A system and method for detecting defects of live pin plugging holes

Abstract

The present invention discloses a system and method for detecting defects in live pin plug holes, which belongs to the field of internal combustion engine manufacturing and quality inspection, and includes: an image acquisition module, including a multi-spectral imaging device, for capturing images of live pin plug holes in different spectral ranges; an image analysis module, for performing multi-dimensional processing and analysis on the acquired images to identify defect types and locations; an evaluation module, for performing severity evaluation on detected defects based on data provided by the image analysis module; and a data output module, for displaying defect detection results and severity ratings. The present invention automatically adjusts spectral settings through an adaptive selection controller; and adopts automatic parameter adjustment based on local image characteristics to provide more accurate and responsive image processing capabilities. By integrating deep learning and user feedback, it not only demonstrates high efficiency during initial deployment, but also continuously optimizes its performance, which is an innovative application rarely seen in industrial inspection systems.

Description

Technical Field

[0001] The invention relates to the technical field of internal combustion engine manufacturing and quality inspection, and in particular to a live pin plug hole defect inspection system and method. Background Art

[0002] The pinhole is a key component in internal combustion engines and other mechanical devices, and its integrity is critical to the performance of the entire machine. Defects in the pinhole, such as cracks, wear or deformation, may cause mechanical failure or performance degradation. Traditional defect detection methods mainly rely on manual visual inspection or simple automated image processing technology, which often cannot accurately identify small or complex defects and are inefficient and cannot meet the needs of high-speed production lines.

[0003] In addition, existing technologies generally lack flexibility in processing multiple spectral data and have difficulty automatically adjusting detection settings according to different types of defects and changing environmental conditions. This limits the applicability and reliability of the detection system, especially in a changing industrial environment.

[0004] Existing technologies also often lack effective data output and interactive interfaces, making it difficult for operators to understand and operate the detection system, thereby reducing the accuracy and efficiency of the detection. Summary of the invention

[0005] In view of the shortcomings of the prior art, the present invention proposes a system and method for detecting live pin plugging holes, which utilizes multi-spectral imaging technology to automatically select the optimal spectrum according to the defect type and ambient light conditions through an improved adaptive spectrum selection controller, and automatically adjusts the filtering parameters through adaptive image preprocessing and optimized edge enhancement filtering algorithms. By utilizing advanced convolutional neural networks and support vector machine models, the system can accurately classify and rate defects, thereby improving the accuracy and reliability of the evaluation.

[0006] In order to achieve the above object, the present invention is implemented by the following technical solutions:

[0007] A live pin plug hole defect detection system, comprising:

[0008] An image acquisition module, including a multispectral imaging device, for capturing images of the piston pin plug hole in different spectral ranges;

[0009] An image analysis module, used to perform multi-dimensional processing and analysis on the collected images to identify defect types and locations;

[0010] An evaluation module for rating the severity of defects based on image analysis results;

[0011] The data output module is used to evaluate the severity of the detected defects based on the data provided by the image analysis module; wherein the multispectral imaging device enhances the system's detection capability for tiny defects and improves detection accuracy through an improved adaptive multispectral image fusion algorithm.

[0012] As a preferred embodiment of the present invention, the multispectral imaging device comprises:

[0013] Infrared, ultraviolet, and visible imaging sensors, each specialized to capture images in a corresponding spectral range;

[0014] Adaptive spectrum selection controller that automatically selects the optimal spectrum setting based on the defect type being detected and ambient light conditions;

[0015] The spectral fusion module is used to integrate different spectral images to enhance the visualization of defect features.

[0016] As a preferred solution of the present invention, the image analysis module further includes an adaptive image preprocessing and optimized edge enhancement filtering algorithm, and the specific steps are as follows:

[0017] Each pixel I(x,y) in the input multispectral image is:

[0018]

[0019] Where x and y are the coordinates of the pixel; max(I) and min(I) are the maximum and minimum values ​​in image I, respectively; I norm (x, y) is the normalized pixel value;

[0020] For each pixel I(x,y), calculate the local variance in the N×N neighborhood, the formula is:

[0021]

[0022] In the formula, σ 2 (x, y) is the calculated local variance; i and j are the offsets in the N×N neighborhood; N is the size of the neighborhood; μ is the average brightness of the N×N neighborhood of pixel (x, y);

[0023] The kernel size of the filter is adjusted according to the local variance, as follows:

[0024]

[0025] Where α is the adjustment strength for controlling the filter kernel size; k is the filter kernel size dynamically adjusted according to the local variance of the image;

[0026] Apply adaptive Gaussian filtering, the formula is:

[0027] σ g =β×local contrast

[0028] In the formula, σ g is the standard deviation of Gaussian filtering; β is the parameter for adjusting the standard deviation of Gaussian filtering; local contrast is the degree of difference in pixel values ​​in a local area;

[0029] The Sobel operator is used for edge detection, where the adaptive threshold T is used to determine the edge. The formula is:

[0030]

[0031] Where Gx and Gy are the gradients of the image in the x and y directions; T is the threshold dynamically calculated based on the global or local characteristics of the image.

[0032] As a preferred solution of the present invention, the evaluation module uses a machine learning model based on image features to evaluate the severity of defects and automatically updates the evaluation criteria by learning historical defect data.

[0033] The vector machine SVM model searches for an optimal segmentation hyperplane in the feature space. The optimization process is as follows:

[0034] Training data set (x i ,y i ), where x i is the eigenvector; y i ∈{+1,-1} is the class label;

[0035] The weight vector w and the bias term b are determined based on the vector machine SVM, and the formula is:

[0036] y i (w*xi+b)≥1;

[0037] Introducing the slack variable ξ i ≥0, the formula is:

[0038]

[0039] In the formula, y i (w·x i +b)≥1-ξi; C is the trade-off between the control error term and the model complexity. The larger the C value, the stricter the model fits the training data. Conversely, the smaller the C value, the looser the model fits the training data. SVM maps the input space to the high-dimensional feature space through the kernel function.

[0040] Select the RBF kernel, and the kernel function expression is:

[0041] K(x i ,x j) = exp(-γ||x i -x j || 2 )

[0042] In the formula, x i and x j is the input vector; γ is the parameter of the kernel function. A larger γ value makes the decision boundary more complex and sharper; a smaller γ value makes the decision boundary smoother.

[0043] As a preferred solution of the present invention, the machine learning model is an advanced deep learning model based on a convolutional neural network (CNN), specifically comprising:

[0044] A network architecture with multiple convolutional layers and pooling layers is used, and the convolution kernel size of each layer ranges from 1×11×1 to 5×55×5.

[0045] The filters in the convolutional layer are designed to capture subtle defects, and the learning rate is dynamically adjusted in combination with a cosine annealing strategy;

[0046] The image is rotated from 0° to 360°, scaled from 0.8x to 1.2x, and data augmentation strategies are integrated.

[0047] As a preferred solution of the present invention, the data output module includes an interactive user interface for displaying defect images in high definition and has the following functions:

[0048] Users can view images of each defect in detail by zooming, rotating and switching between different viewing angles; the severity rating of each defect is displayed in real time and presented through intuitive charts and indicators.

[0049] As a preferred solution of the present invention, the user interface further has the following interactive functions: the user adjusts the detection parameters instantly, and the interface immediately feeds back the effect of the adjustment;

[0050] An integrated user feedback mechanism allows users to submit suggestions for improving the detection algorithm, and the system will automatically adjust the learning model; a guided tutorial is provided to help new users understand how to operate the system and interpret the results.

[0051] A method for detecting defects of a live pin plugging hole, the method comprising:

[0052] Use multispectral imaging equipment to capture images of pin plug holes in infrared, ultraviolet, and visible light spectra; automatically select optimal spectral settings, adjusted based on defect type and ambient light conditions via an adaptive spectral selection controller;

[0053] Apply the spectral fusion module to integrate images of different spectra to enhance the visualization of defect features;

[0054] The image analysis module is used to perform multi-dimensional analysis on the fused image to identify the type and location of the defect.

[0055] As a preferred embodiment of the present invention, the method further comprises:

[0056] Adaptively preprocess and optimize edge enhancement filtering of multispectral images, and adjust filtering parameters to optimize the recognition of defect edges;

[0057] Use the Sobel operator combined with dynamic threshold for edge detection to accurately identify defect edges;

[0058] A deep learning model based on convolutional neural networks is applied to classify the identified defects, determine the severity, and automatically update the assessment criteria through the machine learning model.

[0059] As a preferred embodiment of the present invention, the method further comprises:

[0060] Display detailed defect images and corresponding rating results through an interactive user interface;

[0061] Allow users to adjust detection parameters in real time and immediately provide feedback on the adjustment effect on the interface;

[0062] Collect user feedback and automatically adjust deep learning models to optimize the accuracy of defect detection;

[0063] Educational modules are available to help users understand test results and operate the system.

[0064] Compared with the prior art, the invention has the following beneficial effects: using multispectral imaging technology, it is possible to capture more comprehensive defect images and improve the ability to identify small and complex defects. The improved adaptive spectrum selection controller automatically selects the best spectrum according to the defect type and ambient light conditions, optimizing the image quality and detection efficiency. The adaptive image preprocessing and optimized edge enhancement filtering algorithm automatically adjust the filtering parameters to optimize the edge recognition of defects based on real-time image data. Using advanced convolutional neural network and support vector machine models, the system can accurately classify and rate defects, improving the accuracy and reliability of the evaluation. The interactive user interface provides high-definition defect image display, supports image zooming, rotation and multi-viewing, and enhances the user's operational convenience. The interface allows the user to adjust the detection parameters in real time and immediately feedback the effect, allowing the operator to optimize the detection settings as needed. The integrated user feedback mechanism and automatic learning function enable the system to automatically optimize the detection algorithm based on user feedback and historical data, improving the long-term performance and adaptability of the system. The provided education module helps new users quickly master the use of the system and interpret the detection results, reducing the difficulty of operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor. Among them:

[0066] Figure 1 A system modular structure diagram of an embodiment of the present invention;

[0067] Figure 2 The figure is a flow chart of a method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0068] In order to make the purpose, technical solution and advantages of the embodiment of the present invention clearer, the technical solution of the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings of the embodiment of the present invention. Obviously, the described embodiment is a part of the embodiment of the present invention, not all of the embodiments. Based on the described embodiment of the present invention, all other embodiments obtained by ordinary technicians in this field belong to the scope of protection of the present invention.

[0069] Example 1

[0070] like Figure 1 As shown, an embodiment of the present invention, the image acquisition module includes a multi-spectral imaging device for capturing images of the piston pin plug hole in different spectral ranges;

[0071] The multispectral imaging device comprises:

[0072] Infrared, ultraviolet, and visible imaging sensors, each specialized to capture images in a corresponding spectral range;

[0073] Adaptive spectrum selection controller that automatically selects the optimal spectrum setting based on the defect type being detected and ambient light conditions;

[0074] The spectral fusion module is used to integrate different spectral images to enhance the visualization of defect features.

[0075] In a specific embodiment, the image analysis module is used to perform multi-dimensional analysis on the image to identify the defect type and location;

[0076] The image analysis module further includes an adaptive image preprocessing and optimized edge enhancement filtering algorithm, the specific steps are as follows:

[0077] Each pixel I(x,y) in the input multispectral image is:

[0078]

[0079] Where x and y are the coordinates of the pixel; max(I) and min(I) are the maximum and minimum values ​​in image I, respectively; I norm (x, y) is the normalized pixel value;

[0080] For each pixel I(x,y), calculate the local variance in the N×N neighborhood, the formula is:

[0081]

[0082] In the formula, σ 2 (x, y) is the calculated local variance; i and j are the offsets in the N×N neighborhood; N is the size of the neighborhood; μ is the average brightness of the N×N neighborhood of pixel (x, y);

[0083] The kernel size of the filter is adjusted according to the local variance, as follows:

[0084]

[0085] Where α is the adjustment strength for controlling the filter kernel size; k is the filter kernel size dynamically adjusted according to the local variance of the image;

[0086] Apply adaptive Gaussian filtering, the formula is:

[0087] σ g =β×local contrast

[0088] In the formula, σ g is the standard deviation of Gaussian filtering; β is the parameter for adjusting the standard deviation of Gaussian filtering; local contrast is the degree of difference in pixel values ​​in a local area;

[0089] The Sobel operator is used for edge detection, where the adaptive threshold T is used to determine the edge. The formula is:

[0090]

[0091] Where Gx and Gy are the gradients of the image in the x and y directions; T is the threshold dynamically calculated based on the global or local characteristics of the image.

[0092] In a specific embodiment, an evaluation module is used to rate the severity of the defect based on the image analysis result;

[0093] The evaluation module uses a machine learning model based on image features to assess the severity of defects and automatically updates the evaluation criteria by learning from historical defect data.

[0094] The vector machine SVM model searches for an optimal segmentation hyperplane in the feature space. The optimization process is as follows:

[0095] Training data set (x i ,y i ), where x i is the eigenvector; y i ∈{+1,-1} is the class label;

[0096] The weight vector w and the bias term b are determined based on the vector machine SVM, and the formula is:

[0097] y i (w*xi+b)≥1;

[0098] Introducing the slack variable ξ i ≥0, the formula is:

[0099]

[0100] In the formula, y i (w·xi+b)≥1-ξi; C is the trade-off between the control error term and the model complexity. The larger the C value, the stricter the model fits the training data. Conversely, the smaller the C value, the looser the model fits the training data.

[0101] SVM maps the input space to a high-dimensional feature space through the kernel function;

[0102] Select the RBF kernel, and the kernel function expression is:

[0103] K(x i ,x j ) = exp(-γ||x i -x j || 2 )

[0104] In the formula, x i and x j is the input vector; γ is the parameter of the kernel function. A larger γ value makes the decision boundary more complex and sharper; a smaller γ value makes the decision boundary smoother.

[0105] In a specific embodiment, the machine learning model is an advanced deep learning model based on a convolutional neural network (CNN), specifically including:

[0106] A network architecture with multiple convolutional layers and pooling layers is used, and the convolution kernel size of each layer ranges from 1×11×1 to 5×55×5.

[0107] The filters in the convolutional layer are designed to capture subtle defects, and the learning rate is dynamically adjusted in combination with a cosine annealing strategy;

[0108] The image is rotated from 0° to 360°, scaled from 0.8x to 1.2x, and data augmentation strategies are integrated.

[0109] In a specific embodiment, the data output module is used to display defect detection results and severity ratings; wherein the multispectral imaging device enhances the system's detection capability for tiny defects and improves detection accuracy through an improved adaptive multispectral image fusion algorithm.

[0110] The data output module includes an interactive user interface for displaying defect images in high definition and has the following functions:

[0111] Users can view images of each defect in detail by zooming, rotating, and switching between different viewing angles;

[0112] Displays the severity rating of each defect in real time, presented through intuitive charts and metrics.

[0113] The user interface further has the following interactive features:

[0114] The user can adjust the detection parameters instantly, and the interface will immediately feedback the effect of the adjustment;

[0115] Integrated user feedback mechanism, users submit suggestions for improving the detection algorithm, and the system will automatically adjust the learning model;

[0116] A guided tutorial is provided to help new users understand how to operate the system and interpret the results.

[0117] like Figure 2 FIG. 2 is another embodiment of the present invention, which provides a method for detecting defects of a live pin plug hole, comprising:

[0118] S1: Use a multispectral imaging device to capture images of the pin plug hole in the infrared, ultraviolet, and visible spectra;

[0119] S2: Automatically select the optimal spectrum setting, adjusted according to defect type and ambient light conditions through an adaptive spectrum selection controller;

[0120] S3: Apply the spectral fusion module to integrate images with different spectra to enhance the visualization of defect features;

[0121] S4: Use the image analysis module to perform multi-dimensional analysis on the fused image to identify the type and location of the defect.

[0122] The method further comprises:

[0123] Adaptively preprocess and optimize edge enhancement filtering of multispectral images, and adjust filtering parameters to optimize the recognition of defect edges;

[0124] Use the Sobel operator combined with dynamic threshold for edge detection to accurately identify defect edges;

[0125] A deep learning model based on convolutional neural networks is applied to classify the identified defects, determine the severity, and automatically update the assessment criteria through the machine learning model.

[0126] The method further comprises:

[0127] S5: Display detailed defect images and corresponding rating results through an interactive user interface;

[0128] Allow users to adjust detection parameters in real time and immediately provide feedback on the adjustment effect on the interface;

[0129] Collect user feedback and automatically adjust deep learning models to optimize the accuracy of defect detection;

[0130] S6: Provide educational modules to help users understand test results and operate the system.

[0131] Example 2

[0132] Another embodiment of the present invention provides a data log obtained when testing a pin plug hole defect detection system.

[0133] Table 1. Multispectral imaging equipment performance test table is as follows:

[0134] Spectral type Ambient light conditions Defect detection accuracy Image clarity rating Infrared Low light 88％ 7.5 / 10 UV glare 92％ 8.2 / 10 Visible light Standard light 90％ 8.0 / 10

[0135] In the table:

[0136] Spectral type: refers to the type of imaging technology used, such as infrared, ultraviolet and visible light, each spectrum has different applicability for different material properties and defect types.

[0137] Ambient light conditions: The ambient light conditions when the test is performed, which affect the performance of the imaging device, such as strong or weak light environments.

[0138] Defect Detection Accuracy: Indicates the percentage of defects that the system correctly identifies under specific spectral and lighting conditions.

[0139] Image clarity rating: A rating of image clarity based on subjective or objective criteria, usually from 1 to 10, evaluating the quality and detail displayed in the image.

[0140] It can be seen from Table 1 that a high defect detection accuracy rate is shown under different spectra, proving the ability of the equipment to adapt to different ambient light conditions.

[0141] Table 2. Image analysis module performance test table is as follows:

[0142] Test No. Missed detection rate before filtering Missed detection rate after filtering False detection rate before filtering False detection rate after filtering 1 15％ 5％ 10％ 3％ 2 18％ 6％ 12％ 4％

[0143] In the table:

[0144] Test Number: A sequential number that distinguishes different test settings or conditions.

[0145] Missed detection rate before filtering / Missed detection rate after filtering: The missed detection rate indicates the ratio of defects that the system fails to identify actually existing. The comparison before and after filtering shows the impact of filtering on improving missed detection.

[0146] False detection rate before filtering / false detection rate after filtering: The false detection rate indicates the ratio of defect-free areas incorrectly marked by the system. The comparison before and after filtering is used to demonstrate the effect of filtering on reducing false detections.

[0147] It can be seen from Table 2 that adaptive filtering significantly reduces missed detection and false detection, and improves the accuracy of detection.

[0148] Table 3. The machine learning model evaluation table is as follows:

[0149] Model Type Amount of training data Test accuracy Training time Processing time / image CNN 2000 sheets 95％ 2 hours 150ms SVM 2000 sheets 93％ 3 hours 120ms

[0150] In the table:

[0151] Model type: refers to the machine learning model used for defect classification, such as CNN or SVM.

[0152] Amount of training data: The number of images used to train the model, which affects the generalization ability and accuracy of the model.

[0153] Test accuracy: The percentage of defects that were correctly identified by the model evaluated on an independent test set.

[0154] Training time: The total time required to complete model training.

[0155] Processing time / image: The average time required for the model to process a single image, which affects the real-time performance of the system.

[0156] As can be seen from Table 3, CNN and SVM show efficient training and execution speeds with high accuracy when processing large amounts of data.

[0157] Table 4. The user interface practicality test table is as follows:

[0158] User Type Operation error rate Average study time User satisfaction rating New Users 20％ 30 minutes August 10 Experienced users 5％ 10 minutes 9.5 / 10

[0159] In the table:

[0160] User Type: The experience level of the test user, such as new user or experienced user.

[0161] Operation error rate: The rate at which users make operation errors when using the interface.

[0162] Average learning time: The average time it takes for a user to become familiar with the interface and use it effectively.

[0163] User satisfaction rating: Usually rated from 1 to 10, based on how satisfied users are with the interface's ease of use and functionality.

[0164] As can be seen from Table 4, the differences in operation error rates and learning time between new users and experienced users indicate that the interface has good user guidance and user satisfaction is relatively high.

[0165] Table 5. Defect detection performance of the system under different conditions before and after improvement:

[0166] condition Detection parameters Error rate before improvement Improved error rate Detection time before improvement Improved detection time High ambient light intensity Default Settings 0.15 0.05 200ms 150ms Low ambient light intensity Default Settings 0.2 0.06 220ms 160ms High ambient light intensity Adjusted settings 0.1 0.03 210ms 140ms Low ambient light intensity Adjusted settings 0.18 0.04 230ms 145ms

[0167] In the table:

[0168] Ambient light intensity represents the impact of different ambient light intensities on multispectral imaging.

[0169] The detection parameters indicate whether the system settings have been optimized.

[0170] The error rate represents the rate at which defects are missed (missed detection) or incorrectly marked (false detection) during inspection.

[0171] The test time represents the average time to complete a complete test.

[0172] From the recorded experimental data, we can know that:

[0173] The improved system has significantly reduced error rates under various lighting conditions, demonstrating the effectiveness of adaptive spectrum selection and image processing algorithms. By optimizing algorithm processing, the improved system can complete defect detection faster and improve detection efficiency.

[0174] In summary, compared with traditional single spectrum or manually switched spectrum devices, the present invention automatically adjusts the spectrum settings through an adaptive selection controller, and flexibly adapts according to detection needs and environmental changes, which is a significant innovation in technology. Traditional image processing usually uses fixed parameters or requires manual adjustment of filter settings, while the present invention uses automatic parameter adjustment based on local image characteristics to provide more accurate and responsive image processing capabilities. By integrating deep learning and user feedback, it not only shows high efficiency when it is first deployed, but also continuously optimizes its performance. This is an innovative application that is rarely seen in industrial detection systems.

[0175] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine different embodiments or examples described in this specification and the features of different embodiments or examples, unless they are contradictory.

[0176] Any process or method description in the flow chart or otherwise described herein can be understood to represent a module, fragment or portion of a code including one or more executable instructions for implementing the steps of a specific logical function or process. And the scope of the preferred embodiment of the present application includes other implementations, in which the functions may not be performed in the order shown or discussed, including in a substantially simultaneous manner or in a reverse order according to the functions involved.

[0177] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of various changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A pin plug hole defect detection system, characterized in that: The system comprises: An image acquisition module, including a multi-spectral imaging device, for capturing images of the pin plug hole in different spectral ranges; An image analysis module, used to perform multi-dimensional processing and analysis on the collected images to identify defect types and locations; An evaluation module, used to evaluate the severity of the detected defects based on the data provided by the image analysis module; Data output module for displaying defect detection results and severity ratings; The multispectral imaging device uses an improved adaptive multispectral image fusion algorithm to enhance the system's ability to detect tiny defects and improve detection accuracy; The multispectral imaging device comprises: Infrared, ultraviolet, and visible imaging sensors, each specialized to capture images in a corresponding spectral range; Adaptive spectrum selection controller that automatically selects the optimal spectrum setting based on the defect type being detected and ambient light conditions; Spectral fusion module, used to integrate different spectral images to enhance the visualization of defect features; The image analysis module further includes an adaptive image preprocessing and optimized edge enhancement filtering algorithm, the specific steps are as follows: Every pixel I ( x , y ) in the input multispectral image: ； In the formula, and are the coordinates of the pixel; and Respectively, images The maximum and minimum values ​​in ; is the normalized pixel value; For each pixel ,calculate The local variance within the neighborhood is given by: ； In the formula, is the calculated local variance; and for The offset within the neighborhood; is the size of the neighborhood; It's pixels of The average brightness of the neighborhood; The kernel size of the filter is adjusted according to the local variance, as follows: ； In the formula, To control the adjustment strength of the filter core size; is the filter kernel size that is dynamically adjusted according to the local variance of the image; Apply adaptive Gaussian filtering, the formula is: ； In the formula, is the standard deviation of Gaussian filtering; is the parameter for adjusting the standard deviation of Gaussian filtering; local contrast is the degree of difference in pixel values ​​in a local area; Use Sobel operator for edge detection with adaptive threshold is used to determine the edge, the formula is: ； In the formula, and For images in and Directional gradient; A threshold value that is dynamically calculated based on the global or local characteristics of the image; The evaluation module uses a machine learning model based on image features to evaluate the severity of defects and automatically updates the evaluation criteria by learning from historical defect data. The vector machine SVM model searches for an optimal segmentation hyperplane in the feature space. The optimization process is as follows: Training Dataset ,in, is the feature vector; is the class label; Determine a weight vector based on the vector machine SVM and the deviation term , the formula is: ; Introducing slack variables , the formula is: ； In the formula, ; To control the trade-off between error term and model complexity, The larger the value, the stricter the model fits the training data; conversely, The smaller the value, the looser the model fits the training data. SVM maps the input space to a high-dimensional feature space through the kernel function; Select the RBF kernel, and the kernel function expression is: ； In the formula, and is the input vector; is the parameter of the kernel function; The machine learning model is an advanced deep learning model based on convolutional neural network (CNN), which specifically includes: A network architecture with multiple convolutional layers and pooling layers is used, and the convolution kernel size of each layer ranges from 1×11×1 to 5×55×5. The filters in the convolutional layer are designed to capture subtle defects, and the learning rate is dynamically adjusted in combination with a cosine annealing strategy; Integrated image rotation from 0° to 360°, scaling from 0.8x to 1.2x, and data augmentation strategies; The data output module includes an interactive user interface for displaying defect images in high definition and has the following functions: Users can view images of each defect in detail by zooming, rotating and switching different viewing angles; Display the severity rating of each defect in real time and display it through intuitive charts and indicators; The user interface further has the following interactive functions: The user can adjust the detection parameters instantly, and the interface will immediately feedback the effect of the adjustment; Integrated user feedback mechanism, users submit suggestions for improving the detection algorithm, and the system will automatically adjust the learning model; A guided tutorial is provided to help new users understand how to operate the system and interpret the results.

2. The method for detecting a live pin plug hole defect of a live pin plug hole defect detection system according to claim 1, characterized in that: The method comprises: Using a multispectral imaging device to capture images of the pin plug holes in the infrared, ultraviolet, and visible light spectrums; Automatically select the optimal spectral setting, adjusted according to defect type and ambient light conditions through an adaptive spectral selection controller; Apply the spectral fusion module to integrate images of different spectra to enhance the visualization of defect features; The image analysis module is used to perform multi-dimensional analysis on the fused image to identify the type and location of the defect.

3. The method for detecting defects of a live pin plug hole according to claim 2, characterized in that: The method further comprises: Adaptively preprocess and optimize edge enhancement filtering of multispectral images, and adjust filtering parameters to optimize the recognition of defect edges; Use the Sobel operator combined with dynamic threshold for edge detection to accurately identify defect edges; A deep learning model based on convolutional neural networks is applied to classify the identified defects, determine the severity, and automatically update the assessment criteria through the machine learning model.

4. The method for detecting defects of a live pin plug hole according to claim 2, characterized in that: The method further comprises: Display detailed defect images and corresponding rating results through an interactive user interface; Allow users to adjust detection parameters in real time and immediately provide feedback on the adjustment effect on the interface; Collect user feedback and automatically adjust deep learning models to optimize the accuracy of defect detection; Educational modules are available to help users understand test results and operate the system.

Images