Surface defect detection method
The surface defect detection method uses a neural network model to map detection scores to RGB values, generating a heatmap that enhances the accuracy and clarity of anomaly identification in surface defect detection.
Patent Information
- Authority / Receiving Office
- JP · JP
- Patent Type
- Patents
- Current Assignee / Owner
- GUANGDONG HONHOR SEMICON EQUIP CO LTD
- Filing Date
- 2025-03-04
- Publication Date
- 2026-05-27
Smart Images
Figure 0007866657000008 
Figure 0007866657000009 
Figure 0007866657000001
Abstract
Description
[Technical Field]
[0001] The present invention relates to the field of image detection, and more specifically, to a method for detecting surface defects. [Background technology]
[0002] A heatmap is a visualization tool that displays data density using color. It displays data in a two-dimensional space as hot points of different colors, and is used to show the distribution and concentration of data. This helps in observing data trends and patterns. Heatmaps are generated by mapping numerical data to colors. Generally, larger data values are represented by darker colors, and smaller data values by lighter colors. This can be applied to detecting anomaly and failure types in various products.
[0003] Currently, a common method for detecting wafer anomalies is to present the results using detection boxes. While this is a good approach, too many detection boxes make it difficult to interpret the results. When multiple detection boxes appear on a single diagram, and these boxes overlap, the resulting diagram becomes quite complex and hinders the interpretation of the results. [Overview of the Initiative] [Problems that the invention aims to solve]
[0004] In view of this, the present invention proposes a surface defect detection method aimed at solving the problem of the low accuracy of conventional surface defect detection methods. [Means for solving the problem]
[0005] The present invention relates to a surface defect detection method, The process involves acquiring an image of the object being tested, inputting the image of the detected object into a pre-trained neural network model, and obtaining multiple detection boxes on which detection scores are indicated. The detection scores described in the aforementioned plurality of detection boxes are mapped to a two-dimensional array, Fitting the detection score and substituting the fitted detection score into the RGB polynomial regression equation, Using the RGB polynomial regression equation with the detection score substituted therein to map each detection score described in the two-dimensional array to three-dimensional RGB numerical values, Providing a surface defect detection method including generating a heat map based on the three-dimensional RGB numerical values, superimposing the heat map on the detected object image, and presenting abnormal points displayed in the heat map.
[0006] Furthermore, in the above surface defect detection method, the training method of the neural network model is Taking pictures of images of detected objects with abnormal points, each of which is a multi-channel RGB image, Manually marking the abnormal points in each of the detected object images and surrounding the abnormal points in the images using detection boxes each containing the abnormal points of the detected object, Screening images containing abnormal points to construct a dataset and dividing the dataset into a training set and a test set, Constructing a neural network model based on the EfficientNet-SSD algorithm, inputting and training the image data in the training set, generating a neural network model based on the EfficientNet-SSD algorithm, performing feature extraction on the images using the self-learning ability of the EfficientNet-SSD model to obtain a plurality of feature images with different resolutions, generating a plurality of detection boxes with different resolutions on these feature images based on the plurality of feature images with different resolutions, outputting the probability that each pixel point in the plurality of detection boxes with different resolutions is an abnormal point, and evaluating the neural network model based on the EfficientNet-SSD algorithm using the image data of the test set to obtain an optimized neural network model.
[0007] Furthermore, in the above surface defect detection method, based on the generated neural network model of the EfficientNet-SSD algorithm, position adjustment is performed on the detection boxes with different combinations of resolutions and positions generated, so that these detection boxes are approximated to the true boxes marked with abnormal points, optimizing the neural network model. Thereby, a neural network model based on the optimized EfficientNet-SSD algorithm is obtained, and by evaluating the neural network model based on the optimized EfficientNet-SSD algorithm using the image data of the test set, an optimized neural network model can be obtained. This further includes the fact that it is possible to obtain the optimized neural network model.
[0008] Furthermore, in the above surface defect detection method, mapping the detection scores described in the plurality of detection boxes to a two-dimensional array respectively Establishing a two-dimensional matrix of the same size according to the size of the original detection image According to the coordinates of each pixel point in the detection box, associating and substituting the detection score of each pixel point in the detection box into the corresponding element of the two-dimensional matrix Repeating the above steps, accumulating the detection scores of each pixel point in each detection box into the corresponding elements of the two-dimensional matrix. In this way, each element value in the two-dimensional matrix is the cumulative sum of the detection scores of the corresponding pixel points in each detection box.
[0009] Furthermore, in the above surface defect detection method, mapping each detection score described in the two-dimensional array to three-dimensional RGB values Generating one RGB matrix according to the size of the original detection image, and calculating the detection scores of each element in the two-dimensional matrix by R, G, B polynomial regression equations respectively to obtain the values of each RGB channel in the RGB matrix.
[0010] Furthermore, in the above surface defect detection method, substituting the fitted detection score into the RGB polynomial regression equation Customize the relationship between color and detection score, and Utilize the relationship between the detection score and the RGB values of the color to obtain the polynomial regression equation f(x) = a , 3 , , , 2 , 4 , 6 , 1 , 2 , , 1 , 4 ,
[0012] , 3 , , 5 , 7 , x n + a n-1 x (n-1) + … + a2x 2 + a1x + a o Identify each constant in the equation and further obtain the solution of the following polynomial regression equation.
[0011] R = 1.06105X 7 - 3.12570X 6 + 3.40036X 5 - 1.68407X 4 + 3.86728X 3 - 3.68337X 2 + 1.01210X 1 + 1.38694, G = 2.12211X 7 - 7.73688X 6 + 1.12571X 5 - 8.26694X 4 + 3.14362X 3 - 5.52839X 2 + 3.37545X 1 - 1.98135, B = - 2.12211X 7 + 8.29394X 6 - 1.29283X 5 + 1.01249X 4 - 4.07428X 3 + 7.41653X 2 - 3.57669X 1 + 7.92540.
[0012] Furthermore, in the above surface defect detection method, the relationship between the color and the detection score is as follows: The RGB values corresponding to a detection score of 1 are 255, 0, 0 respectively. The RGB values corresponding to a detection score of 0.875 are 255, 127.5, 0 respectively. The RGB values corresponding to a detection score of 0.75 are 255, 255, 0 respectively. When the detection score is 0.625, the corresponding RGB values are 127.5, 255, and 0, respectively. When the detection score is 0.5, the corresponding RGB values are 0, 255, and 0, respectively. When the detection score is 0.375, the corresponding RGB values are 0, 127.5, and 127.5, respectively. When the detection score is 0.25, the corresponding RGB values are 0, 0, and 255, respectively. When the detection score is 0.125, the corresponding RGB values are 0, 0, and 127.5, respectively. When the detection score is 0, the corresponding RGB values are 0, 0, and 0, respectively.
[0013] Furthermore, in the above surface defect detection method, in the RGB polynomial regression equation into which the fitted detection score is substituted, a specific coefficient R 2 The score is an indicator used to evaluate the fit of a model. JPEG0007866657000001.jpg13170
[0014] Furthermore, the above surface defect detection method further includes normalizing each detection score listed in the two-dimensional array and then mapping it to (0~1), and the normalization formula is as follows: JPEG0007866657000002.jpg24170
[0015] Furthermore, in the above surface defect detection method, after preprocessing the captured image containing the object to be detected, it is input into a pre-trained neural network model based on the EfficientNet-SSD algorithm, and each pixel point in the detection box, with multiple resolution and position combinations, generates an anomaly detection score. [Effects of the Invention]
[0016] In the surface defect detection method of the present invention, an image containing the object to be detected is preprocessed and then input into a pre-trained neural network model. Each pixel point in the detection box, which is a combination of multiple resolutions and positions, generates an anomaly detection score. Furthermore, the detection scores of the corresponding pixel points are mapped to a two-dimensional array. Using a polynomial regression equation that converts the two-dimensional detection scores into three-dimensional RGB values, the two-dimensional array with the detection scores is mapped to three-dimensional RGB values. This generates a heatmap, and by superimposing the heatmap onto the original image of the object to be detected, anomalies in the image of the object to be detected can be determined intuitively, quickly, and accurately. [Brief explanation of the drawing]
[0017] A variety of other advantages and benefits will become apparent to those skilled in the art by reading the detailed description of the preferred embodiments below. The drawings are used solely for the purpose of illustrating preferred embodiments and are not intended to limit the invention. Furthermore, the same parts are indicated by the same reference numerals throughout the drawings. [Figure 1] This is a flowchart of a surface defect detection method according to an embodiment of the present invention. [Figure 2] This diagram shows the relationship between detection score and color as defined in the surface defect detection method according to an embodiment of the present invention. [Modes for carrying out the invention]
[0018] Hereinafter, exemplary embodiments of the present disclosure will be described in more detail with reference to the accompanying drawings. While the drawings show exemplary embodiments of the present disclosure, it should be understood that the present disclosure should not be limited to the embodiments described herein and can be realized in various forms. Rather, these embodiments are provided to allow for a more complete understanding of the present disclosure and to fully convey its scope to those skilled in the art. Where there is no inconsistency, embodiments and features of the present invention may be combined with each other. Hereinafter, the present invention will be described in detail with reference to the drawings and in relation to the embodiments.
[0019] Referring to Figure 1, the surface defect detection method of the embodiment of the present invention includes the following steps.
[0020] Step 1: An image of the object to be detected is acquired, and the image of the detected object is input into a pre-trained neural network model to obtain multiple detection boxes on which detection scores are written.
[0021] Specifically, after preprocessing the captured images containing the detected objects, they are input into a pre-trained neural network model based on the EfficientNet-SSD algorithm. This model generates an anomaly detection score for each pixel point in the detection box, based on multiple resolution and location combinations.
[0022] In this embodiment, the object to be detected may be a semiconductor element, more specifically, a wafer. The tool used for imaging may be a high-resolution camera, such as a 25-megapixel camera. Preprocessing may involve cropping the image to a certain resolution, for example, cropping it to a 448*448 (or other resolution such as 800*800) pixel multi-channel RGB image. The detection score is the probability that a pixel consisting of any one pixel point and other pixel points in the image is an anomaly. Here, the combination of resolution and position means that each detection box has a different size (resolution) and a different position in the image, and since each detection box is not shifted with spacing between them, there may be overlapping areas in each detection box.
[0023] In this embodiment, each pixel point in the detection box corresponds to one detection score.
[0024] Step 2: The detection scores listed in the plurality of detection boxes are mapped to a two-dimensional array.
[0025] Specifically, in step 1, a two-dimensional matrix of the same size is established according to the size of the original detected image (let L be the resolution in the length direction, W be the resolution in the width direction, and L*W be the size of the image), and the two-dimensional matrix has L rows and W columns.
[0026] Step 3: Substitute the detection score for each pixel point within the detection box into the corresponding element of the two-dimensional matrix, according to the coordinates of each pixel point within the detection box.
[0027] The above steps are repeated, and the detection score of each pixel point within each detection box is accumulated in the corresponding element of the two-dimensional matrix.
[0028] Thus, each element value in the two-dimensional matrix is the cumulative sum of the detection scores of the corresponding pixel points within each detection box.
[0029] To explain, if the detection score value of each pixel point within the detection box is between 0 and 1, then the cumulative sum value will be between 0 and N, where N is a known quantity, which is the cumulative sum of detection scores within the detection box for the same pixel. In this way, the detection score data can be further normalized and remapped to be between 0 and 1.
[0030] Step S4: The detection score is fitted, and the fitted detection score is substituted into the RGB polynomial regression equation.
[0031] In the specific implementation, an RGB matrix is generated according to the size of the original detection image, and the detection score of each element in the two-dimensional matrix in step 2 above is calculated using R, G, and B polynomial regression equations to obtain the numerical values of each RGB channel in the RGB matrix.
[0032] Specifically, the polynomial regression equation is f(x)=a n x n +a n-1 x (n-1) +…+a2x 2 +a1x+a o That is the case.
[0033] In this example, the process for obtaining the polynomial is as follows: The relationship between custom colors and detection scores is shown in Figure 2.
[0034] When the detection score is 1, the corresponding RGB values are 255, 0, and 0, respectively. When the detection score is 0.875, the corresponding RGB values are 255, 127.5, and 0, respectively. When the detection score is 0.75, the corresponding RGB values are 255, 255, and 0, respectively. When the detection score is 0.625, the corresponding RGB values are 127.5, 255, and 0, respectively. When the detection score is 0.5, the corresponding RGB values are 0, 255, and 0, respectively. When the detection score is 0.375, the corresponding RGB values are 0, 127.5, and 127.5, respectively. When the detection score is 0.25, the corresponding RGB values are 0, 0, and 255, respectively. When the detection score is 0.125, the corresponding RGB values are 0, 0, and 127.5, respectively. When the detection score is 0, the corresponding RGB values are 0, 0, and 0, respectively.
[0035] Table 1 below shows the results of 20 runs using the data shown in Figure 2.
[0036] JPEG0007866657000003.jpg117170
[0037] As can be seen from the table above, when N=7, R² is already equal to 1, and it is also equal to 1 for N=8 to 20. However, as the polynomial regression equation becomes more complex, the computation time of the system increases, so when N=7, it can be considered that it already belongs to the optimal solution. Therefore, when N=7, the solution to the polynomial regression is obtained by the following equation. R=1.06105X 7 -3.12570X 6 +3.40036X 5 -1.68407X 4 +3.86728X 3 -3.68337X 2 +1.01210X 1 +1.38694, G=2.12211X 7 -7.73688X 6 +1.12571X 5 -8.26694X 4 +3.14362X 3 -5.52839X 2 +3.37545X 1 -1.98135, B = -2.12211X 7 +8.29394X 6 -1.29283X 5 +1.01249X 4 -4.07428X 3 +7.41653X 2 -3.57669X 1 +7.92540.
[0038] The R2 score is used to determine how well a model fits the data. The specific coefficient (R2 score): R2 is an index that evaluates the fit of a model, indicating the degree to which the model interprets changes in the data. The closer the R2 value is to 1, the better the model fits the data.
[0039] When generating an equation specifically for mapping detection scores to R, G, and B using a polynomial regression equation, the R-Squared calculation formula is as follows:
[0040] JPEG0007866657000004.jpg14170
[0041] By using the RGB polynomial regression equation into which the fitted detection scores from the embodiment of the present invention have been substituted, the RGB values corresponding to any pixel point can be obtained. For example, if the detection score of one pixel point in one detection box is 0.32 (32%), the above R, G, B polynomial regression equation converts 0.32 into three-dimensional data with R=0, G=70, and B=195, and presents or displays the corresponding color after rendering.
[0042] If the detection score for a single pixel point in a detection box is 0.79 (79%), the above polynomial regression equation converts 0.79 into three-dimensional data with R=255, G=227, and B=0, and presents or displays the corresponding color after rendering.
[0043] Step S5: A heat map is generated based on the three-dimensional RGB values, and the heat map is superimposed on the image of the detected object to display the anomalies shown in the heat map.
[0044] Therefore, a polynomial regression equation can be used to convert data with detection scores ranging from 0 to 1 (0 to 100%) into corresponding R, G, and B parameters. In this way, different detection results can be displayed in different colors (for example, red indicates a high probability that this point is an anomaly), and presented as a heatmap.
[0045] As can be seen from the above, in this invention, by defining the relationship between the anomaly detection score and RGB, constant parameters of the polynomial regression equation are identified, an equation that maps the detection score to R, G, and B is generated by the polynomial regression equation, the detection scores within the detection box are accumulated, a two-dimensional matrix of the accumulated detection scores is generated, the detection score data is further normalized and remapped between 0 and 1, then the values of R, G, and B are recalculated using the polynomial regression equation, and finally a heatmap advantageous for identifying anomalies is generated.
[0046] In other words, based on the size of each detection box, the detection score of the corresponding pixel point is mapped to a two-dimensional array, and the two-dimensional array with the detection score is remapped between 0 and 1 using a polynomial regression equation that converts the two-dimensional detection score to three-dimensional RGB, generating a three-dimensional array of RGB values, converting the three-dimensional array of RGB values into a heatmap, and superimposing and presenting the heatmap result onto the original image of the detected object.
[0047] In this way, by observing the heat map, a person can determine the abnormal points on the object being detected based on this heat map.
[0048] As can be seen from the above embodiment, in the present invention, after preprocessing an image containing the object to be detected, it is input into a pre-trained neural network model to generate an anomaly detection score for each pixel point in the detection box with multiple resolution and position combinations. Furthermore, the detection scores of the corresponding pixel points are mapped to a two-dimensional array, and the two-dimensional detection scores are converted to three-dimensional RGB values using a polynomial regression equation. This maps the two-dimensional array with detection scores to three-dimensional RGB values, thereby generating a heatmap. By superimposing the heatmap onto the original image of the object to be detected, it is possible to intuitively, quickly, and accurately determine anomalies in the image of the object to be detected, and is also advantageous for identifying areas with defects on the surface of the object to be detected.
[0049] In the above embodiment, the selected neural network model is EfficientNet-SSD, and the training process of this neural network model based on the EfficientNet-SSD algorithm is as follows.
[0050] Step 1: Capture images of the detected object containing anomalies, each of which is a multi-channel RGB image.
[0051] Specifically, the object to be detected may be a wafer, and a 25-megapixel camera may be used to capture images of multiple wafers having the same anomaly or multiple wafers having different anomalies. Subsequently, each image may be cut into a multi-channel RGB image with a resolution of 448*448 (or other resolutions such as 800*800).
[0052] In the specific implementation, wafers with abnormalities are placed on a detection stand, a 25-megapixel camera is used to photograph the wafers, and the images are stored in a designated folder.
[0053] Step 2: Manually mark the abnormal points in each of the detected object images, and surround the abnormal points in the image using detection boxes, each containing the abnormal points of the detected object.
[0054] Specifically, the detection box can be selected from options such as a rectangular box or an elliptical box.
[0055] Step 3: Screen images containing anomalies to construct a dataset, and divide the dataset into a training set and a test set. For example, 80% of the dataset may be selected as the training set, and the remaining 20% of the dataset may be used as the test set.
[0056] Step 4: Construct a neural network model based on the EfficientNet-SSD algorithm and train it by inputting image data from the training set. The neural network model based on the EfficientNet-SSD algorithm is used to predict detection scores for detecting detection boxes corresponding to each of multiple resolution and location combinations in the input image. The resolution and location of the detection boxes in the image are adapted and identified during the training process of the neural network model based on the EfficientNet-SSD algorithm based on the sample set, training set, and training accuracy. The detection scores are used to indicate the probability that each corresponding pixel point is an anomaly after detecting detection boxes corresponding to each resolution and location combination in the image.
[0057] To make the neural network model more complete, this embodiment may further include step 5, in which the neural network model based on the generated EfficientNet-SSD algorithm is repositioned to bring the detection boxes of different resolution and position combinations closer to the true boxes marked with anomalies, thereby optimizing the neural network model, and an optimized neural network model based on the EfficientNet-SSD algorithm is obtained, and the optimized neural network model is obtained by evaluating the optimized neural network model based on the EfficientNet-SSD algorithm using image data from a test set.
[0058] In this example, the neural network model is optimized using the Momentum optimization algorithm, which is a reference momentum optimization algorithm for optimizers.
[0059] The specific implementation process of the embodiment of the present invention is as follows: The object to be detected is placed on a detection platform, the shooting path of the detection camera (for example, a 25-megapixel camera may be selected) is set, an image of the object is acquired, and the image is input into a pre-trained neural network model. The neural network model outputs multiple detection boxes with detection scores, generates a two-dimensional array based on the size of the captured image, maps the detection scores to the two-dimensional array according to the size of the detection boxes, generates a three-dimensional array according to the size of the captured image, obtains a polynomial regression equation that converts the detection scores into three-dimensional RGB, normalizes the two-dimensional array with detection scores, stores the result back into the two-dimensional array, inputs the normalized two-dimensional array of detection scores into a three-dimensional RGB polynomial regression equation, obtains three-dimensional RGB values, stores the result back into the previously generated three-dimensional array, generates a heatmap based on the RGB values of this three-dimensional array, and superimposes the heatmap onto the original detection image.
[0060] The normalization formula for a two-dimensional array is as follows: JPEG0007866657000005.jpg25170
[0061] In summary, in the embodiment of the present invention, an image of the object to be detected is input to a pre-trained neural network model, and a plurality of detection boxes on which detection scores are recorded are obtained and defined. The detection score of each pixel point in each detection box is accumulated in the corresponding element of the two-dimensional matrix, so that each element value in the two-dimensional matrix is the cumulative sum of the detection scores of the corresponding pixel points in each detection box. The detection scores are then fitted, and the fitted detection scores are substituted into an RGB polynomial regression equation. Using the RGB polynomial regression equation with the detection scores substituted, each detection score recorded in the two-dimensional array is mapped to a three-dimensional RGB value. Finally, a heat map that is advantageous for identifying abnormal locations is generated, making the detection results more intuitive and accurate. Clearly, those skilled in the art can make various modifications and variations to the present invention without departing from the spirit and scope of the invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and the equivalent art, the present invention is intended to include these modifications and variations.
Claims
1. A method for detecting surface defects, The process involves acquiring an image of the object to be detected, inputting the image of the object to a pre-trained neural network model, and obtaining multiple detection boxes on which detection scores are indicated. The detection scores described in the aforementioned plurality of detection boxes are mapped to a two-dimensional array, The detection score is fitted, and the fitted detection score is substituted into the RGB polynomial regression equation. Using the RGB polynomial regression equation into which the aforementioned detection scores are substituted, each detection score listed in the two-dimensional array is mapped to a three-dimensional RGB value. This includes generating a heatmap based on the three-dimensional RGB values, superimposing the heatmap onto the image of the detected object, and displaying the anomalies displayed within the heatmap. The training method for the aforementioned neural network model is: Each image is a multi-channel RGB image of the detected object containing anomalies, and The abnormal points in each image of the detected object are manually marked, and detection boxes containing the abnormal points of the detected object are used to surround the abnormal points in the image. This involves screening images containing anomalies to construct a dataset, and then dividing the dataset into a training set and a test set. A surface defect detection method characterized by comprising: constructing a neural network model based on the EfficientNet-SSD algorithm and training it by inputting image data from a training set; generating a neural network model based on the EfficientNet-SSD algorithm; using the autonomous learning ability of the EfficientNet-SSD model to extract features from images; obtaining multiple feature images of different resolutions; generating multiple detection boxes of different resolutions on these feature images based on the multiple feature images of different resolutions; outputting the probability that each pixel point in the multiple detection boxes of different resolutions is an anomaly; and evaluating the neural network model based on the EfficientNet-SSD algorithm using image data from a test set to obtain an optimized neural network model.
2. The surface defect detection method according to claim 1, further comprising: adjusting the position of the detection boxes with different resolution and position combinations generated by the generated EfficientNet-SSD algorithm to bring these detection boxes closer to the true boxes marked with anomalies, thereby optimizing the neural network model; obtaining an optimized neural network model based on the EfficientNet-SSD algorithm; and evaluating the optimized neural network model based on the EfficientNet-SSD algorithm using image data from a test set to obtain the optimized neural network model.
3. Mapping the detection scores described in the plurality of detection boxes into a two-dimensional array is, Establish a two-dimensional matrix of the same size according to the size of the original detected image, The detection score for each pixel point within the detection box is assigned to the corresponding element of the two-dimensional matrix according to the coordinates of each pixel point within the detection box, and The surface defect detection method according to claim 1, characterized in that the above steps are repeated and the detection score of each pixel point in each detection box is accumulated into the corresponding element of the two-dimensional matrix, wherein each element value in the two-dimensional matrix is the cumulative sum of the detection scores of the corresponding pixel points in each detection box.
4. Mapping each detection score listed in the aforementioned two-dimensional array to a three-dimensional RGB value is: The surface defect detection method according to claim 2, characterized in that it includes generating an RGB matrix according to the size of the original detection image, and calculating the detection score of each element in the two-dimensional matrix using R, G, and B polynomial regression equations to obtain numerical values for each RGB channel in the RGB matrix.
5. A method for detecting surface defects, The process involves acquiring an image of the object to be detected, inputting the image of the object to a pre-trained neural network model, and obtaining multiple detection boxes on which detection scores are indicated. The detection scores described in the aforementioned plurality of detection boxes are mapped to a two-dimensional array, The detection score is fitted, and the fitted detection score is substituted into the RGB polynomial regression equation. Using the RGB polynomial regression equation into which the aforementioned detection scores are substituted, each detection score listed in the two-dimensional array is mapped to a three-dimensional RGB value. This includes generating a heatmap based on the three-dimensional RGB values, superimposing the heatmap onto the image of the detected object, and displaying the anomalies displayed within the heatmap. Substituting the fitted detection score into the RGB polynomial regression equation as described above means that Customizing the relationship between color and detection score, Using the relationship between the detection score and the RGB values of the color, a polynomial regression equation f(x) = a ｎ x ｎ +a ｎ－１ x （ｎ－１） +...+a ２ x ２ +a １ x + a ｏ A surface defect detection method characterized by including identifying each constant in and obtaining the solution to the following polynomial regression equation. R=1.06105X ７ -3.12570X ６ +3.40036X ５ -1.68407X ４ +3.86728X ３ -3.68337X ２ +1.01210X １ +1.38694、 G=2.12211X ７ -7.73688X ６ +1.12571X ５ -8.26694X ４ +3.14362X ３ -5.52839X ２ +3.37545X １ -1.98135、 B=-2.12211X ７ +8.29394X ６ -1.29283X ５ +1.01249X ４ -4.07428X ３ +7.41653X ２ -3.57669X １ +7.92540
6. The relationship between the aforementioned color and the detection score is as follows: When the detection score is 1, the corresponding RGB values are 255, 0, and 0, respectively. When the detection score is 0.875, the corresponding RGB values are 255, 127.5, and 0, respectively. When the detection score is 0.75, the corresponding RGB values are 255, 255, and 0, respectively. When the detection score is 0.625, the corresponding RGB values are 127.5, 255, and 0, respectively. When the detection score is 0.5, the corresponding RGB values are 0, 255, and 0, respectively. When the detection score is 0.375, the corresponding RGB values are 0, 127.5, and 127.5, respectively. When the detection score is 0.25, the corresponding RGB values are 0, 0, and 255, respectively. When the detection score is 0.125, the corresponding RGB values are 0, 0, and 127.5, respectively. The surface defect detection method according to claim 5, characterized in that the RGB values corresponding to the detection score of 0 are 0, 0, and 0, respectively.
7. When fitting the detection score to identify the constants of the RGB polynomial regression equation, the specific coefficient R is used as an index to evaluate the fit of the model. ２ Using Score, the specific coefficient R ２ The constants are specified such that Score approaches 1, and the specified coefficient R ２ Score is,
8. The process further includes normalizing each detection score listed in the two-dimensional array and then mapping it to (0-1), the normalization formula being as follows:
9. The surface defect detection method according to claim 1, characterized in that, after preprocessing the captured image containing the object to be detected, the image is input into a pre-trained neural network model based on the EfficientNet-SSD algorithm, and each pixel point in the detection box with multiple resolution and position combinations generates an anomaly detection score.