Silicon carbide crystal ingot concentration spot image positioning method, system, medium and equipment

Abstract

The invention discloses a silicon carbide crystal ingot concentration spot image positioning method, system, medium and equipment, and the method comprises the following steps: S1, collecting an image of the surface of a positioned crystal ingot through a line scanning image collection device, and carrying out the enhancement processing and filtering preprocessing of the collected image, so as to obtain a calculation image which can be used for calculation; s2, obtaining a trained model; s3, using the trained model to predict the image to obtain a preliminary concentration spot region, then through connected domain analysis, selecting the concentration spot region with the largest area as an undetermined concentration spot region A1, through feature analysis, carrying out coarse positioning on the crystal ingot, then setting a search region proportion, selecting a region geometric limitation region A2 in which a limited concentration spot may appear, and carrying out geometric positioning on the crystal ingot; selecting an intersection area of A1 and A2 as a concentration spot area A; and S4, the external rectangle of the concentration spot area A is calculated, and the starting point and the ending point of each laser cutting line are determined. The method has the advantages of improving the positioning precision and recognition rate of the silicon carbide crystal ingot concentration spots.

Description

Technical Field

[0001] The present invention relates to the field of semiconductors, and particularly to a method, a system, a medium and a device for locating the concentration spot image of a silicon carbide ingot. Background Art

[0002] Due to its excellent physical and chemical properties, silicon carbide (SiC) is widely regarded as the most promising third-generation semiconductor substrate material. Similar to the processing process of traditional semiconductor wafers, SiC substrate wafers are obtained by cutting SiC ingots and subsequent processes such as grinding, polishing, and detection. Among them, ingot cutting is currently widely achieved by laser cutting.

[0003] Due to the limitations of the previous production process, SIC ingots are extremely prone to concentration spot regions (such as Figure 5 the green region in ), and the resistivity value of this region is usually relatively high. If the same laser power as other regions is used during cutting, the ingot in this region cannot be separated. At this time, it is necessary to perform supplementary cutting on this concentration spot region, and accurate positioning of this region is required for supplementary cutting. Through the traceability of screening concentration spots in past ingots, it is found that concentration spots appear in specific orientations of the ingot. When using traditional concentration plate positioning, once an abnormal region exceeding the threshold is screened out, it is determined as a concentration spot, resulting in misjudgment of the concentration spot.

[0004] Traditional positioning is usually based on image positioning. Concentration spots usually appear as irregular closed regions with low gray values in the image. The traditional algorithm performs feature analysis on the entire image. First, several candidate low-gray value regions are roughly segmented through threshold segmentation and connected component analysis, and then the regions are screened by area to locate the concentration spot region. However, in actual situations, the gray value difference between the concentration spot region and the background region is not large, and the shape is not a regular polygon, and the area is not fixed. Therefore, the recognition rate of the traditional algorithm is low, and even if it is recognized, the positioning accuracy is not high. Deep learning methods are very popular technologies in recent years. Through the learning of multiple sample images, a model with multi-dimensional parameters is trained, which can greatly improve the defect recognition rate and positioning accuracy.

[0005] In view of this, it is necessary to provide a method, a system, a medium and a device for locating the concentration spot image of a silicon carbide ingot. Summary of the Invention

[0006] The method, the system, the medium and the device for locating the concentration spot image of a silicon carbide ingot provided by the present invention effectively solve the problems of low recognition rate and low positioning accuracy of the existing silicon carbide ingot concentration spots.

[0007] The technical solution adopted by the present invention is as follows:

[0008] A method for locating the concentration spot image of a silicon carbide ingot includes the following steps: S1. Use a line-scan image acquisition device to acquire the image of the surface of the located ingot, and perform enhancement processing and filtering preprocessing on the acquired image to obtain a calculation image that can be used for calculation; S2. Mark the concentration spot area of the calculation image, sample the marked image, and use a semantic segmentation model to train the sampled marked image to obtain a trained model; S3. Use the trained model to predict the image to obtain a preliminary concentration spot area, and then through connected component analysis, select the concentration spot area with the largest area as the to-be-determined concentration spot area A1. Coarse-locate the ingot through feature analysis to obtain the surface area of the ingot, and then set the search area ratio to select the geometric limit area A2 where the limited concentration spot may appear, and select the intersection area of A1 and A2 as the concentration spot area A; S4. Calculate the circumscribed rectangle of the concentration spot area A, draw vertical cutting lines at equal intervals according to the set step size, and determine the starting point and ending point of each laser cutting line.

[0009] Further: The ingot is positioned by a vacuum stage, the ingot is centered in the X direction and Y direction of the image, the line-scan graphic acquisition device is a camera, the acquisition mode of the camera is the position trigger mode, and the accuracy of the camera in the X direction and Y direction is 0.055mm / Pixel.

[0010] Further: The formula for the enhancement processing in S2 is:

[0011] g′(x,y)=g(x,y) τ (τ=1.8)

[0012]

[0013] where, g(x,y) is the pixel gray value of the coordinate (x,y), τ is the power operation coefficient, g′(x,y) is the image after the power operation; Max(g′(x,y)) is the maximum gray value of the image after the power operation, Min(g′(x,y)) is the minimum gray value of the image after the power operation, and f(x,y) is the finally enhanced image. This method is beneficial to improving the contrast of the image.

[0014] Further: The filtering process in S2 is Gaussian filtering, and the formula for the Gaussian filtering is:

[0015]

[0016] where, w(i,j) is the weight factor, the weight of the Gaussian filtering follows a two-dimensional normal distribution, σ is the standard deviation, n is the Gaussian filtering kernel radius, σ = 1.5, n = (int)3*σ.

[0017] Further: S2 specifically includes the following steps:

[0018] S21. Set two types of labels, namely background and concentration spot labels, and annotate the images with concentration spots.

[0019] S22. Downsample the annotated images, and resize the original images from 4000×4000 to 400×400.

[0020] S23. Load a pre-trained model, set the training parameters, including the number of Epochs as 20.0, the learning rate as 1e-04, the batch size as 1, and evaluate the model performance after training.

[0021] Furthermore: The specific steps of S3 are as follows: S31. Use the trained model to predict the collected images to obtain the preliminary concentration spot regions; S32. Adjust the preliminary obtained concentration spot regions by s times to obtain the concentration spot regions of the actual size; S33. Perform connected component analysis on the concentration spot regions of the actual size, and select the region with the largest area, which is designated as the to-be-determined concentration spot region A1; S34. Perform threshold segmentation on the collected images, select the region with the largest area after connected component analysis, calculate the minimum bounding rectangle region, and then set the search region ratio for the minimum bounding rectangle region to obtain the geometric limited region A2 where the concentration spots may appear; S35. Obtain the concentration spot region A = A1 ∩ A2.

[0022] Furthermore: The specific steps of S4 are as follows: S41. Calculate the bounding rectangle of the concentration spot region A to obtain the coordinates P lt (x1, y1) of the upper left corner point of the rectangle, the width image_width = w of the rectangle, and the height image_height = h of the rectangle; S42. Draw vertical cutting lines l1, l2... l n , n = (int) W / S at equal intervals according to the set step size step = s. The intersection points of each line with the upper and lower limits of the region A are the starting points P start = Up(l n ∩ A) and the ending points P end = Down(l n ∩ A).

[0023] A silicon carbide ingot concentration spot image positioning system, including an image acquisition module, which is used to collect and image the line-scanned image of the ingot, cooperate with a vacuum stage for position triggering and distortion correction; an image preprocessing module, which performs enhancement processing and Gaussian filtering preprocessing on the collected image; a model training module, which is used to label the concentration spot area of the preprocessed image, downsample the labeled image and then input it into a semantic segmentation model for training to obtain a trained model; a region prediction module, which is used to use the trained model to predict the image to obtain a preliminary concentration spot area, and determine the final concentration spot area A through connected domain analysis and feature analysis; an image postprocessing module, which is used to calculate the circumscribed rectangle of the concentration spot area A, draw vertical cutting lines at equal intervals according to a set step size, and determine the starting point and ending point of each laser cutting line.

[0024] A computer-readable storage medium stores a computer program, and when the computer program is processed and executed, it implements the steps of the silicon carbide ingot concentration spot image positioning method.

[0025] A computer device includes a processor, a communication interface, a memory, and a communication bus. The processor, the communication interface, and the memory complete mutual communication through the communication bus. Among them:

[0026] The memory is used to store a computer program;

[0027] The processor is used to execute the steps of the silicon carbide ingot concentration spot image positioning method by running the program stored on the memory.

[0028] Advantages of the invention:

[0029] 1. The present invention combines a deep learning semantic segmentation model with traditional image processing algorithms, realizes high-precision positioning and high-efficiency recognition of the concentration spots of silicon carbide ingots, and overcomes the problems that traditional image processing algorithms are not easy to identify due to the small color difference between the concentration spots of silicon carbide ingots and the background area and the irregular concentration spots are not easy to locate.

[0030] 2. Through the preprocessing methods of enhancing the image and Gaussian filtering, the visibility of the low-contrast concentration spot area is effectively enhanced, and at the same time, the image noise is suppressed. Further improve the recognition rate of the concentration spot area.

[0031] 3. By selecting the intersection of the to-be-determined concentration spot area A1 and the geometric limit area A2 as the concentration spot area A, false detections are excluded, ensuring the accuracy of the concentration spot area A, and further improving the recognition rate and recognition accuracy of the concentration spot area.

[0032] 4. The image enhancement and filtering algorithms can achieve high-precision acquisition with only ordinary industrial cameras, without relying on expensive high-dynamic range devices, effectively saving production costs. Description of the Drawings

[0033] Figure 1 It is a flowchart of the method for locating the concentration spot image of a silicon carbide ingot provided by the embodiment of the present application.

[0034] Figure 2 It is a flowchart of the enhancement processing algorithm for the method for locating the concentration spot image of a silicon carbide ingot provided by the embodiment of the present application.

[0035] Figure 3 It is a flowchart of the Gaussian filtering processing algorithm for the method for locating the concentration spot image of a silicon carbide ingot provided by the embodiment of the present application.

[0036] Figure 4 It is a flowchart of the model training for the method for locating the concentration spot image of a silicon carbide ingot provided by the embodiment of the present application.

[0037] Figure 5 It is a schematic diagram of the ingot spot imaging.

[0038] Figure 6 It is a schematic diagram when the concentration spot area is completely within the geometric limit area when using the method for locating the concentration spot image of a silicon carbide ingot of the present application.

[0039] Figure 7 It is a schematic diagram when part of the concentration spot area is within the geometric limit area when using the method for locating the concentration spot image of a silicon carbide ingot of the present application. Detailed Embodiments

[0040] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following detailed description of the specific embodiments of the present invention will be made with reference to the accompanying drawings.

[0041] As Figure 1 shown, the first embodiment provided by the present application is a method for locating the concentration spot image of a silicon carbide ingot, including the following steps: S1. Use a line-scan image acquisition device to acquire the image of the surface of the located ingot, and perform enhancement processing and filtering preprocessing on the acquired image to obtain a calculation image that can be used for calculation; S2. Use image marking software (such as CVAT) to mark the concentration spot area of the calculation image, sample the marked image, and use a semantic segmentation model to train the sampled marked image to obtain a trained model; S3. Use the trained model to predict the image to obtain a preliminary concentration spot area (i.e., Figure 6 and Figure 7(the irregular area with the cross - hatching shown), and then through connected - component analysis, select the concentration spot area with the largest area as the to - be - determined concentration spot area A1. Through feature analysis, roughly locate the ingot to obtain the ingot surface area, that is Figure 6 and Figure 7 the ingot surface area referred to in. After that, set the search - area ratio, select the geometric restricted area A2 where the defined concentration spot may appear, and select the intersection area of A1 and A2 as the concentration spot area A; S4. Calculate the circumscribed rectangle of the concentration spot area A, draw vertical cutting lines at equal intervals according to the set step size, and determine the starting point and ending point of each laser cutting line.

[0042] The ingot ring - spot statistics are as shown in Table 1 below:

[0043] Ingot number Primary ring spot Secondary ring spot Tertiary ring spot N67008 Success Success Success N75008 Success Success Success N67008 Success Success Success N80015 Success Success Success

[0044] In the above design, the positioning of the ingot concentration - spot image combines deep learning and traditional image analysis, and the positioning accuracy reaches the sub - pixel level. Down - sampling and pre - trained models are used to shorten the processing time, which is suitable for industrial real - time detection. Compared with the previous detection method for concentration spots where the to - be - determined concentration spot area A1 outside the area where the concentration spot should appear is still marked as the concentration - spot area, this application realizes the discrimination of the concentration - spot area A by screening the to - be - determined concentration spot area A1 in the geometric restricted area A2, effectively improving the operation efficiency and the success rate of ring - spot detection.

[0045] Specifically: The ingot is positioned by a vacuum stage, the ingot is centered in the X - direction and Y - direction of the image, the line - scan image acquisition device is a camera, the acquisition mode of the camera is the position - trigger mode, and the accuracy of the camera in both the X - direction and Y - direction is 0.055 mm / Pixel.

[0046] The positioning of the ingot is achieved by adsorbing the lower end face of the ingot by the vacuum stage, ensuring that the ingot does not have relative movement with the vacuum stage during the image - acquisition process. The acquisition mode of the camera is the position - trigger mode, that is, the camera trigger mode is synchronized with the stage movement, and the X / Y - direction centering calibration is performed to eliminate image distortion.

[0047] In the above design, using a vacuum stage to position the ingot avoids the jitter of the ingot during image acquisition. The accuracy of the camera in both the X - direction and Y - direction being 0.055 mm / Pixel is a fixed image accuracy, which can ensure the repeatability of measurement.

[0048] Specifically: As Figure 2 shown, the formula for the enhancement processing is:

[0049] g′(x,y) = g(x,y) τ (τ = 1.8)

[0050]

[0051] Among them, g(x, y) is the pixel gray value of coordinates (x, y), τ is the power operation coefficient, and g′(x, y) is the image after the power operation; Max(g′(x, y)) is the maximum gray value of the image after the power operation, Min(g′(x, y)) is the minimum gray value of the image after the power operation, and f(x, y) is the finally enhanced image. This method is beneficial to improving the contrast of the image.

[0052] In the above design, the power operation of the enhancement process is used to highlight the gray difference between the concentration spot and the background, improve the visibility of low contrast, and improve the subsequent segmentation effect, and can effectively filter out the interference of noise.

[0053] Specifically: as Figure 3 shown, the filtering process is a Gaussian filtering process, and the formula for the Gaussian filtering process is:

[0054]

[0055] Among them, w(i, j) is the weight factor, and the weights of the Gaussian filter follow a two-dimensional normal distribution. σ is the standard deviation, n is the Gaussian filter kernel radius, σ = 1.5, and n = (int)3*σ. The larger σ is, the more uniform the weight distribution is, and the better the filtering effect is. At the same time, the more blurred the image is. On the contrary, the smaller σ is, the more the weight distribution is biased towards the center point of the window, the worse the filtering effect is, and the more the image can retain its original sharpness.

[0056] In the above design, the weights in the Gaussian filtering process are calculated according to the two-dimensional normal distribution, which can smooth the noise while retaining the sharpness of the concentration spot edge, facilitating the imaging of the concentration spot area.

[0057] Specifically: as Figure 4 shown, the S2 specifically includes the following steps:

[0058] S21. Set two types of labels for the background and the concentration spot, and label the images with concentration spots; for the images with concentration spots, use a polyline to draw a closed area, and adjust each corner point of the polyline so that the contour line can more accurately enclose the concentration spot area; repeat the operation until all the collected images are labeled.

[0059] S22. Downsample the labeled images, and adjust the original image from 4000×4000 size to 400×400; that is, perform downsampling by s = 10 times. Essentially, it is to turn the image within the s*s window of the original image into a single pixel, and the value of this pixel is the average value of all the pixels within the window. The formula is as follows:

[0060]

[0061] Where s is the downsampling factor, and using this method can greatly reduce the time required for training and prediction;

[0062] S23. Load the pre-trained model, set the training parameters, load all the images together with the annotation data, create a split, set the training dataset to 70%, the validation dataset to 15%, and the test dataset to 15%. Set the image channel to single-channel, the number of Epochs to 20.0, the learning rate to 1e-04, the number of iterations to 1780, the batch size to 1, and the learning rate to 1e-04. Evaluate the model performance after training. After training, use the test images to evaluate the model, calculate the average IOU of the test images = 94.04%, which meets the requirements, and exit the model training.

[0063] In the above design, downsampling is used to reduce the computational resource requirements and improve the speed of model training. At the same time, the pre-trained model can adapt to concentration spots of different morphologies.

[0064] Specifically: S3 specifically includes the following steps: S31. Use the trained model to predict the collected images to obtain the preliminary concentration spot region; S32. Adjust the preliminary concentration spot region by s times to obtain the concentration spot region of the actual size, that is, restore the downsampled pixels in S22 to the original graph; S33. Perform connected component analysis on the concentration spot region of the actual size, and select the region with the largest area as the to-be-determined concentration spot region A1; S34. Perform threshold segmentation on the collected images, select the region with the largest area after connected component analysis, calculate the minimum bounding rectangle region, and then set the search region ratio for the minimum bounding rectangle region to obtain the geometric constraint region A2 where the concentration spot may appear. The search region ratio of the present invention is 0.5, and thus the position region A2 where the concentration spot may appear can be limited; S35. Obtain the concentration spot region A = A1 ∩ A2.

[0065] In the above design, the intersection of the model prediction region A1 and the geometric constraint region A2 is taken to exclude false detections and ensure the accuracy of the concentration spot region A.

[0066] Specifically: S4 specifically includes the following steps: S41. Calculate the bounding rectangle of the concentration spot region A to obtain the coordinates P lt (x1, y1) of the upper left corner point of the rectangle, the width image_width = w of the rectangle, and the height image_height = h of the rectangle; S42. Draw vertical cutting lines l1, l2... l n , n = (int) W / S at equal intervals according to the set step size step = s. The intersection points of each line with the upper and lower limits of the region A are the starting points P start = Up(l n ∩A) and the ending points Pend = Down(l n ∩ A).

[0067] In the above design, the equally spaced vertical lines are adapted to the laser cutting path planning to improve the accuracy of laser cutting.

[0068] The second embodiment provided by the present application is a concentration spot image positioning system for a silicon carbide ingot, including an image acquisition module for capturing and imaging a line-scan image of the ingot, and cooperating with a vacuum stage for position triggering and distortion correction; an image preprocessing module for performing enhancement processing and Gaussian filtering preprocessing on the acquired image; a model training module for annotating the concentration spot area of the preprocessed image, downsampling the annotated image, and then inputting it into a semantic segmentation model for training to obtain a trained model; a region prediction module for using the trained model to predict the image to obtain a preliminary concentration spot area, and determining the final concentration spot area A through connected component analysis and feature analysis; an image postprocessing module for calculating the circumscribed rectangle of the concentration spot area A, drawing vertical cutting lines at equal intervals according to a set step size, and determining the starting point and ending point of each laser cutting line.

[0069] The third embodiment provided by the present application is a computer-readable storage medium storing a computer program, and when the computer program is processed and executed, it implements the steps of the concentration spot image positioning method for a silicon carbide ingot. In addition, the computer-readable storage medium of this embodiment can adopt any combination of one or more readable storage media, where the readable storage medium includes a system, device, or component of electricity, light, electromagnetism, infrared, or semiconductor, or any combination of the above.

[0070] The fourth embodiment provided by the present application is a computer device, including a processor, a communication interface, a memory, and a communication bus, where the processor, the communication interface, and the memory complete mutual communication through the communication bus: where:

[0071] The memory is used to store a computer program;

[0072] The processor is used to execute the steps of the concentration spot image positioning method for a silicon carbide ingot by running the program stored in the memory. As an implementation manner of the present invention, the communication bus mentioned in the above terminal may be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into an address bus, a data bus, a control bus, etc.

[0073] As an implementation manner of the present invention, the communication interface is used for communication between the above terminal and other devices.

[0074] As an implementation manner of the present invention, the memory may include a random access memory (RAM), or may also include a non-volatile memory, such as at least one disk memory. Optionally, the memory may also be at least one storage device located away from the aforementioned processor.

[0075] As an implementation manner of the present invention, the above processor may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0076] For further detailed description, it should be understood that the above are only specific embodiments of the present invention and are not used to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for locating the concentration spot image of a silicon carbide ingot, characterized in that: The method includes the following steps: S1. Use a line-scan image acquisition device to acquire an image of the surface of the positioned ingot, and perform enhancement processing and filtering preprocessing on the acquired image to obtain a calculation image that can be used for calculation; S2. Mark the concentration spot area of the calculation image, sample the marked image, and use a semantic segmentation model to train the sampled marked image to obtain a trained model; S3. Use the trained model to predict the image to obtain a preliminary concentration spot area, and then through connected component analysis, select the concentration spot area with the largest area as the to-be-determined concentration spot area A1. Coarse position the ingot through feature analysis, and then set the search area ratio to select the geometric limiting area A2 where the defined concentration spot may appear, and select the intersection area of A1 and A2 as the concentration spot area A; S4. Calculate the circumscribed rectangle of the concentration spot area A, draw vertical cutting lines at equal intervals according to the set step size, and determine the starting point and ending point of each laser cutting line.

2. The method for locating the concentration spot image of the silicon carbide ingot according to claim 1, wherein: The ingot is positioned by a vacuum stage. The ingot is centered in the X and Y directions of the image. The line-scan image acquisition device is a camera. The acquisition mode of the camera is the position trigger mode. The accuracy of the camera in the X and Y directions is 0.055 mm / Pixel.

3. The method for locating the concentration spot image of the silicon carbide ingot according to claim 1, wherein: The formula for the enhancement processing is: g′(x,y) = g(x,y) τ (τ = 1.8) where g(x, y) is the pixel gray value at coordinates (x, y), τ is the power operation coefficient, and g′(x, y) is the image after the power operation; Max(g′(x, y)) is the maximum gray value of the image after the power operation, Min(g′(x, y)) is the minimum gray value of the image after the power operation, and f(x, y) is the finally enhanced image. This method is beneficial to improving the contrast of the image.

4. The method for locating the concentration spot image of the silicon carbide ingot according to claim 1, wherein: The filtering process is Gaussian filtering, and the formula for the Gaussian filtering is: where w(i, j) is the weight factor, and the weights of the Gaussian filter follow a two-dimensional normal distribution. σ is the standard deviation, n is the radius of the Gaussian filter kernel, σ = 1.5, and n = (int)3*σ.

5. The method for locating the concentration spot image of a silicon carbide ingot according to claim 1, characterized in that: The specific steps of S2 include the following: S21. Set two types of labels, background and concentration spot, and mark the image with concentration spots. S22. Downsample the marked image, and adjust the original image from a size of 4000×4000 to 400×400. S23. Load a pre-trained model, set the training parameters, and evaluate the model performance after training.

6. The method for positioning the concentration spot image of the silicon carbide ingot according to claim 1, wherein: The specific steps of S3 include the following: S31. Use the trained model to predict the acquired image to obtain a preliminary concentration spot area; S32. Adjust the preliminary obtained concentration spot area by s times to obtain the concentration spot area of the actual size; S33. Perform connected component analysis on the concentration spot area of the actual size, and select the area with the largest area, which is selected as the to-be-determined concentration spot area A1; S34. Perform threshold segmentation on the acquired image, select the area with the largest area after connected component analysis, calculate the minimum circumscribed circle area, and then set the search area ratio to obtain the geometric limiting area A2 where the concentration spot may appear; S35. Obtain the concentration spot area A = A1∩A2.

7. The method for locating the concentration spot image of a silicon carbide ingot according to claim 1, wherein: The specific steps of S4 are as follows: S41. Calculate the circumscribed rectangle of the concentration spot area A to obtain the coordinates P of the upper left corner point of the rectangle lt (x1, y1), the width of the rectangle image_width = w, and the height of the rectangle image_height = h; S42. Draw vertical cutting lines l1, l2... l n , n = (int)w / s, and the intersection points of each line with the upper and lower limits of area A are the starting points P start = Up(l n ∩ A) and the end point P end = Down(l n ∩ A).

8. A silicon carbide ingot concentration spot image positioning system, characterized in that: including An image acquisition module, which is used to acquire and form an image of the ingot by line scanning, cooperate with a vacuum stage for position triggering and distortion correction; An image preprocessing module, which performs enhancement processing and Gaussian filtering preprocessing on the acquired image; A model training module, which is used to label the concentration spot area of the preprocessed image, downsample the labeled image and then input it into a semantic segmentation model for training to obtain a trained model; A region prediction module, which is used to predict the image using the trained model to obtain a preliminary concentration spot area, and determine the final concentration spot area A through connected component analysis and feature analysis; An image postprocessing module, which is used to calculate the circumscribed rectangle of the concentration spot area A, draw vertical cutting lines at equal intervals according to a set step size, and determine the starting point and ending point of each laser cutting line.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is processed and executed, it implements the steps of the method for locating the concentration spot image of the silicon carbide ingot according to any one of claims 1 to 7.

10. A computer device, characterized in that: It includes a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory complete communication with each other through the communication bus: wherein: The memory is used to store a computer program; The processor is used to execute the steps of the method for locating the concentration spot image of the silicon carbide ingot according to any one of claims 1 to 7 by running the program stored on the memory.

Images