilinear interpolation to calculate the pixel value of each pixel after correction.

[0067] When the distortion type is perspective distortion, the four corner points in the optical image are calibrated, the ideal positions of the four corner points in the optical image are defined, and the four corner points are corrected by the perspective transformation matrix, thereby eliminating perspective distortion;

[0068] When the distortion type is chromatic aberration, the color balance of the optical image is adjusted by using image processing software, and the color deviation is automatically corrected by an algorithm.

[0069] Set up a pixel cell window, which consists of 6x6 pixel cells;

[0070] The optical image is divided into several image sub-regions using pixel cell windows;

[0071] The unit brightness and unit contrast of each pixel cell in each image sub-region are obtained, and the unit brightness and unit contrast of all pixel cells in each image sub-region are accumulated to generate the regional image brightness and regional image contrast of each image sub-region.

[0072] The image sub-regions are labeled 1, 2, 3, ..., k, where k is a natural number greater than 0;

[0073] tS[k] represents the region image brightness of the image sub-region numbered k;

[0074] Db[k] represents the region image contrast of the image sub-region numbered k;

[0075] Set the brightness threshold and contrast threshold, and denote them as D respectively. 亮度 and D 对比度 ;

[0076] When tS[k]≥D 亮度 At that time, the corresponding image sub-region is marked as a brightness imbalance region, and brightness adjustment is performed. When tS[k] < D 亮度 At that time, no operation is performed;

[0077] When Db[k]≥D 对比度 When the corresponding image sub-region is marked as a contrast-disproportioned region, contrast correction is performed. When Db[k] < D 对比度 At that time, no operation is performed;

[0078] Each optical image is cropped to a preset standard size, thereby generating a standard style optical image.

[0079] It should be noted that in practical applications, usually only the first few distortion parameters need to be considered, because the influence of higher-order distortion parameters is smaller and more difficult to estimate accurately through the calibration process. In general, p1, p2, q1, and q2 are sufficient to describe the radial distortion of most lenses. If higher correction accuracy is required, p3, p4, or higher-order distortion parameters may be considered.

[0080] It should be further explained that, in the specific implementation process, the process of extracting features from the standard style optical image to obtain the corresponding image optical feature information includes:

[0081] Image key points in standard style optical maps are detected using the SIFT algorithm. These key points include corner points, edge points, spots, endpoints, T-intersections, and high curvature points.

[0082] Corner point: The point formed by the intersection of two edges in a standard optical diagram;

[0083] Edge points: Points on the edges of a standard optical diagram, used to represent the outline of an LED circuit board;

[0084] Spots: Areas with the same texture or color in a standard optical pattern, used to represent the location distribution of circuits on an LED circuit board;

[0085] Endpoint: The start or end point of an edge in a standard optical diagram;

[0086] T-junction: In a standard optical diagram, the point where one edge segment terminates at the junction of two edge segments;

[0087] The edge contours of the LED circuit board and the edge contours of the circuits distributed on the LED circuit board in the standard style optical diagram are identified by the Canny edge detection algorithm and combined as the edge contour information of the standard style optical diagram.

[0088] Feature extraction of standard optical patterns includes shape feature extraction, texture feature extraction, color feature extraction and frequency domain feature extraction, thereby obtaining the geometric attributes, texture information, color distribution information and frequency domain features of the LED circuit board and the circuits distributed on the LED circuit board corresponding to the current standard optical pattern;

[0089] Geometric properties include area, perimeter, rectangularity, and circularity;

[0090] Texture information is extracted using local binary patterns and gray-level co-occurrence matrices;

[0091] Color distribution information is obtained by calculating color histograms and constructing color moments;

[0092] The frequency domain features are extracted by using Fourier transform to show the changes in the standard pattern diagram in the frequency domain.

[0093] The key points, edge contour information, geometric attributes, texture information, color distribution information, and frequency domain features of the image corresponding to the standard style optical map are integrated and then used as the image optical feature information of the corresponding standard style optical map.

[0094] It should be further explained that, in the specific implementation process, the process of building the defect identification model includes:

[0095] The initial CNN model is constructed using convolutional neural network technology;

[0096] Select LED circuit boards that have been classified into defect types and LED circuit boards without defects, and obtain the optical feature information corresponding to their respective standard style optical images after processing, and construct the information pooling layer and data output layer of the CNN model.

[0097] Optical feature information is input into the information pooling layer to obtain the corresponding main optical features and the pooling coefficients corresponding to the main optical features. The pooling coefficients are used to characterize the overall features of the LED circuit board corresponding to the current standard style optical pattern, and are used for subsequent defect type identification.

[0098] The portion of optical feature information with the highest data proportion is considered the primary optical feature, as described below:

[0099]

[0100] Where D1 is the amount of data corresponding to key points in the image in the optical feature information, D2 is the amount of data corresponding to edge contour information, D3 is the amount of data after the accumulation of geometric attributes, texture information, color distribution information and frequency domain features, and D is the amount of data corresponding to the entire optical feature information.

[0101] Key[Switch] is used to select the optical feature information corresponding to the highest value separated by commas within {} as the main optical feature;

[0102] Set the weighting coefficients a, b, and c;

[0103] Set the correction coefficients α and β;

[0104]

[0105] Where a, b, and c are all natural numbers greater than 0, 0 < α < 1, 0 < β < 1;

[0106] Set the defect judgment range and denote it as Ω. 缺陷 Ω 缺陷 = [0.6, 1);

[0107] When Ch∈Ω 缺陷 When a defect is identified in the LED circuit board, the data output layer outputs the type of defect.

[0108] It should be noted that when determining whether there is a defect in the LED circuit board, further detailed defect types are determined through automatic optical inspection, X-ray inspection, flying probe testing, and functional testing. The identified defect types are then output by a defect identification model. The defect identification model uses the relationship between the defect type of the defective LED circuit board and the corresponding optical feature information as the identification reference data for subsequent defect identification models. Thus, the defect identification model can automatically identify the defect type of the LED circuit board in the future.

[0109] When Ch∉Ω 缺陷 When the current LED circuit board is identified, it is determined that there are no defects, and the identification result is output by the data output layer;

[0110] The identification results are compared with the actual results of LED circuit boards with identified defect types and those without defects to obtain the model's identification accuracy. When the model's identification accuracy is lower than the preset accuracy threshold, the optical feature information of LED circuit boards with identified defect types and those without defects is added as a training set. The CNN model is trained using the training set, and the weight coefficients and correction coefficients are changed until the model's identification accuracy is not lower than the accuracy threshold. At this point, the model training is stopped, and the final defect identification model is constructed.

[0111] It should be noted that the defect judgment interval can be changed according to the actual model verification and identification results. The weight coefficient, correction coefficient, and accuracy threshold can all be changed and adjusted according to the actual model verification results and classification and identification requirements.

[0112] It should be further explained that, in the specific implementation process, the process of inputting the image optical feature information into the defect recognition model, and then classifying and identifying the corresponding defect type of the LED circuit board, includes:

[0113] Once the defect identification model is successfully built, the optical feature information of the image corresponding to the LED circuit board that is currently not classified and identified is input into the defect identification model, and then the defect type of the current LED circuit board is classified and identified. The defect types include solder joint defects, electrical connection defects, circuit defects and interlayer defects.

[0114] Solder joint defects include cold solder joints, excessive solder, insufficient solder, solder balls, and open circuits.

[0115] Specifically as follows:

[0116] Cold welding: The weld joint is not completely melted, resulting in a weak weld;

[0117] Too much or too little solder: affects electrical connections and mechanical strength;

[0118] Solder balls: Excess solder balls formed during the soldering process may cause short circuits;

[0119] Open circuit: The solder joint was not formed, causing the circuit to be broken;

[0120] Electrical connection defects include short circuits and open circuits;

[0121] Specifically as follows:

[0122] Short circuit: An electrical connection occurs where parts of a conductive path that should not be connected become electrically connected.

[0123] Open circuit: The conductive path is interrupted, causing current to be unable to flow;

[0124] Circuit defects include board surface cracks, board holes, and uneven conductor width;

[0125] Specifically as follows:

[0126] Board surface cracks: Cracks appear in the copper foil or circuitry on the LED circuit board;

[0127] Hole in the board: The hole on the LED circuit board is not drilled correctly or the hole wall is defective;

[0128] Uneven conductor width: Inconsistent conductor width may affect the uniform distribution of current;

[0129] Interlayer defects include interlayer short circuits and poor interlayer insulation;

[0130] Specifically as follows:

[0131] Interlayer short circuit: A short circuit occurs in the conductive path between different layers in a multilayer board;

[0132] Poor interlayer insulation: Damage to the interlayer insulation layer leads to a decrease in electrical performance.

[0133] It should be further explained that, in the specific implementation process, the classification and identification results of the defect identification model for LED circuit boards are fed back to the corresponding operation terminal on the production line. The operation terminal then controls the robotic arms deployed on the production line to pick out the abnormal LED circuit boards. The process includes:

[0134] The final defect identification model will feed back the identification results to the corresponding operation terminal on the production line in real time. The operation terminal will locate the defective LED circuit board and the corresponding circuit board storage area on the production line based on the identification results, circuit board code and storage area code.

[0135] Each circuit board storage area is equipped with a corresponding robotic arm. When the identification result indicates that the LED circuit board has a defect, the corresponding LED circuit board is marked as an abnormal LED circuit board, and the robotic arm is controlled to pick out the abnormal LED circuit boards in each circuit board storage area and classify and store them according to the defect type of the LED circuit board.

[0136] It should be noted that each circuit board storage area on the production line corresponds to a processing stage of the LED circuit board. To further determine whether an LED circuit board in the corresponding storage area is abnormal, it is necessary to consider the current processing stage and all previous processing operations. For example, if there is no soldering operation in the current processing stage and all previous processing operations, but the defect type identified by the defect identification model of the LED circuit board in the corresponding storage area is a soldering operation-related defect, then it cannot be determined that the LED circuit board in the current storage area is abnormal. However, if the defect type corresponding to the current processing stage and all previous processing operations exists, then it can be determined that the LED circuit board is abnormal. For example, if the current processing stage is the lamination stage, and the defect identification model identifies a lamination stage-related defect type, then it can be determined that the current LED circuit board is abnormal.

[0137] The above formulas are all numerical calculations after removing dimensions. The formulas are obtained by software simulation based on a large amount of data and are closest to the real situation. The preset parameters and preset thresholds in the formulas are set by those skilled in the art according to the actual situation or obtained by simulation based on a large amount of data.

[0138] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A method for identifying defects in LED circuit boards based on optical imaging, characterized in that, Includes the following steps: Step S1: The LED circuit board is imaged and captured using an optical imaging device to obtain an optical image of the LED circuit board; Step S2: Perform image processing on the optical image to generate a corresponding standard style optical image, extract features from the standard style optical image, and obtain the corresponding image optical feature information; Step S3: Construct a defect recognition model and input the image optical feature information into the defect recognition model to classify and identify the defect type corresponding to the LED circuit board; Step S4: Feed back the classification and identification results of the LED circuit boards by the defect identification model to the corresponding operation terminal on the production line. The operation terminal controls the robotic arm arranged on the production line to pick out the abnormal LED circuit boards. The process of image processing of optical images to generate corresponding standard style optical diagrams includes: Determine whether there is distortion in an optical image, and what type of distortion it is; Perform corresponding distortion correction processing on the optical image according to the type of distortion; If the optical image has distortion, and the distortion type is radial distortion, a distortion model is constructed; Obtain distortion parameters, including barrel distortion parameters and pincushion distortion parameters; The barrel distortion parameter corresponds to barrel distortion in radial distortion, which manifests as straight lines bulging outward in the optical image. The barrel distortion parameter is a positive value and is denoted as p1, p2, p3, ..., pn, where n is an integer greater than 1. Among them, the order of p1-pn gradually increases, p1 is the main control parameter, and p2-pn is used to describe the distortion details in barrel distortion. The pincushion distortion parameter corresponds to the pincushion distortion in radial distortion, which manifests as the inward concavity of straight lines in the optical image. The pincushion distortion parameter is negative and is denoted as q1, q2, q3, ..., qm, where m is an integer greater than 1. Among them, the order of q1-qm gradually increases, q1 is the main control parameter, and q2-qm is used to describe the distortion details in pincushion distortion. Taking barrel distortion in radial distortion as an example, it can be expressed by the following formula: X-distorted=X(1+p1×r2+p2×r2+p3×r2+……pn×r2); Y-distorted=Y(1+p1×r2+p2×r2+p3×r2+……pn×r2); Where (X, Y) are the coordinates of a point in an ideal, distortion-free optical image, (X-distorted, Y-distorted) are the coordinates of a point in a distorted optical image, and r2 = X2 + Y2 is the square of the distance from the point to the center of the optical image; The distortion parameters of the optical image are input into the distortion model, which calculates the ideal position of each pixel in the optical image, thereby correcting the distortion of the optical image, and using bilinear interpolation to calculate the pixel value of each pixel after correction. When the distortion type is perspective distortion, the four corner points in the optical image are calibrated, the ideal positions of the four corner points in the optical image are defined, and the four corner points are corrected by the perspective transformation matrix, thereby eliminating perspective distortion; When the distortion type is chromatic aberration, the color balance of the optical image is adjusted by using image processing software, and the color deviation is automatically corrected by an algorithm. Set up a pixel cell window to divide the optical image into several image sub-regions, and obtain the regional image brightness and regional image contrast of each image sub-region; Determine whether the brightness and contrast of a sub-region of an image meet the requirements, and process sub-regions that do not meet the requirements. Each optical image is cropped to a preset standard size, thereby generating a standard style optical image.

2. The method for identifying LED circuit board defects based on optical imaging according to claim 1, characterized in that, The process of capturing an optical image of an LED circuit board using an optical imaging device includes: Set the device parameters for the optical imaging equipment; For each processing stage of LED circuit boards, there is a corresponding circuit board storage area on the production line. Optical imaging equipment is arranged in the circuit board storage area on the production line. The circuit board storage area is used to store LED circuit boards. The optical imaging equipment is activated to capture images of the LED circuit board, obtaining optical images of the LED circuit board from several angles. When the imaging area captured cannot cover the surface area of ​​the LED circuit board, the layout coordinates of the optical imaging equipment on the production line are adjusted to take pictures in different areas. The regional optical images obtained from the separate area shooting are then stitched together to obtain a complete optical image of the LED circuit board.

3. The method for identifying LED circuit board defects based on optical imaging according to claim 2, characterized in that, The process of determining whether the brightness and contrast of a sub-region of an image meet the requirements, and processing sub-regions that do not meet the requirements, includes: Image sub-regions are labeled 1, 2, 3, ..., k, where k is a natural number greater than 0. tS[k] represents the brightness of the image sub-region numbered k, and Db[k] represents the contrast of the image sub-region numbered k. Brightness and contrast thresholds are set and denoted as D... 亮度 and D 对比度 ; When tS[k]≥D 亮度 At that time, the corresponding image sub-region is marked as a brightness imbalance region, and brightness adjustment is performed. When tS[k] < D 亮度 At that time, no operation is performed; When Db[k]≥D 对比度 When the corresponding image sub-region is marked as a contrast-disproportioned region, contrast correction is performed. When Db[k] < D 对比度 At that time, no operation is performed.

4. The method for identifying LED circuit board defects based on optical imaging according to claim 3, characterized in that, The process of extracting features from a standard optical image to obtain the corresponding image optical feature information includes: Key points in a standard style optical map are detected using the SIFT algorithm. The edge contours of the LED circuit board and the edge contours of the circuits distributed on the LED circuit board in the standard style optical diagram are identified by the Canny edge detection algorithm and combined as the edge contour information of the standard style optical diagram. Feature extraction of standard optical patterns includes shape feature extraction, texture feature extraction, color feature extraction and frequency domain feature extraction, thereby obtaining the geometric attributes, texture information, color distribution information and frequency domain features of the LED circuit board and the circuits distributed on the LED circuit board corresponding to the current standard optical pattern; The key points, edge contour information, geometric attributes, texture information, color distribution information, and frequency domain features of the image corresponding to the standard style optical map are integrated and then used as the image optical feature information of the corresponding standard style optical map.

5. The method for identifying LED circuit board defects based on optical imaging according to claim 4, characterized in that, The process of building a defect identification model includes: The initial CNN model is constructed using convolutional neural network technology; Select LED circuit boards that have been classified into defect types and LED circuit boards without defects, and obtain the optical feature information corresponding to their respective standard style optical images after processing, and construct the information pooling layer and data output layer of the CNN model. Optical feature information is input into the information pooling layer to obtain the corresponding main optical features and the pooling coefficients corresponding to the main optical features. The part with the highest data proportion in the optical feature information is taken as the main optical feature, and the defect judgment interval is set. Based on the relationship between the pooling coefficient of the main optical features and the defect judgment interval, it is determined whether there is a defect in the current LED circuit board and the type of defect identified. The data output layer outputs the identification result, compares the identification result with the actual result, and decides whether to train the CNN model based on the comparison result, thereby constructing the final defect identification model.

6. The method for identifying LED circuit board defects based on optical imaging according to claim 5, characterized in that, The process of inputting image optical feature information into a defect recognition model, and then classifying and identifying the corresponding defect type of the LED circuit board, includes: Once the defect identification model is successfully built, the optical feature information of the image corresponding to the currently unclassified LED circuit board is input into the defect identification model to classify and identify the defect types of the LED circuit board. The defect types include solder joint defects, electrical connection defects, circuit defects, and interlayer defects.

7. The method for identifying LED circuit board defects based on optical imaging according to claim 6, characterized in that, The process of feeding back the classification and identification results of LED circuit boards by the defect identification model to the corresponding operation terminal on the production line, and then controlling the robotic arms deployed on the production line to pick out the abnormal LED circuit boards includes: The defect identification model feeds back the identification results in real time to the corresponding operation terminal on the production line, and the operation terminal locates the defective LED circuit board and the corresponding circuit board storage area on the production line. Each circuit board storage area is equipped with a corresponding robotic arm. When the identification result indicates that the LED circuit board has a defect, the corresponding LED circuit board is marked as an abnormal LED circuit board. The robotic arm is then controlled to pick out the abnormal LED circuit boards in each circuit board storage area and classify and store them according to the defect type.

Images