Ridge detection method and device based on deep learning model, equipment and storage medium

Abstract

The invention discloses a ridge detection method and device based on a deep learning model, equipment and a storage medium. The method comprises the following steps: acquiring a first to-be-measured image in a monocrystalline silicon shouldering process and a second to-be-measured image in an equal-diameter process in a single-cycle flow of monocrystalline silicon production; and inputting the first to-be-detected image and the second to-be-detected image into a deep learning model for processing, so that the deep learning model outputs a first detection result representing whether the first to-be-detected image has a shouldering broken line ridge in the first gray area, the first detection result is used for representing whether the second to-be-detected image has equal-diameter broken-line ridges or not in the second gray area, and the first detection result is determined based on the number of the shouldering ridges in the first gray area and the protruding size of the shouldering ridges. The second detection result is determined based on the number of equal-diameter ridges in the second gray area and the protruding size of the equal-diameter ridges. According to the invention, the broken line detection efficiency of the monocrystalline silicon ridge line can be improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of crystalline silicon manufacturing, and specifically relates to an edge detection method, device, equipment and storage medium based on a deep learning model. Background Art

[0002] In the production of monocrystalline silicon, it is necessary to detect whether the crystal edges of the monocrystalline silicon are broken during the shouldering process. At present, manual inspection is mostly used to check whether the monocrystalline silicon edges are broken. However, in the actual production environment, manual inspection has the defect of low detection efficiency.

[0003] Therefore, there is an urgent need for a technical solution that can improve the efficiency of detecting broken lines on the edges of single crystal silicon during the shouldering process. Summary of the invention

[0004] In view of the shortcomings of the prior art, the purpose of the present invention is to provide an edge detection method, device, equipment and storage medium based on a deep learning model to solve the problems of low efficiency and low accuracy of edge detection in the prior art.

[0005] According to one aspect of the present application, a method for edge detection based on a deep learning model is disclosed, the method comprising:

[0006] Acquire a first image to be measured in the process of shouldering the single crystal silicon and a second image to be measured in the process of equalizing the diameter of the single crystal silicon in a single cycle process of single crystal silicon production;

[0007] The first image to be tested and the second image to be tested are input into a deep learning model for processing, so that the deep learning model calls a first comparison image under the first same attribute information to compare with the first image to be tested and then outputs a first detection result corresponding to the first image to be tested, and calls a second comparison image under the second same attribute information to compare with the second image to be tested and then outputs a second detection result corresponding to the second image to be tested, wherein the first detection result represents whether there is a shoulder-breaking line edge in the first grayscale area of ​​the first image to be tested, and the second detection result represents whether there is an equal-diameter broken line edge in the second grayscale area of ​​the second image to be tested, the first grayscale area is the area where the shoulder-breaking line is located, and when the number of shoulder-breaking lines and the protruding size of the shoulder-breaking line in the first grayscale area meet a first condition, it is determined that the shoulder-breaking line edge exists in the first grayscale area; the second grayscale area is the area where the equal-diameter edge is located, and when the number of equal-diameter edges and the protruding size of the equal-diameter edges in the second grayscale area meet a second condition, it is determined that the equal-diameter broken line edge exists in the second grayscale area;

[0008] According to the first detection result and the second detection result, the edge target detection result of the single crystal silicon in the single cycle process is determined.

[0009] In some embodiments, the method further comprises:

[0010] Training deep learning models, including:

[0011] Acquire a shoulder-release sample image and an equal-diameter sample image, wherein the shoulder-release sample image includes shoulder-release sample attribute information, the shoulder-release sample attribute information includes shoulder-release sample time and shoulder-release sample rotation speed, and the equal-diameter sample image includes equal-diameter sample attribute information, the equal-diameter sample attribute information includes equal-diameter sample time and equal-diameter sample rotation speed;

[0012] Performing grayscale value detection on the shoulder-released sample image and the equal-diameter sample image respectively to obtain a shoulder-released grayscale region and an equal-diameter grayscale region, wherein the shoulder-released grayscale region includes shoulder-released sample edges with shoulder-released edge protrusion information, and the equal-diameter grayscale region includes equal-diameter sample edges with equal-diameter edge protrusion information, the shoulder-released grayscale region satisfies a first target grayscale value, and the equal-diameter grayscale region satisfies a second target grayscale value, wherein the second target grayscale value is greater than or equal to the first target grayscale value, and the first target grayscale value is greater than a reference grayscale value of a blank image;

[0013] The deep learning model is trained based on the shoulder-dropped sample edges and the equal-diameter sample edges.

[0014] In some embodiments, the first same attribute information is determined based on the shoulder-released sample attribute information, the first same attribute information includes a first same time and a first same speed, the second same attribute information is determined based on the equal-diameter sample attribute information, the second same attribute information includes a second same time and a second same speed, the deep learning model calls the first comparison image under the first same attribute information to compare with the first image to be tested and then outputs a first detection result corresponding to the first image to be tested, and calls the second comparison image under the second same attribute information to compare with the second image to be tested and then outputs a second detection result corresponding to the second image to be tested, including:

[0015] calling the first comparison image corresponding to the first image to be tested at the first same time and the first same speed to compare with the first image to be tested, so as to obtain the number of shoulder ridge lines and the protrusion size of the shoulder ridge lines in the first grayscale area of ​​the first image to be tested;

[0016] Compare the number of shoulder-released ridge lines and the protruding size of the shoulder-released ridge lines with the called first condition to obtain the first detection result;

[0017] calling the second comparison image corresponding to the second image to be tested at the second same time and the second same speed to compare with the second image to be tested, so as to obtain the number of equal-diameter ridges and the protrusion size of the equal-diameter ridges in the second grayscale area of ​​the second image to be tested;

[0018] The number of the equal-diameter ridges and the protruding size of the equal-diameter ridges are compared with the called second condition to obtain the second detection result.

[0019] In some embodiments, the first condition includes a first quantity threshold and a first similarity threshold, the second condition includes a second quantity threshold and a second similarity threshold, and the comparing the number of shoulder-dropped ridge lines and the protruding size of the shoulder-dropped ridge lines with the called first condition to obtain the first detection result includes:

[0020] Comparing the number of the shoulder ridge lines with the first number threshold to obtain a first comparison result;

[0021] Comparing the protrusion size of the shoulder edge line with the edge protrusion size value of the first comparison image for similarity, to obtain a first similarity result, wherein the first similarity result indicates the similarity between the shoulder edge line of the first image to be tested and the first comparison edge line of the first comparison image;

[0022] Obtaining the first detection result according to the first comparison result and the first similarity result;

[0023] The step of comparing the number of equal-diameter ridges and the protruding size of the equal-diameter ridges with the called second condition to obtain the second detection result includes:

[0024] Compare the number of the equal-diameter ridgelines with the second number threshold to obtain a second comparison result;

[0025] Comparing the protrusion size of the equal-diameter edge line with the edge protrusion size value of the second comparison image for similarity, to obtain a second similarity result, wherein the second similarity result represents the degree of similarity between the equal-diameter edge line of the second image to be tested and the second comparison edge line of the second comparison image;

[0026] The second detection result is obtained according to the second comparison result and the second similarity result.

[0027] In some embodiments, obtaining the first detection result according to the first comparison result and the first similarity result includes:

[0028] When at least one of the first comparison result is that the number of the shoulder-dropping ridges is less than the first number threshold, or the first similarity result is that the degree of similarity between the shoulder-dropping ridges and the first comparison ridges of the first comparison image is less than the first similarity threshold, determining that the first detection result is that the first image to be tested has a shoulder-dropping broken line ridge in the first grayscale region;

[0029] The obtaining the second detection result according to the second comparison result and the second similarity result includes:

[0030] When at least one of the second comparison result is that the number of the equal-diameter edges is less than the second number threshold, or the second similarity result is that the degree of similarity between the equal-diameter edges and the second comparison edges of the second comparison image is less than the second similarity threshold, it is determined that the second detection result is that the second image to be tested has equal-diameter broken lines in the second grayscale area.

[0031] In some embodiments, the method further comprises:

[0032] Determine a third grayscale area in the second image to be tested that does not overlap with the second grayscale area, wherein the third grayscale area is a part of the first grayscale area;

[0033] A non-overlapping detection result in the first detection result is determined, where the non-overlapping detection result indicates whether the first image to be tested has a shoulder-breaking edge in the third grayscale region.

[0034] In some embodiments, determining the edge target detection result of the single crystal silicon in a single cycle process according to the first detection result and the second detection result includes:

[0035] When the first detection result is that the first image to be tested has a shoulder-breaking edge in the first grayscale region, or the second detection result is that the second image to be tested has an equal-diameter breaking edge in the second grayscale region, at least one of the two exists, it is determined that the single crystal silicon has a breaking edge in a single cycle process;

[0036] or,

[0037] Determining the edge target detection result of the single crystal silicon in the single cycle process according to the first detection result and the second detection result includes:

[0038] When the non-overlapping detection result is that the first image to be tested has a shoulder-breaking edge in the third grayscale region, it is determined that the single crystal silicon has a breaking edge in a single cycle process.

[0039] According to another aspect of the present application, a ridge detection device based on a deep learning model is also disclosed, the device comprising:

[0040] An image acquisition module for acquiring a first image to be tested in the single crystal silicon shoulder release process and a second image to be tested in the single crystal silicon equal diameter process in a single cycle process of single crystal silicon production;

[0041] The image input module is used to input the first image to be tested and the second image to be tested into a deep learning model for processing, so that the deep learning model calls a first comparison image under the first same attribute information to compare with the first image to be tested and then outputs a first detection result corresponding to the first image to be tested, and calls a second comparison image under the second same attribute information to compare with the second image to be tested and then outputs a second detection result corresponding to the second image to be tested, wherein the first detection result indicates whether there is a shoulder-breaking line edge in the first grayscale area of ​​the first image to be tested, and the second detection result indicates whether there is an equal-diameter broken line edge in the second grayscale area of ​​the second image to be tested, the first grayscale area is the area where the shoulder-breaking line is located, and when the number of shoulder-breaking lines and the protruding size of the shoulder-breaking line in the first grayscale area meet a first condition, it is determined that the shoulder-breaking line edge exists in the first grayscale area; the second grayscale area is the area where the equal-diameter edge is located, and when the number of equal-diameter edges and the protruding size of the equal-diameter edges in the second grayscale area meet a second condition, it is determined that the equal-diameter broken line edge exists in the second grayscale area;

[0042] The target detection result determination module is used to determine the edge target detection result of the single crystal silicon in a single cycle process according to the first detection result and the second detection result.

[0043] According to another aspect of the present application, an electronic device is also disclosed, which includes a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the electronic device executes each step of the edge detection method based on the deep learning model as described in any one of the above items.

[0044] According to another aspect of the present application, a computer-readable storage medium is also disclosed, on which instructions are stored, characterized in that when the instructions are executed by a processor, the various steps of the edge detection method based on the deep learning model as described in any of the above items are implemented.

[0045] The present invention includes but is not limited to the following beneficial effects: (1) By introducing a deep learning model, the present invention can learn and identify the features of edges in a large amount of data, thereby improving the accuracy of edge detection. After the image to be tested is input into the deep learning model, the deep learning model can detect and output the detection results by itself without manual intervention, thereby greatly improving the detection efficiency. (2) By detecting the edges of monocrystalline silicon in a single-cycle process, it can help determine whether production parameters need to be adjusted, thereby improving production efficiency and product quality. In addition, when detecting in a single-cycle process, the images to be tested will not be confused with each other, and the comparison is clear, thereby improving the detection accuracy. (3) The deep learning model of the present application introduces sample images with rich attribute information, such as shoulder release sample time, shoulder release sample rotation speed, equal-diameter sample time and equal-diameter sample rotation speed, which can provide more accurate information for the deep learning model. More training data can be obtained to improve the performance of the model, so that multi-dimensional comparison can be performed when the image to be tested is input into the deep learning model for detection; (4) The present invention can obtain more accurate edge information by performing grayscale value detection on the sample image, thereby improving the accuracy of edge detection; (5) The present invention can accurately compare the number of edges and the protrusion size by setting the number threshold and the similarity threshold, so that the detection of edges is more sensitive and can respond to subtle changes; (6) The present invention improves the comprehensiveness of detection by performing broken line detection on both shoulder edges and equal-diameter edges, making the judgment of edges more accurate and comprehensive, and improving the accuracy of broken line detection; (7) The present invention can perform independent detection on non-overlapping areas, which can increase the sensitivity of detection and make earlier discovery and warning of possible shoulder broken lines. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art are briefly introduced below.

[0047] Figure 1 It is a flow chart of an edge detection method based on a deep learning model in an embodiment of the present application;

[0048] Figure 2 is a flow chart of a training method for a deep learning model according to an embodiment of the present application;

[0049] Figure 3 is another flow chart of the edge detection method based on the deep learning model in the embodiment of the present application;

[0050] Figure 4 is another flow chart of the edge detection method based on the deep learning model in the embodiment of the present application;

[0051] Figure 5 is another flow chart of the edge detection method based on the deep learning model in the embodiment of the present application;

[0052] Figure 6 is another flow chart of the edge detection method based on the deep learning model in the embodiment of the present application;

[0053] Figure 7 is another flow chart of the edge detection method based on the deep learning model in the embodiment of the present application;

[0054] Figure 8 It is a structural block diagram of an edge detection device based on a deep learning model in an embodiment of the present application;

[0055] Fig. 9 It is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0056] An embodiment of the present invention provides an edge detection method based on a deep learning model, the method comprising: obtaining a first image to be tested in the single crystal silicon shoulder release process and a second image to be tested in the single crystal silicon equal diameter process in a single cycle process of single crystal silicon production; inputting the first image to be tested and the second image to be tested into a deep learning model for processing, so that the deep learning model calls a first comparison image under first identical attribute information to compare with the first image to be tested, and then outputs a first detection result corresponding to the first image to be tested, and calls a second comparison image under second identical attribute information to compare with the second image to be tested, and then outputs a second detection result corresponding to the second image to be tested, wherein the first detection result represents the first image to be tested. Whether there is a shoulder-breaking line edge in the first grayscale area of ​​the measured image, the second detection result indicates whether there is an equal-diameter broken line edge in the second grayscale area of ​​the second measured image, the first grayscale area is the area where the shoulder-breaking line is located, when the number of shoulder-breaking lines and the protruding size of the shoulder-breaking line in the first grayscale area meet the first condition, it is determined that the shoulder-breaking line edge exists in the first grayscale area; the second grayscale area is the area where the equal-diameter edge is located, when the number of equal-diameter edges and the protruding size of the equal-diameter edges in the second grayscale area meet the second condition, it is determined that the equal-diameter broken line edge exists in the second grayscale area; according to the first detection result and the second detection result, the edge target detection result of the single crystal silicon in the single cycle process is determined. This technical solution detects broken lines through a deep learning model without manual intervention, greatly improving the detection efficiency, and can detect both the number of edges and the protruding size of the edges, improving the comprehensiveness of the detection, making the judgment of the edges more accurate and comprehensive, and improving the accuracy of broken line detection.

[0057] The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0058] For ease of understanding, the specific process of the embodiment of the present invention is described below. Specifically, Figure 1 This is a flow chart of the edge detection method based on the deep learning model of the present application, and the steps include:

[0059] S100, obtaining a first image to be measured in a process of shouldering single crystal silicon and a second image to be measured in a process of equalizing the diameter of single crystal silicon in a single cycle process of single crystal silicon production.

[0060] Specifically, in this embodiment, an industrial camera can be used to capture images of monocrystalline silicon starting from shoulder release to obtain multiple frames of first images to be tested, and an industrial camera can be used to capture images of monocrystalline silicon starting from equal diameter to obtain multiple frames of second images to be tested. Alternatively, in this embodiment, the first image to be tested and the second image to be tested captured by the industrial camera can be read from an image database. In this embodiment, the industrial camera is used to capture images of monocrystalline silicon according to a preset target duration, and the target duration can be seconds. Thus, a frame of the first image to be tested or a frame of the second image to be tested can be obtained every target duration. Afterwards, in this embodiment, step S102 can be executed for each frame of the image to be tested.

[0061] S102. Input the first image to be tested and the second image to be tested into the deep learning model for processing, so that the deep learning model calls the first comparison image under the first same attribute information to compare with the first image to be tested and then outputs a first detection result corresponding to the first image to be tested, and calls the second comparison image under the second same attribute information to compare with the second image to be tested and then outputs a second detection result corresponding to the second image to be tested.

[0062] Specifically, the first detection result indicates whether there is a shoulder-breaking line in the first grayscale area of ​​the first image to be tested, and the second detection result indicates whether there is an equal-diameter broken line in the second grayscale area of ​​the second image to be tested. The first grayscale area is the area where the shoulder-breaking line is located. When the number of shoulder-breaking lines in the first grayscale area and the protruding size of the shoulder-breaking line meet the first condition, it is determined that there is a shoulder-breaking line in the first grayscale area; the second grayscale area is the area where the equal-diameter line is located. When the number of equal-diameter lines in the second grayscale area and the protruding size of the equal-diameter lines meet the second condition, it is determined that there is an equal-diameter broken line in the second grayscale area. The grayscale value of the first grayscale area in the first image to be tested is greater than the grayscale value of other areas in the second image to be tested, and the grayscale value of the second grayscale area in the second image to be tested is greater than the grayscale value of other areas in the second image to be tested.

[0063] It can be understood that in the embodiment of the application, by introducing a deep learning model, the characteristics of the edge can be learned and identified in a large amount of data, thereby improving the accuracy of edge detection, and after the image to be tested is input into the deep learning model, the deep learning model can detect and output the detection result by itself, without the need for human intervention, thereby greatly improving the detection efficiency, and, by detecting the edges of monocrystalline silicon in a single-cycle process, it can help determine whether the production parameters need to be adjusted, thereby improving production efficiency and product quality, and in the single-cycle process, the images to be tested will not be confused with each other, the comparison is clearer, and the detection accuracy is improved. Furthermore, the present invention improves the comprehensiveness of detection by performing broken line detection on both shoulder edges and equal-diameter edges, making the judgment of edges more accurate and comprehensive, and improving the accuracy of broken line detection.

[0064] S104. Determine the edge target detection result of the single crystal silicon in a single cycle process according to the first detection result and the second detection result.

[0065] The deep learning model can be a model based on machine learning, for example, a machine learning model based on a deep neural network. Specifically, the deep learning model needs to be pre-trained. In one example, the training of the deep learning model in the present application at least includes Figure 2 The following steps are shown:

[0066] S200, obtaining a shoulder-removed sample image and an equal-diameter sample image.

[0067] The shoulder-releasing sample image includes shoulder-releasing sample attribute information, which includes shoulder-releasing sample time and shoulder-releasing sample rotation speed. The equal-diameter sample image includes equal-diameter sample attribute information, which includes equal-diameter sample time and equal-diameter sample rotation speed. Specifically, the shoulder-releasing sample time refers to the specific time when the shoulder-releasing sample image is obtained during the production process, and the shoulder-releasing sample rotation speed refers to the equipment rotation speed when the shoulder-releasing sample image is obtained during the production process; similarly, the equal-diameter sample time refers to the specific time when the equal-diameter sample image is obtained during the production process, and the equal-diameter sample rotation speed refers to the equipment rotation speed when the equal-diameter sample image is obtained during the production process.

[0068] S202, respectively performing grayscale value detection on the shoulder-removed sample image and the equal-diameter sample image to obtain a shoulder-removed grayscale region and an equal-diameter grayscale region.

[0069] In this step, the grayscale value detection is performed on the shoulder-released sample image and the equal-diameter sample image, mainly to identify the edges in the image. Since the color of the edge is different from the surrounding color, the grayscale value corresponding to the edge in the grayscale image will also be different. By detecting the grayscale value, the grayscale area corresponding to these edges can be found. In one example, the edges in the image are black and the background is white. After converting this image to a grayscale image, the grayscale value corresponding to the edge will be close to 0 (black), and the grayscale value corresponding to the background will be close to 255 (white). By setting a suitable threshold, such as 128, the area with a grayscale value less than 128 can be identified as a shoulder-released grayscale area. Similarly, by setting a suitable threshold, such as 120, the area with a grayscale value less than 120 can be identified as an equal-diameter grayscale area.

[0070] S204. Train a deep learning model based on the shoulder-dropped sample edges and the equal-diameter sample edges.

[0071] It can be understood that the deep learning model in this technical solution introduces sample images with rich attribute information, such as shoulder release sample time, shoulder release sample rotation speed, equal-diameter sample time and equal-diameter sample rotation speed, which can provide more training data for the deep learning model, thereby improving the performance of the model, so that when the image to be tested is input into the deep learning model for detection, multi-dimensional comparison can be performed, and by performing grayscale value detection on the sample image, more accurate edge information can be obtained, thereby improving the accuracy of edge detection.

[0072] Furthermore, Figure 3This is another flow chart of an edge detection method based on a deep learning model according to an embodiment of the present invention. In this method, the first same attribute information is determined based on the shoulder-released sample attribute information, the first same attribute information includes the first same time and the first same speed, the second same attribute information is determined based on the equal-diameter sample attribute information, the second same attribute information includes the second same time and the second same speed, and the method steps are further instructions for the steps: the deep learning model calls the first comparison image under the first same attribute information to compare with the first image to be tested and then outputs the first detection result corresponding to the first image to be tested, and calls the second comparison image under the second same attribute information to compare with the second image to be tested and then outputs the second detection result corresponding to the second image to be tested, refer to Figure 3 , including the following steps:

[0073] S300, calling a first comparison image corresponding to the first image to be tested at the first same time and the first same rotation speed to compare with the first image to be tested, so as to obtain the number of shoulder ridges and the protrusion size of the shoulder ridges in the first grayscale area of ​​the first image to be tested.

[0074] The protruding size of the shoulder edge may refer to the size of a shoulder edge in the first image to be tested protruding from the first comparison edge in the first comparison image at the first same time and the first same speed, which may be the protruding length or the protruding area.

[0075] S302, comparing the number of shoulder-released ridge lines and the protruding sizes of the shoulder-released ridge lines with the called first condition to obtain a first detection result.

[0076] In one example, the first condition may include a first quantity threshold and a first similarity threshold. The number of shoulder ridges and the protrusion size of the shoulder ridges are compared with the called first condition, and the first detection result may be specifically: Figure 4 For the related introduction of steps S400-S404 in FIG.

[0077] S304, calling a second comparison image corresponding to the second image to be tested at the second same time and the second same speed to compare with the second image to be tested, so as to obtain the number of equal-diameter ridges in the second grayscale area of ​​the second image to be tested and the protrusion size of the equal-diameter ridges.

[0078] Similarly, the protrusion size of the equal-diameter edge line may refer to the size of a certain equal-diameter edge line in the second image to be tested protruding from the second comparison edge line in the second comparison image at the second same time and the second same speed, which may be the protrusion length or the protrusion area.

[0079] S306, comparing the number of equal-diameter ridges and the protruding sizes of the equal-diameter ridges with the called second condition to obtain a second detection result.

[0080] In one example, the second condition may include a second quantity threshold and a second similarity threshold. The number of equal-diameter ridges and the protrusion size of the equal-diameter ridges are compared with the called second condition. The second detection result may be: Figure 4 For the related introduction of steps S406-S410 in FIG.

[0081] Furthermore, Figure 4 This is another flow chart of the edge detection method based on the deep learning model of the embodiment of the present application, referring to Figure 4 , including the following steps:

[0082] S400: Compare the number of shoulder ridge lines with a first number threshold to obtain a first comparison result.

[0083] Specifically, the number of shoulder ridges is compared with a first quantity threshold to obtain a first comparison result. Specifically, the number of shoulder ridges is compared with the first quantity threshold to obtain a result that the number of shoulder ridges is less than the first quantity threshold, or a result that the number of shoulder ridges is equal to the first quantity threshold, or a result that the number of shoulder ridges is greater than the first quantity threshold.

[0084] S402: Compare the shoulder edge line with the first comparison edge line of the first comparison image to obtain a first similarity result.

[0085] Specifically, the first similarity result represents the similarity between the shoulder edge of the first image to be tested and the first comparison edge of the first comparison image. Specifically, in one example, the first similarity comparison result can be based on the fact that the protrusion size of the shoulder edge of the first image to be tested relative to the first comparison edge at the first same time and the first same speed is smaller than the overall target ratio.

[0086] S404: Obtain a first detection result according to the first comparison result and the first similarity result.

[0087] Specifically, the first comparison result and the first similarity result may be judged separately or in combination, so as to obtain the first detection result based on different judgment results.

[0088] S406: Compare the number of equal-diameter ridges with a second number threshold to obtain a second comparison result.

[0089] Similarly, the number of equal-diameter edges is compared with the second quantity threshold to obtain a second comparison result, which can be specifically comparing the number of equal-diameter edges with the second quantity threshold to obtain a result that the number of equal-diameter edges is less than the second quantity threshold, or a result that the number of equal-diameter edges is equal to the second quantity threshold, or a result that the number of equal-diameter edges is greater than the second quantity threshold.

[0090] S408 , performing a similarity comparison between the equal-diameter edge line and the second comparison edge line protrusion size value of the second comparison image to obtain a second similarity result.

[0091] Specifically, the second similarity result represents the degree of similarity between the equal-diameter edge line of the second image to be tested and the second comparison edge line of the second comparison image. Specifically, in one example, the second similarity comparison result can be based on the fact that the protrusion size of the equal-diameter edge line of the second image to be tested relative to the second comparison edge line at the second same time and the second same speed is smaller than the overall target ratio.

[0092] S410: Obtain a second detection result according to the second comparison result and the second similarity result.

[0093] Specifically, the second comparison result and the second similarity result may be judged separately or in combination, so as to obtain the second detection result based on different judgment results.

[0094] Furthermore, Figure 5 This is another flow chart of the edge detection method based on the deep learning model in the embodiment of the present application. Specifically, Figure 5 is an exemplary description of step S406 and step S410, wherein Figure 5 Step S500 is a specific exemplary description of step S406, and step S502 is a specific exemplary description of step S408. Figure 5 , including the following steps:

[0095] S500, when the first comparison result is that the number of shoulder edges is less than a first number threshold, or the first similarity result is that the degree of similarity between the shoulder edges and the first comparison edges of the first comparison image is less than a first similarity threshold, at least one of the following is determined, the first detection result is determined to be that the first image to be tested has shoulder broken line edges in the first grayscale area.

[0096] S502, when the second comparison result is that the number of equal-diameter edges is less than the second number threshold, or the second similarity result is that the degree of similarity between the equal-diameter edges and the second comparison edges of the second comparison image is less than the second similarity threshold, determine that the second detection result is that the second image to be tested has equal-diameter broken line edges in the second grayscale area.

[0097] It is understandable that the embodiment of the present application can accurately compare the number of ridges and the protrusion size by setting the number threshold and the similarity threshold, so that the detection of ridges is more sensitive and can respond to subtle changes.

[0098] It is understandable that the thresholds in this application can be set in advance and are not specifically limited here.

[0099] Furthermore, Figure 6 This is another flow chart of the edge detection method based on the deep learning model in the embodiment of the present application, referring to Figure 6 , including the following steps:

[0100] S600: Determine a third grayscale area in the second image to be tested that does not overlap with the second grayscale area.

[0101] Specifically, the third grayscale area is a part of the first grayscale area;

[0102] S602: Determine a non-overlapping detection result in the first detection result, where the non-overlapping detection result indicates whether there is a shoulder-breaking edge in the third grayscale region of the first image to be tested.

[0103] It can be understood that in the production process of single crystal silicon, the equal diameter step is after the shoulder release step. When determining the second grayscale area, the second grayscale area may overlap with the first grayscale area or may not overlap. The non-overlapping third grayscale area is directly determined, and it can be further determined whether there is a shoulder release broken line edge in the third grayscale area as the edge target detection result of the single crystal silicon in the single cycle process, for example Figure 7 In step S702, when the non-overlapping detection result is that the first image to be tested has a shoulder-breaking edge in the third grayscale area, it is determined that the single crystal silicon has a broken edge in the single cycle process. Therefore, the present application can independently detect the non-overlapping area, increase the sensitivity of the detection, and detect and warn the possible shoulder-breaking edge earlier.

[0104] Furthermore, Figure 7 This is another flow chart of the edge detection method based on the deep learning model in the embodiment of the present application, referring to Figure 7 , including the following steps:

[0105] S700, when the first detection result is that the first image to be tested has a shoulder broken line in the first grayscale area, or the second detection result is that the second image to be tested has an equal-diameter broken line in the second grayscale area, at least one of the two exists, it is determined that the single crystal silicon has a broken line in a single cycle process.

[0106] S702: When the non-overlapping detection result indicates that the first image to be tested has a shoulder-breaking edge in the third grayscale region, it is determined that the single crystal silicon has a breaking edge in a single cycle process.

[0107] Further, according to one aspect of the present application, an edge detection device based on a deep learning model is also disclosed, such as Figure 8 As shown, the device comprises:

[0108] An image acquisition module for obtaining a first image to be measured in a single crystal silicon shoulder release process and a second image to be measured in a single crystal silicon equal diameter process in a single cycle process of single crystal silicon production;

[0109] The image input module is used to input the first image to be tested and the second image to be tested into the deep learning model for processing, so that the deep learning model calls the first comparison image under the first same attribute information to compare with the first image to be tested and then outputs the first detection result corresponding to the first image to be tested, and calls the second comparison image under the second same attribute information to compare with the second image to be tested and then outputs the second detection result corresponding to the second image to be tested, wherein the first detection result represents whether there is a shoulder-breaking line edge in the first grayscale area of ​​the first image to be tested, and the second detection result represents whether there is an equal-diameter broken line edge in the second grayscale area of ​​the second image to be tested, the first grayscale area is the area where the shoulder-breaking line is located, and when the number of shoulder-breaking lines and the protruding size of the shoulder-breaking line in the first grayscale area meet the first condition, it is determined that there is a shoulder-breaking line edge in the first grayscale area; the second grayscale area is the area where the equal-diameter edge is located, and when the number of equal-diameter edges and the protruding size of the equal-diameter edges in the second grayscale area meet the second condition, it is determined that there is an equal-diameter broken line edge in the second grayscale area;

[0110] The target detection result determination module is used to determine the edge target detection result of the single crystal silicon in a single cycle process according to the first detection result and the second detection result.

[0111] Furthermore, the device also includes:

[0112] The sample image acquisition module is used to acquire shoulder-release sample images and equal-diameter sample images. The shoulder-release sample images include shoulder-release sample attribute information, which includes shoulder-release sample time and shoulder-release sample rotation speed. The equal-diameter sample images include equal-diameter sample attribute information, which includes equal-diameter sample time and equal-diameter sample rotation speed.

[0113] A grayscale value detection module is used to perform grayscale value detection on a shoulder-released sample image and an equal-diameter sample image respectively to obtain a shoulder-released grayscale region and an equal-diameter grayscale region, wherein the shoulder-released grayscale region includes a shoulder-released sample edge line with shoulder-released edge line protrusion information, and the equal-diameter grayscale region includes an equal-diameter sample edge line with equal-diameter edge line protrusion information, the shoulder-released grayscale region satisfies a first target grayscale value, and the equal-diameter grayscale region satisfies a second target grayscale value, wherein the second target grayscale value is greater than or equal to the first target grayscale value, and the first target grayscale value is greater than a reference grayscale value of a blank image.

[0114] The training module is used to train the deep learning model based on the shoulder-dropped sample edges and the equal-diameter sample edges.

[0115] Further, the first same attribute information is determined based on the shoulder-released sample attribute information, the first same attribute information includes a first same time and a first same speed, the second same attribute information is determined based on the equal-diameter sample attribute information, the second same attribute information includes a second same time and a second same speed, and the input module further includes:

[0116] The first calling unit is used to call a first comparison image corresponding to the first image to be tested at the first same time and the first same speed to compare with the first image to be tested, so as to obtain the number of shoulder ridges and the protrusion size of the shoulder ridges in the first grayscale area of ​​the first image to be tested.

[0117] The first comparison unit is used to compare the number of shoulder-released ridge lines and the protruding sizes of the shoulder-released ridge lines with the called first condition to obtain a first detection result.

[0118] The second calling unit is used to call a second comparison image corresponding to the second image to be tested at the second same time and the second same speed to compare with the second image to be tested, so as to obtain the number of equal-diameter ridges in the second grayscale area of ​​the second image to be tested and the protrusion size of the equal-diameter ridges.

[0119] The second comparison unit is used to compare the number of equal-diameter ridges and the protruding size of the equal-diameter ridges with the called second condition to obtain a second detection result.

[0120] Further, the first condition includes a first quantity threshold and a first similarity threshold, the second condition includes a second quantity threshold and a second similarity threshold, and the first comparison unit includes:

[0121] The first quantity comparison subunit is used to compare the quantity of the shoulder ridge lines with a first quantity threshold to obtain a first comparison result.

[0122] The first similarity comparison subunit is used to compare the protrusion size of the shoulder edge with the edge protrusion size value of the first comparison image to obtain a first similarity result, which represents the similarity between the shoulder edge of the first image to be tested and the first comparison edge of the first comparison image.

[0123] The first detection result determination subunit is used to obtain a first detection result according to the first comparison result and the first similarity result.

[0124] The second quantity comparison subunit is used to compare the quantity of equal-diameter ridges with a second quantity threshold to obtain a second comparison result.

[0125] The second similarity comparison subunit is used to compare the protrusion size of the equal-diameter edge line with the edge protrusion size value of the second comparison image to obtain a second similarity result, which represents the degree of similarity between the equal-diameter edge line of the second image to be tested and the second comparison edge line of the second comparison image.

[0126] The second detection result determination subunit is used to obtain a second detection result according to the second comparison result and the second similarity result.

[0127] Furthermore, the first detection result determination subunit is specifically used to determine that the first detection result is that there are shoulder and broken line edges in the first grayscale area of ​​the first image to be tested when at least one of the first comparison result is that the number of shoulder edges is less than a first number threshold, or the first similarity result is that the degree of similarity between the shoulder edges and the first comparison edges of the first comparison image is less than a first similarity threshold.

[0128] The second detection result determination subunit is specifically used to determine that the second detection result is that there are equal-diameter broken line edges in the second grayscale area of ​​the second image to be tested when at least one of the second comparison result is that the number of equal-diameter edges is less than a second number threshold, or the second similarity result is that the degree of similarity between the equal-diameter edges and the second comparison edges of the second comparison image is less than a second similarity threshold.

[0129] Furthermore, the device also includes:

[0130] The third grayscale region determining module is used to determine a third grayscale region in the second image to be tested that does not overlap with the second grayscale region, wherein the third grayscale region is a part of the first grayscale region.

[0131] The non-overlapping detection result determination module is used to determine the non-overlapping detection result in the first detection result, and the non-overlapping detection result indicates whether there is a shoulder and line break edge in the third grayscale area of ​​the first image to be tested.

[0132] Furthermore, the target detection result determination module is specifically used for:

[0133] When the first detection result is that the first image to be tested has a shoulder-breaking edge in the first grayscale region, or the second detection result is that the second image to be tested has an equal-diameter breaking edge in the second grayscale region, at least one of the two exists, it is determined that the single crystal silicon has a breaking edge in the single cycle process;

[0134] or,

[0135] When the non-overlapping detection result is that the first image to be tested has a shoulder-breaking edge in the third grayscale region, it is determined that the single crystal silicon has a breaking edge in the single cycle process.

[0136] The application introduction of the relevant modules of the device in this example can refer to the relevant introduction of the principle of the above method, which will not be repeated here.

[0137] According to another aspect of the present application, the present application also discloses an electronic device, which includes a memory and at least one processor, wherein instructions are stored in the memory; at least one processor calls the instructions in the memory so that the electronic device executes the various steps of the above-mentioned edge detection method based on the deep learning model.

[0138] above Figure 8 The edge detection device based on the deep learning model in the embodiment of the present invention is described in detail from the perspective of modular functional entities, and the electronic device in the embodiment of the present invention is described in detail from the perspective of hardware processing.

[0139] Fig. 9 is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. The electronic device 900 may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) 910 (for example, one or more processors) and a memory 920, and one or more storage media 930 (for example, one or more mass storage devices) storing application programs 933 or data 932. Among them, the memory 920 and the storage medium 930 can be short-term storage or permanent storage. The program stored in the storage medium 930 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations in the electronic device 900. Furthermore, the processor 910 can be configured to communicate with the storage medium 930 to execute a series of instruction operations in the storage medium 930 on the electronic device 900.

[0140] The electronic device 900 may also include one or more power supplies 940, one or more wired or wireless network interfaces 950, one or more input and output interfaces 950, and / or one or more operating systems 931, such as Windows Serve, Mac OS X, Unix, Linux, FreeBSD, etc. It will be appreciated by those skilled in the art that Fig. 9 The structure of the electronic device shown does not constitute a limitation on the electronic device, and may include more or less components than shown in the figure, or combine some components, or arrange the components differently.

[0141] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of an edge detection method based on a deep learning model.

[0142] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described system, device, or unit can refer to the corresponding process in the aforementioned method embodiment and will not be repeated here.

[0143] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc., various media that can store program codes.

[0144] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A ridge detection method based on a deep learning model, characterized in that: The method comprises: Acquire a first image to be measured in the process of shouldering the single crystal silicon and a second image to be measured in the process of equalizing the diameter of the single crystal silicon in a single cycle process of single crystal silicon production; The first image to be tested and the second image to be tested are input into a deep learning model for processing, so that the deep learning model calls a first comparison image under the first same attribute information to compare with the first image to be tested and then outputs a first detection result corresponding to the first image to be tested, and calls a second comparison image under the second same attribute information to compare with the second image to be tested and then outputs a second detection result corresponding to the second image to be tested, wherein the first detection result represents whether there is a shoulder-breaking line edge in the first grayscale area of ​​the first image to be tested, and the second detection result represents whether there is an equal-diameter broken line edge in the second grayscale area of ​​the second image to be tested, the first grayscale area is the area where the shoulder-breaking line is located, and when the number of shoulder-breaking lines and the protruding size of the shoulder-breaking line in the first grayscale area meet a first condition, it is determined that the shoulder-breaking line edge exists in the first grayscale area; the second grayscale area is the area where the equal-diameter edge is located, and when the number of equal-diameter edges and the protruding size of the equal-diameter edges in the second grayscale area meet a second condition, it is determined that the equal-diameter broken line edge exists in the second grayscale area; According to the first detection result and the second detection result, an edge target detection result of the single crystal silicon in a single cycle process is determined.

2. The edge detection method based on deep learning model according to claim 1, characterized in that: The method further comprises: Training deep learning models, including: Acquire a shoulder-release sample image and an equal-diameter sample image, wherein the shoulder-release sample image includes shoulder-release sample attribute information, the shoulder-release sample attribute information includes shoulder-release sample time and shoulder-release sample rotation speed, and the equal-diameter sample image includes equal-diameter sample attribute information, the equal-diameter sample attribute information includes equal-diameter sample time and equal-diameter sample rotation speed; Performing grayscale value detection on the shoulder-released sample image and the equal-diameter sample image respectively to obtain a shoulder-released grayscale region and an equal-diameter grayscale region, wherein the shoulder-released grayscale region includes shoulder-released sample edges with shoulder-released edge protrusion information, and the equal-diameter grayscale region includes equal-diameter sample edges with equal-diameter edge protrusion information, the shoulder-released grayscale region satisfies a first target grayscale value, and the equal-diameter grayscale region satisfies a second target grayscale value, wherein the second target grayscale value is greater than or equal to the first target grayscale value, and the first target grayscale value is greater than a reference grayscale value of a blank image; The deep learning model is trained based on the shoulder-dropped sample edges and the equal-diameter sample edges.

3. The edge detection method based on deep learning model according to claim 2 is characterized in that: The first same attribute information is determined based on the shoulder-released sample attribute information, the first same attribute information includes a first same time and a first same speed, the second same attribute information is determined based on the equal-diameter sample attribute information, the second same attribute information includes a second same time and a second same speed, the deep learning model calls the first comparison image under the first same attribute information to compare with the first image to be tested and then outputs a first detection result corresponding to the first image to be tested, and calls the second comparison image under the second same attribute information to compare with the second image to be tested and then outputs a second detection result corresponding to the second image to be tested, including: calling the first comparison image corresponding to the first image to be tested at the first same time and the first same speed to compare with the first image to be tested, so as to obtain the number of shoulder ridge lines and the protrusion size of the shoulder ridge lines in the first grayscale area of ​​the first image to be tested; Compare the number of shoulder-released ridge lines and the protruding size of the shoulder-released ridge lines with the called first condition to obtain the first detection result; calling the second comparison image corresponding to the second image to be tested at the second same time and the second same speed to compare with the second image to be tested, so as to obtain the number of equal-diameter ridges and the protrusion size of the equal-diameter ridges in the second grayscale area of ​​the second image to be tested; The number of the equal-diameter ridges and the protruding size of the equal-diameter ridges are compared with the called second condition to obtain the second detection result.

4. The edge detection method based on deep learning model according to claim 3 is characterized in that: The first condition includes a first quantity threshold and a first similarity threshold, the second condition includes a second quantity threshold and a second similarity threshold, and the comparison of the number of shoulder-dropped ridgelines and the protrusion size of the shoulder-dropped ridgelines with the called first condition to obtain the first detection result includes: Comparing the number of the shoulder ridge lines with the first number threshold to obtain a first comparison result; Comparing the protrusion size of the shoulder edge line with the edge protrusion size value of the first comparison image for similarity, to obtain a first similarity result, wherein the first similarity result indicates the similarity between the shoulder edge line of the first image to be tested and the first comparison edge line of the first comparison image; Obtaining the first detection result according to the first comparison result and the first similarity result; The step of comparing the number of equal-diameter ridges and the protruding size of the equal-diameter ridges with the called second condition to obtain the second detection result includes: Compare the number of the equal-diameter ridges with the second number threshold to obtain a second comparison result; Comparing the protrusion size of the equal-diameter edge line with the edge protrusion size value of the second comparison image for similarity, to obtain a second similarity result, wherein the second similarity result represents the degree of similarity between the equal-diameter edge line of the second image to be tested and the second comparison edge line of the second comparison image; The second detection result is obtained according to the second comparison result and the second similarity result.

5. The edge detection method based on deep learning model according to claim 4 is characterized in that: The obtaining the first detection result according to the first comparison result and the first similarity result includes: When at least one of the first comparison result is that the number of the shoulder-dropping ridges is less than the first number threshold, or the first similarity result is that the degree of similarity between the shoulder-dropping ridges and the first comparison ridges of the first comparison image is less than the first similarity threshold, determining that the first detection result is that the first image to be tested has a shoulder-dropping broken line ridge in the first grayscale region; The obtaining the second detection result according to the second comparison result and the second similarity result includes: When at least one of the second comparison result is that the number of the equal-diameter edges is less than the second number threshold, or the second similarity result is that the degree of similarity between the equal-diameter edges and the second comparison edges of the second comparison image is less than the second similarity threshold, it is determined that the second detection result is that the second image to be tested has equal-diameter broken lines in the second grayscale area.

6. The edge detection method based on deep learning model according to claim 5, characterized in that: The method further comprises: Determine a third grayscale area in the second image to be tested that does not overlap with the second grayscale area, wherein the third grayscale area is a part of the first grayscale area; A non-overlapping detection result in the first detection result is determined, where the non-overlapping detection result indicates whether the first image to be tested has a shoulder-breaking edge in the third grayscale region.

7. The edge detection method based on deep learning model according to claim 6 is characterized in that: Determining the edge target detection result of the single crystal silicon in the single cycle process according to the first detection result and the second detection result includes: When the first detection result is that the first image to be tested has a shoulder-breaking edge in the first grayscale region, or the second detection result is that the second image to be tested has an equal-diameter breaking edge in the second grayscale region, at least one of the two exists, it is determined that the single crystal silicon has a breaking edge in a single cycle process; or, Determining the edge target detection result of the single crystal silicon in the single cycle process according to the first detection result and the second detection result includes: When the non-overlapping detection result is that the first image to be tested has a shoulder-breaking edge in the third grayscale region, it is determined that the single crystal silicon has a breaking edge in a single cycle process.

8. An edge detection device based on a deep learning model, characterized in that: The device comprises: An image acquisition module for acquiring a first image to be tested in the single crystal silicon shoulder release process and a second image to be tested in the single crystal silicon equal diameter process in a single cycle process of single crystal silicon production; The image input module is used to input the first image to be tested and the second image to be tested into a deep learning model for processing, so that the deep learning model calls a first comparison image under the first same attribute information to compare with the first image to be tested and then outputs a first detection result corresponding to the first image to be tested, and calls a second comparison image under the second same attribute information to compare with the second image to be tested and then outputs a second detection result corresponding to the second image to be tested, wherein the first detection result indicates whether there is a shoulder-breaking line edge in the first grayscale area of ​​the first image to be tested, and the second detection result indicates whether there is an equal-diameter broken line edge in the second grayscale area of ​​the second image to be tested, the first grayscale area is the area where the shoulder-breaking line is located, and when the number of shoulder-breaking lines and the protruding size of the shoulder-breaking line in the first grayscale area meet a first condition, it is determined that the shoulder-breaking line edge exists in the first grayscale area; the second grayscale area is the area where the equal-diameter edge is located, and when the number of equal-diameter edges and the protruding size of the equal-diameter edges in the second grayscale area meet a second condition, it is determined that the equal-diameter broken line edge exists in the second grayscale area; The target detection result determination module is used to determine the edge target detection result of the single crystal silicon in a single cycle process according to the first detection result and the second detection result.

9. An electronic device, characterized in that: The electronic device includes a memory and at least one processor, wherein instructions are stored in the memory; the at least one processor calls the instructions in the memory so that the electronic device executes each step of the edge detection method based on a deep learning model as described in any one of claims 1-7.

10. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the various steps of the edge detection method based on the deep learning model as described in any one of claims 1 to 7 are implemented.

Images