Sample data labeling method for image detection model

Abstract

Disclosed in the present invention is a sample data labeling method for an image detection model. The method comprises the following steps: S1, determining an anomaly-free image as a basic image; S2, acquiring images to be subjected to detection, and comparing, with the basic image, the images to be subjected to detection, wherein the images to be subjected to detection and the basic image are subjected to inverse-color overlay or inverse-color deduplication to screen out an anomalous image from among the images to be subjected to detection; S3, by means of contour detection, independently extracting and splitting anomalous positions in the anomalous image to form a plurality of independent graphs; S4, determining a minimum bounding rectangle on the basis of an edge pixel position of each independent graph; and S5, digitizing the determined minimum bounding rectangle, and using same as sample data for training an image detection model.

Description

A Sample Data Labeling Method for an Image Detection Model Technical Field

[0001] This invention relates to the technical field of visual recognition data labeling, and in particular to a sample data labeling method for an image detection model. Background Technology

[0002] In the semiconductor manufacturing industry, the showerhead is used in etching equipment for semiconductor integrated circuit manufacturing processes. During the wafer etching process, it serves as an electrode for applying voltage and as a pathway for etching gas to enter the cavity, making it an essential core component. The key factor affecting the production quality of the etching process is the inspection of patterned wafers. Specifically, this involves identifying and analyzing defects or errors that may occur on the wafer surface during pattern lithography, such as defects, misalignments, and missing features. Therefore, those skilled in the art generally utilize visual inspection technology for wafer quality control to assist in identifying defects in the manufacturing process and improve yield.

[0003] Traditional visual inspection relies on recognition software that is limited by the dataset, requiring long-term manual labeling of anomalous images. However, these anomalous images have two main problems: 1) It is difficult to determine the anomalous range because the boundaries between good and bad areas have overlapping colors, which may lead to inaccurate labeling when manually labeling anomalous data; 2) Labeling is tedious and time-consuming because each data point in the anomalous area needs to be labeled one by one, which is tedious and time-consuming when processing large amounts of data. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art and provide a sample data labeling method for an image detection model that is efficient, accurate and stable.

[0005] To achieve the above objectives, the present invention provides a sample data labeling method for an image detection model, comprising the following steps:

[0006] S1. Establish an image without anomalies as the base image;

[0007] S2. Obtain the image to be tested and compare it with the base image. The image to be tested and the base image are compared by inverting colors to filter out abnormal images in the image to be tested.

[0008] S3. Contour detection is used to independently extract and split the abnormal locations in the abnormal image into multiple independent graphics;

[0009] S4. Determine the minimum bounding rectangle based on the edge pixel position of each individual graphic;

[0010] S5. The established minimum bounding rectangle is digitized and used as sample data for training the image detection model.

[0011] Furthermore, the inversion deduplication includes the following steps: First, the base image is inverted; then, based on the RGB matrix difference between the inverted base image and the image to be tested, the region corresponding to the RGB matrix difference is used as the abnormal image in the image to be tested.

[0012] Furthermore, the inverted color overlay includes the following steps: First, the contour range image of the image to be tested is obtained through contour detection. Then, the area enclosed by the contour range is filled with black. Next, the filled contour range image is overlaid on the base image to form an overlay image. Finally, the overlay image is inverted and used as an abnormal image in the image to be tested.

[0013] Furthermore, abnormal locations in the abnormal image can be manually marked, and the manually marked pixel locations can be digitized and used as sample data for training the image detection model.

[0014] Further, in step S4, calculating the minimum bounding rectangle includes the following steps: First, based on the point set given by the independent graphic obtained from contour detection; then, find the minimum and maximum values ​​in the X-axis and Y-axis directions respectively from the point set; next, calculate and confirm the coordinates of the upper left endpoint and the rectangle size of the independent graphic based on the minimum and maximum values ​​respectively, wherein the coordinates of the upper left endpoint are the minimum value of the X-axis and the maximum value of the Y-axis, the width of the rectangle is the difference between the maximum and minimum values ​​of the X-axis and the width of the rectangle is the difference between the maximum and minimum values ​​of the Y-axis; finally, establish the minimum bounding rectangle based on the coordinates of the upper left endpoint and the rectangle size.

[0015] The present invention employs the above-mentioned scheme, and its beneficial effects are as follows: 1. Improved detection efficiency: By establishing a base image and performing image comparison, abnormal parts in the image under test can be quickly identified, thereby improving the overall detection efficiency. 2. Accurate anomaly identification: Utilizing inverse color overlay or inverse color deduplication techniques, abnormal regions in the image can be identified more accurately, reducing false positives and false negatives. 3. Automated processing: Through contour detection and calculation of the minimum bounding box, the automatic extraction and labeling of abnormal regions can be achieved, reducing manual intervention and improving the level of automation. 4. Data-driven samples: Datafication of the minimum bounding box of abnormal images provides standardized training samples for the image detection model, which helps improve the model's accuracy and generalization ability. 5. Easy implementation: The entire method has clear steps, is easy to program and implement, and is easy to integrate into existing image processing and detection systems. 6. Improved model performance: By providing high-quality training samples, the performance of the image detection model is improved, making it more reliable and effective in practical applications. Attached Figure Description

[0016] Figure 1 is a flowchart illustrating the sample data labeling method.

[0017] Figure 2 is a schematic diagram of the process of color inversion and deduplication.

[0018] Figure 3 is a schematic diagram of the process of color inversion overlay.

[0019] Figure 4 is a schematic diagram of the contour detection process. Detailed Implementation

[0020] To facilitate understanding of the present invention, a more complete description is given below with reference to the accompanying drawings. Preferred embodiments of the invention are shown in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. These embodiments are provided to provide a thorough and complete understanding of the disclosure of the invention.

[0021] Referring to Figures 1-4, in this embodiment, a sample data labeling method for an image detection model includes the following steps:

[0022] Step S1. Establish an image without anomalies as the base image. For ease of explanation, take the black and white image in Figure 2 as an example. The image without anomalies is fixed, with a complete white circle in the middle and a hollow area with black edges.

[0023] Step S2. Obtain the image to be tested and compare it with the base image. The image to be tested and the base image are screened out by inverting the colors and overlaying or deduplicating the colors.

[0024] In this embodiment, abnormal images can be screened out from the images to be tested using two methods: inverted color overlay or inverted color deduplication, as detailed below:

[0025] Referring to Figure 3, the color inversion deduplication includes the following steps: First, the base image is inverted (i.e., the original white areas are turned into black, and the original black areas are turned into white). Then, based on the RGB matrix difference between the inverted base image and the image to be tested (i.e., the RGB matrix of the inverted base image is: A=[RA, GA, BA], the RGB matrix between the images to be tested is: [RB, GB, BB], and the RGB matrix difference C=[|RA−RB|,|GA−GB|,|BA−BB|]), finally, the region corresponding to the RGB matrix difference is used as the abnormal image in the image to be tested. Specifically, in the inverted base image and the image under test, the white areas are defined as R=255, G=255, B=255, and the black areas as R=0, G=0, B=0. The black areas with an RGB matrix difference of R=0, G=0, B=0 are identified as outliers. For the remaining areas, the calculated RGB matrix difference is either R=255, G=255, B=255 or R=-255, G=-255, B=-255. Taking the absolute value converts the negative RGB values ​​to positive RGB values, thus converting all other areas to white. Based on this, outliers in the image under test can be quickly identified.

[0026] Referring to Figure 4, the color inversion overlay includes the following steps: First, the contour range image of the image to be tested is obtained through contour detection (i.e., the contour image of the white area is obtained through contour detection). Then, the area enclosed by the contour range is filled with black. Next, the filled contour range image is overlaid on the base image to form an overlay image (i.e., the remaining white area after the black contour range image is overlaid with the base image is the abnormal position). Finally, the overlay image is inverted and used as the abnormal image in the image to be tested (by inverting the overlay image to convert the remaining white area to black, the inverted black area is used as the abnormal image in the image to be tested).

[0027] In summary, by using the two methods of color inversion overlay or color inversion deduplication, abnormal images in the test image can be screened out quickly and accurately, which is more convenient for subsequent processing. This effectively solves the problem that binarized images are not conducive to marking because the black abnormal positions near the edge will be mixed with the black parts in the normal positions.

[0028] Furthermore, the abnormal locations in the abnormal images screened out by the above-mentioned inverse color overlay or inverse color deduplication methods can be manually marked, and the manually marked pixel locations can be digitized as sample data for training the image detection model.

[0029] Step S3. Extract and split the abnormal locations in the abnormal image independently to form multiple independent graphics by contour detection. Specifically, as shown in Figure 4, extract and split the abnormal locations (black areas) in the abnormal image independently to form multiple independent graphics by contour detection.

[0030] Specifically, contour detection includes the following steps: 1) Boundary tracking: First, scan from the top left corner of the anomalous image to establish the first foreground pixel (contour starting point). Then, starting from this foreground pixel, scan and track along the boundary of the anomalous location until returning to the foreground pixel, thus completing the scanning and recognition of a contour. The boundary tracking above uses an eight-neighbor connection method to confirm the next foreground pixel and the next contour, ultimately extracting and separating each anomalous location independently. 2) Contour direction and hierarchical structure: While performing boundary tracking, record the direction of the contour, which is helpful in confirming the hierarchical relationship between multiple contours (a contour can contain other contours, or be contained by other contours). 3) Recursive search for sub-contours: After completing the boundary tracking of a contour, continue to explore the region inside the contour to find possible sub-contours, and repeat the boundary tracking and hierarchical structure establishment for the found sub-contours. 4) Termination condition: Contour detection ends when all foreground pixels in the anomalous image have been visited and all contours and their sub-contours have been identified.

[0031] Step S4. Determine the minimum bounding rectangle based on the edge pixel positions of each independent graphic. Specifically, in step S4, calculating the minimum bounding rectangle includes the following steps: First, based on the point set of the independent graphic obtained by contour detection; then, find the minimum and maximum values ​​in the X-axis and Y-axis directions respectively from the point set; next, calculate and confirm the coordinates of the upper left endpoint and the rectangle size of the independent graphic based on the minimum and maximum values ​​respectively, where the coordinates of the upper left endpoint are the minimum value of the X-axis and the maximum value of the Y-axis (i.e., the upper left endpoint is (Xmin, Ymax)), the width of the rectangle is the difference between the maximum and minimum values ​​of the X-axis (i.e., width S = Xmax - Xmin), and the width of the rectangle is the difference between the maximum and minimum values ​​of the Y-axis (length L = Ymax - Ymin); finally, establish the minimum bounding rectangle based on the coordinates of the upper left endpoint and the rectangle size.

[0032] S5. The established minimum bounding rectangle is digitized (saved as an XML file) and used as sample data for training the image detection model. This effectively solves the problem that only abnormal points are manually marked in traditional images, automates the marking process, and saves data marking time.

[0033] In conclusion,

[0034] This invention provides an abnormal image detection method based on image comparison technology. It quickly filters out abnormal images in the image to be tested by two methods: inverse color overlay and inverse color deduplication. Through contour detection and calculation of the minimum bounding rectangle, it realizes the accurate marking and digitization of abnormal positions, providing efficient and accurate sample data for the training of image detection models.

[0035] Furthermore, the method of this invention can be combined with deep learning technology to further improve the accuracy and robustness of anomaly image detection. By training a deep learning model, features in images can be automatically learned and extracted, thereby achieving more intelligent and automated anomaly image detection.

[0036] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any way. Any modifications or variations made by those skilled in the art, without departing from the scope of the present invention, using the disclosed technical content, are equivalent embodiments of the present invention. Therefore, all equivalent changes made based on the concept of the present invention without departing from the scope of the present invention should be covered within the protection scope of the present invention.

Claims

1. A method for labeling sample data of an image detection model, characterized in that: Includes the following steps: S1. Establish an image without anomalies as the base image; S2. Obtain the image to be tested and compare it with the base image. The image to be tested and the base image are compared by inverting colors to filter out abnormal images in the image to be tested. S3. Contour detection is used to independently extract and split the abnormal locations in the abnormal image into multiple independent graphics; S4. Determine the minimum bounding rectangle based on the edge pixel position of each individual graphic; S5. The established minimum bounding rectangle is digitized and used as sample data for training the image detection model.

2. The sample data labeling method for an image detection model according to claim 1, characterized in that: The color inversion deduplication includes the following steps: First, the base image is inverted; then, based on the RGB matrix difference between the inverted base image and the image to be tested, the region corresponding to the RGB matrix difference is used as the abnormal image in the image to be tested.

3. The sample data labeling method for an image detection model according to claim 1, characterized in that: The inverted color overlay includes the following steps: First, the contour range image of the image to be tested is obtained through contour detection. Then, the area enclosed by the contour range is filled with black. Next, the filled contour range image is overlaid on the base image to form an overlay image. Finally, the overlay image is inverted and used as an abnormal image in the image to be tested.

4. A sample data labeling method for an image detection model according to claim 2 or 3, characterized in that: Abnormal locations in abnormal images can be manually marked, and the manually marked pixel locations can be digitized and used as sample data for training the image detection model.

5. The sample data labeling method for an image detection model according to claim 1, characterized in that: In step S4, calculating the minimum bounding rectangle includes the following steps: First, based on the point set given by the independent graphics obtained from contour detection; Then, find the minimum and maximum values ​​in the X and Y axes respectively from the point set; next, calculate and confirm the coordinates of the upper left endpoint and the size of the rectangle of the independent figure based on the minimum and maximum values, where the coordinates of the upper left endpoint are the minimum value of the X axis and the maximum value of the Y axis, the width of the rectangle is the difference between the maximum and minimum values ​​of the X axis and the width of the rectangle is the difference between the maximum and minimum values ​​of the Y axis; finally, establish the minimum bounding rectangle based on the coordinates of the upper left endpoint and the size of the rectangle.

Images