Method for training image-recognition model, method for inspecting spinneret, device for training image-recognition model, device for inspecting spinneret, electronic device and program

Abstract

To provide a method for training an image-recognition model, a method for inspecting a spinneret, a device for training an image-recognition model, a device for inspecting a spinneret, an electronic device and a program.SOLUTION: A method includes: processing a spinneret sample image based on a first image-recognition model; and acquiring label-state information of each small hole in the spinneret sample image, the spinneret sample image being a projection image of light emitted from a light source and formed on an imaging member after passing through each small hole of the spinneret sample and being enlarged by a magnifying member and the label-state information of each small hole being used for representing a shape state of each small hole corresponding thereto. The method also includes: training a pre-set image-recognition model and acquiring a target image-recognition model, based on the spinneret sample image and the label-state information of each small hole; and automatically inspect the state information of each small hole in a spinneret by the target image-recognition model.SELECTED DRAWING: Figure 3

Description

Technical Field

[0001] The present disclosure relates to the field of computer technology, particularly to the field of image recognition and deep learning technology, and specifically relates to a method for training an image recognition model, a method and device for inspecting a spinneret.

Background Art

[0002] In the spinning manufacturing process, the spinneret converts a viscous molten polymer or polymer solution into a thin stream having a specific cross-sectional shape through pores, and forms a yarn after solidification through a coagulation medium or coagulation bath. Particles such as mechanical impurities, gels, carbon fibers, and thermal cracks present in the melt tend to block the pores of the spinneret, thereby causing non-uniformity in the fineness of the raw yarn and defects such as neps, thin yarns, and hairy yarns. Therefore, it is necessary to regularly inspect the spinneret.

Summary of the Invention

Problems to be Solved by the Invention

[0003] The present disclosure provides a method for training an image recognition model, a method for inspecting a spinneret, a device for training an image recognition model, a device for inspecting a spinneret, an electronic device, and a program to solve or alleviate one or more technical problems in the prior art.

Means for Solving the Problems

[0004] In a first aspect, the present disclosure provides a method for training an image recognition model, the method comprising: processing a spinneret sample image based on a first image recognition model to obtain label state information of each pore in the spinneret sample image, where the spinneret sample image is a projection image formed on an imaging member after light emitted from a light source passes through each pore of the spinneret sample and is enlarged by an enlarging member, and the label state information of each pore is used to represent the shape state of the corresponding pore; Based on the spinneret sample image and the label status information of each pore, training a preset image recognition model to obtain a target image recognition model, where the target image recognition model is used to process the spinneret projection image to obtain the status information of each pore in the spinneret, and, including.

[0005] In a second aspect, the present disclosure provides a spinneret inspection method, the method comprising: A gripping member for gripping the spinneret, a light source and an imaging member respectively arranged on both sides of the gripping member, and a magnifying member arranged between the imaging member and the gripping member, wherein the light emitted from the light source passes through each inspection target pore of the spinneret and is magnified by the magnifying member, and then forms a spinneret projection image on the imaging member, and a magnifying member, and is a spinneret inspection method used in a spinneret inspection device comprising: Obtaining a spinneret projection image on the imaging member; Processing the spinneret projection image using the target image recognition model to obtain the status information of each inspection target pore in the spinneret, wherein the target image recognition model is trained based on any one of the training methods in the above embodiments, and, including. Determining the inspection result of the spinneret based on the status information of each inspection target pore, and, including.

[0006] In a third aspect, the present disclosure provides a training device for an image recognition model, the device comprising: A processing unit for processing the spinneret sample image based on the first image recognition model to obtain the label status information of each pore in the spinneret sample image, where the spinneret sample image is a projection image formed on the imaging member after the light emitted from the light source sequentially passes through each pore of the spinneret sample and is magnified by the magnifying member, and the label status information of each pore is used to represent the shape status of the corresponding pore, and a processing unit; A training unit that trains a preset image recognition model based on a spinneret sample image and label state information of each pore to obtain a target image recognition model, wherein the target image recognition model is used to process a spinneret projection image to obtain state information of each pore in the spinneret, and the training unit is provided.

[0007] In a fourth aspect, the present disclosure provides a spinneret inspection device, which A gripping member for gripping a spinneret, a light source and an imaging member respectively arranged on both sides of the gripping member, and an enlarging member arranged between the imaging member and the gripping member, wherein the light emitted from the light source passes through each inspection target pore of the spinneret and is enlarged by the enlarging member, and then forms a spinneret projection image on the imaging member, and the enlarging member is a spinneret inspection device used in a spinneret inspection device provided with An acquisition unit for acquiring a spinneret projection image on the imaging member, and A prediction unit for processing a spinneret projection image using a target image recognition model to obtain state information of each inspection target pore in the spinneret, wherein the target image recognition model is trained based on any one of the above training methods, and the prediction unit is provided. A determination unit for determining an inspection result of the spinneret based on the state information of each inspection target pore is provided.

[0008] In a fifth aspect, the present disclosure provides an electronic device, which At least one processor, and A memory communicatively connected to the at least one processor, and The memory stores instructions executable by the at least one processor, and when the instructions are executed by the at least one processor, they cause any one of the methods in the embodiments of the present disclosure to be executed.

[0009] In a sixth aspect, there is provided a non-transitory computer-readable storage medium storing computer instructions for causing a computer to execute any one of the methods in the embodiments of the present disclosure.

[0010] In a seventh aspect, there is provided a program which, when executed by a processor, implements any one of the methods in the embodiments of the present disclosure.

[0011] With the image recognition model training method, spinneret inspection method and device according to the embodiments of the present disclosure, it is possible to automatically inspect the state information of each inspection target pore of the spinneret and determine the inspection result of the spinneret, eliminating the need for manual inspection, reducing the manpower, and improving the inspection efficiency.

[0012] It should be understood that the content described herein is not intended to describe the key points or important features of the embodiments of the present disclosure, nor is it used to limit the scope of the present disclosure. For other features of the present disclosure, understanding is promoted through the following specification.

Brief Description of the Drawings

[0013] In the accompanying drawings, unless otherwise specified, the same reference numerals indicate the same or similar components or elements throughout the plurality of accompanying drawings. These accompanying drawings are not necessarily drawn to scale. It should be understood that these drawings show only some embodiments provided by the present disclosure and should not be regarded as limiting the scope of the present disclosure.

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Mode for Carrying Out the Invention

[0014] Hereinafter, the present disclosure will be described in more detail with reference to the accompanying drawings. In the accompanying drawings, the same reference numerals represent the same or similar elements. Also, in the accompanying drawings, although various aspects of the embodiments are shown, these accompanying drawings are not necessarily drawn to scale unless otherwise stated.

[0015] Furthermore, in order to better explain the present disclosure, many specific details are described in the following specific embodiments. Those skilled in the art should understand that the present disclosure can be implemented similarly even without some details. In some embodiments, methods, means, components, circuits, etc. that are well known to those skilled in the art are not described in detail so that the gist of the present disclosure can be made clear.

[0016] In the related art, the inspection of the spinneret needs to be carried out manually, and the operator needs to inspect whether there are blockages or cracks in each pore on the spinneret. However, since the number of pores on the spinneret is large and the size of the pores is relatively small, it takes time to inspect the spinneret.

[0017] To solve at least one of the above problems, in an embodiment of the present disclosure, there are provided a method for training an image recognition model, a spinneret inspection method, a device for training an image recognition model, a spinneret inspection device, an electronic device, and a program. Based on a first image recognition model, a spinneret sample image is processed to obtain label state information of each pore in the spinneret sample image. Here, the spinneret sample image is a projection image formed on an imaging member after light emitted from a light source sequentially passes through a spinneret sample and a magnifying member. The label state information of each pore is used to represent the shape state of the corresponding pore. Based on the spinneret sample image and the label state information of each pore, a preset image recognition model is trained to obtain a target image recognition model. The target image recognition model is used to process a spinneret projection image to obtain state information of each pore in the spinneret. By using the target image recognition model, the state information of each pore of the spinneret can be automatically inspected, so that the inspection result of the spinneret can be determined, manual inspection is unnecessary, the number of personnel can be reduced, and the inspection efficiency can be improved.

[0018] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings.

[0019] FIG. 1 is a schematic diagram showing the configuration of a spinneret inspection apparatus according to an embodiment of the present disclosure. FIG. 2 is a schematic diagram showing the configuration of the spinneret in FIG. 1.

[0020] Referring to FIGS. 1 and 2, an embodiment of the present disclosure provides a spinneret inspection apparatus including a gripping member 110, a light source 120, an imaging member 130, and a magnifying member 140. The gripping member 110 is used to grip a spinneret 200. The light source 120 and the imaging member 130 are respectively disposed on both sides of the gripping member 110. The magnifying member 140 is disposed between the imaging member 130 and the gripping member 110. After the light emitted from the light source 120 passes through each inspection target pore 210 of the spinneret 200 and is magnified by the magnifying member 140, a spinneret projection image can be formed on the imaging member 130.

[0021] In addition, the spinneret 200 may have a plurality of minute holes abbreviated as pores. The pores of the spinneret to be inspected are referred to as inspection target pores 210, and the shape of the inspection target pores 210 may be various shapes such as circular, triangular, rice-shaped, linear, caterpillar-shaped, etc.

[0022] In the spinning manufacturing process, the spinneret converts a viscous molten polymer or polymer solution into a thin stream having a specific cross-sectional shape through the pores, and forms a yarn by coagulating it through a coagulation medium or coagulation bath. Fine particles such as mechanical impurities, gels, carbon fibers, and thermal cracks present in the melt tend to block the pores of the spinneret (the shape of the pores becomes irregular and the area shrinks), or cause problems such as cracking of the pores (the shape of the pores becomes irregular and the area expands). These problems change the shape of the pores and further affect the quality of the formed fiber filaments.

[0023] The main types of fiber filaments according to the solution of the embodiment of the present disclosure can include one or more of partially oriented yarns (POY), fully drawn yarns (FDY), polyester staple fiber, etc. Specific examples of the types of filaments include, for example, polyester partially oriented yarns, polyester fully drawn yarns, polyester drawn yarns, polyester staple fiber, etc.

[0024] The gripping member 110 of the spinneret inspection device can be used to grip or fix the spinneret, and this may have a structure like a grip.

[0025] The light source 120 has a structure capable of emitting a light beam, and may be, for example, a light-emitting diode or the like. In some embodiments, the light source 120 can emit parallel light, the diameter of the light source is larger than the diameter of the spinneret, and their center lines may overlap.

[0026] The enlarging member 140 may be a general device that enables amplification of an optical signal, such as a structure like a convex lens.

[0027] The imaging member 130 may have a structure capable of presenting an image such as a screen, etc., and may be perpendicular to the axis of the light emitted from the light source.

[0028] The light emitted from the light source 120 can pass through the spinneret 200, pass through the inspection target pores 210 of the spinneret 200 to form an imaging beam. After passing through the enlarging member 140 and being enlarged, the imaging beam is irradiated onto the imaging member 130 and imaged, so that the shape of each inspection target pore is enlarged and projected onto the imaging member 130 and presented. When inspecting a spinneret sample through the spinneret inspection device, the projected image on the imaging member 130 can be used as the spinneret sample image, and by training a preset image recognition model, a target image recognition model capable of accurately outputting the state information of each inspection target pore in the spinneret can be obtained.

[0029] Hereinafter, a method for training the image recognition model will be described. FIG. 3 is a schematic flowchart of a method for training an image recognition model according to another embodiment of the present disclosure. Referring to FIG. 3, the embodiment of the present disclosure further provides a method 300 for training an image recognition model including the following steps S301 to S302.

[0030] In step S301, the spinneret sample image is processed based on the first image recognition model to obtain the label state information of each pore in the spinneret sample image. Here, the spinneret sample image is a projection image formed on the imaging member after the light emitted from the light source passes through each pore of the spinneret sample and is enlarged by the enlarging member. The label state information of each pore is used to represent the shape state of the corresponding pore.

[0031] In step S302, based on the spinneret sample image and the label state information of each pore, a preset image recognition model is trained to obtain a target image recognition model. The target image recognition model is used to process the spinneret projection image to obtain the state information of each pore in the spinneret.

[0032] The spinneret sample image may be a spinneret projection image obtained by passing through the above spinneret inspection device. This projection image contains the shape of each pore in the spinneret sample. Generally, since the size of the pore is small, if the spinneret image is directly used as the sample image, the quality of the sample is poor and it is likely to affect the accuracy of the finally obtained trained model. On the other hand, in this embodiment, since the spinneret projection image is enlarged by the enlarging member, the quality of the sample can be improved and the accuracy of the target image recognition model can be improved.

[0033] The first image recognition model may be an image recognition, object detection, or image segmentation model. For example, it may be a Segment Anything (SA) model, or a YOLOv8 model (you only look once version 8), etc. Using these models, the pore and pore label status information in the image can be identified relatively quickly. This pore label status information can be used to represent the shape status of the pore. Taking a circular pore as an example, its label status information is a perfect circle. Taking a clogged or cracked pore as an example, its label status information can be non-circular. Of course, the pores may have other shapes, and this can be inferred accordingly.

[0034] It can be understood that the first image recognition model is a pre-trained model in the related art and can obtain relatively accurate pore label status information.

[0035] In step S301, the spinneret sample image is input into the first image recognition model to obtain the label status information of each pore.

[0036] The first image recognition model can quickly obtain the label status information of each pore in the spinneret sample image without manual labeling.

[0037] In step S302, a preset image recognition model is trained using the spinneret sample image and the label status information to obtain a trained image recognition model, that is, a target image recognition model.

[0038] The target image recognition model can be used to process the spinneret projection image obtained using the spinneret inspection device, thereby obtaining the status information of each pore. Furthermore, it can be understood that the pass or fail of this spinneret can be inspected. In this process, manual inspection is not required, saving labor and time, and having higher inspection efficiency and accuracy.

[0039] In some embodiments, in step S301, processing the spinneret sample image based on the first image recognition model to obtain the label state information of each pore in the spinneret sample image means that inputting the spinneret sample image into a first sub-model in the first image recognition model to obtain a first information set for each pore, where the first information set includes the first state information of each pore, the first position information of each pore, and the first confidence corresponding to the first state information of each pore, and inputting the spinneret sample image into a second sub-model in the first image recognition model to obtain a second information set for each pore, where the second information set includes the second state information of each pore, the second position information of each pore, and the second confidence corresponding to the second state information of each pore, and determining the label state information of each pore based on the first information set and the second information set.

[0040] In this embodiment, the first image recognition model can include two sub-models such as a first sub-model and a second sub-model, and they may be different models.

[0041] By processing the spinneret sample image using the first sub-model, a first information set can be obtained, and this first information set can include the first state information of each pore, the first position information, and the first confidence.

[0042] Here, the first state information is used to represent the shape state of this pore, and the first position information is the position of this pore in the spinneret sample image. Since there are many pores in the spinneret, it can be understood that each pore can be identified by the first position information. The first confidence is the probability of the first state information. When the first state information is a perfect circle, the first confidence is the probability of that perfect circle.

[0043] Similarly, by processing the spinneret sample image using the second sub-model, a second information set can be obtained, and this second information set can include the second state information of each pore, the second position information, and the second reliability.

[0044] Here, the second state information is used to represent the shape state of this pore, the second position information is the position of the pore in the spinneret sample image, and since there are many pores in the spinneret, it can be understood that each pore can be identified by the second position information. The second reliability is the probability of the second state information. When the second state information is a perfect circle, the second reliability is the probability of that perfect circle.

[0045] Based on the first information set and the second information set, the label state information of each pore can be obtained.

[0046] It can be understood that by recognizing the spinneret sample image through different models and synthesizing the results of both to obtain the label state information of each pore, the accuracy of the label state information can be improved and the sample quality can be enhanced.

[0047] In one embodiment, the first sub-model can be, for example, a segmenting model, and higher accuracy of the first information set can be obtained. The second sub-model can be, for example, a YOLOv8 model, and the second information set can be obtained more quickly.

[0048] In other embodiments, the first image recognition model may be a model that can output the label state information of the pores, trained using artificially labeled samples.

[0049] In some embodiments, determining the label state information of each pore based on the first information set and the second information set is Determining a subset of pore state information based on the first position information in the first information set and the second position information in the second information set, where the subset of state information includes the first state information and the second state information of the same pore, For the subset of the state information of the first pore in each pore, when the first state information of the first pore does not match the second state information of the first pore, determining the overall confidence level of the first pore based on the first confidence level corresponding to the first state information and the second confidence level corresponding to the second state information, Determining the labeled state information of the first pore based on the overall confidence level of the first pore and the confidence threshold, Determining the labeled state information of each pore based on at least the labeled state information of the first pore,

[0050] Since both the first submodel and the second submodel are processing the spinneret sample image, it can be understood that the position information obtained from both can be located in the same coordinate system.

[0051] By matching the first position information of the pore and the second position information of the pore, it is possible to determine whether two pores belong to the same pore. When both belong to the same pore, it is possible to determine to add the first state information and the second state information of this pore to the subset of the state information of this pore, whereby one subset of state information can be determined for each pore.

[0052] Also, taking the first pore in each pore as an example, in the subset of the state information of this first pore, when the first state information and the second state information in the first pore match (their meanings are the same), this first state information is used as the labeled state information of the first pore. When the first state information and the second state information in the first pore do not match (for example, one is a perfect circle and the other is a non-perfect circle, and their meanings are different), the first confidence level and the second confidence level of the first pore are obtained, and the overall confidence level of the first pore can be determined from the first confidence level and the second confidence level.

[0053] For example, the first reliability indicates that the probability of the first pore being a perfect circle is 85%, and the second reliability indicates that the probability of the first pore being non-circular is 65%. Using 1 - 65% = 35%, it is possible to obtain that the probability of the first pore being a perfect circle is 35%. Based on 85% and 35%, the overall reliability of the first pore can be obtained by the method of cumulative addition, and the overall reliability can determine the overall probability that the first pore is a perfect circle. This method comprehensively considers each reliability, and the obtained overall reliability is relatively accurate. Then, by comparing the integrated reliability threshold with the reliability threshold, the final state information of the first pore, that is, the label state information, is determined.

[0054] By performing the above processing on each pore, the label state information of each pore can be obtained. In this embodiment, when the first state information and the second state information do not match, by determining the overall reliability, more accurate label state information can be obtained, which is beneficial for improving the quality of the sample and the accuracy of the trained model.

[0055] In some embodiments, in step S302, based on the spinneret sample image and the label state information of each pore, training a preset image recognition model to obtain a target image recognition model includes: processing the spinneret sample image based on the preset image recognition model to obtain the predicted state information of each pore; determining a first loss function based on the predicted state information of each pore and the label state information of each pore; obtaining a second loss function by adding a temperature scalar to the first loss function, where the temperature scalar balances the sensitivity of the preset image inspection model to the differences between pores of different shape states; and adjusting the parameters of the preset image recognition model based on the second loss function to obtain a target image recognition model.

[0056] In this embodiment, by inputting the spinneret sample image into a preset image recognition model, the predicted state information of each pore, that is, the predicted shape state of each pore, can be obtained.

[0057] And based on the predicted state information and the label state information of each pore, the first loss function can be determined.

[0058] The temperature scalar is used to adjust the temperature parameter in the first loss function in order to balance the sensitivity of the preset image inspection model to the differences between pores of different shape categories, and to improve the accuracy of classification and recognition. The temperature parameter is a parameter that controls the relative distance between categories.

[0059] In this embodiment, the first loss function can be realized by using the following formula.

Equation

[0060] Adding the temperature scalar w to the first loss function loss 1 (x, y), that is, adding the temperature scalar w as an adjustable variable in the second loss function to the second loss function, and the obtained second loss function loss 2 (x, y, w) can be realized using the following formula.

Equation

[0061] By the second loss function of the above embodiment, the parameters of the preset image recognition model can be adjusted to obtain a target loss function. The temperature scalar is advantageous for enhancing the flexibility of the loss function and further improving the performance of the model.

[0062] In some embodiments, obtaining the second loss function by adding the temperature scalar to the first loss function Based on the temperature scalar, determining a dynamic temperature function for representing the change relationship of the temperature scalar with time Adding the dynamic temperature function to the first loss function to obtain the second loss function, includes.

[0063] In this embodiment, the dynamic temperature function f can also be introduced by the temperature scalar w w (t),

Equation

[0064] Adding the dynamic temperature function to the first loss function, the obtained second loss function loss 2(x, y, w) is as follows.

Number

[0065] Here, δ(y i ) is the classification scaling coefficient corresponding to y, and δ(y) is the classification scaling coefficient corresponding to y. i Since the temperature can rise or fall with time, the second loss function having a dynamic temperature function can make the prediction distribution smoother or sharper, and can facilitate the convergence of the second loss function.

[0066]

[0067] In some embodiments, determining the first loss function based on the predicted state information of each pore and the labeled state information of each pore includes determining the first loss function based on the predicted state information of each pore, the labeled state information of each pore, and the loss function of the first image recognition model.

[0068] In this embodiment, the first loss function can also be obtained by associating with the loss function of the first image recognition model. For example, by using the loss function of the first image recognition model as one reference item of the first loss function, the first loss function can be obtained by associating with referring to the first image recognition model, and the determination process of the loss function can be simplified.

[0069] ​FIG. 4 is a schematic flowchart of a spinneret inspection method according to an embodiment of the present disclosure. Referring to FIGS. 1, 2, and 4, an embodiment of the present disclosure includes a gripping member 110 that grips a spinneret 200, a light source 120 and an imaging member 130 disposed on both sides of the gripping member 110, and an enlarging member 140 disposed between the imaging member 130 and the gripping member 110. The light emitted from the light source 120 passes through each inspection target pore 210 of the spinneret 200 and is enlarged by the enlarging member 140, and then a spinneret projection image is formed on the imaging member 130. The present disclosure provides a spinneret inspection method 400 used in a spinneret inspection apparatus including the enlarging member 140. The method 400 includes steps S401 to S403.

[0070] In step S401, a spinneret projection image on the imaging member is acquired.

[0071] In step S402, the spinneret projection image is processed using a target image recognition model to obtain state information of each inspection target pore in the spinneret. The target image recognition model is trained based on any of the above training methods.

[0072] In step S403, based on the state information of each inspection target pore, an inspection result of the spinneret is determined.

[0073] Regarding the configuration of the spinneret inspection device, reference can be made to the description of the above embodiment, and thus it will not be repeatedly described. In step S401, an imaging component such as a camera in the spinneret inspection device can capture the projection image on the imaging member to obtain a spinneret projection image.

[0074] Next, the spinneret projection image can be processed using a target image recognition model trained using the above training method, and the state information of each inspection target pore in the spinneret, that is, the shape state of each inspection target pore, can be output.

[0075] In step S403, based on the state information of each pore to be inspected, the inspection result of the spinneret can be obtained.

[0076] In the inspection method of this embodiment, by using the target image recognition model to process the spinneret projection image, the state information of each pore to be inspected can be obtained. Furthermore, it can be understood that the pass / fail of this spinneret can be inspected, eliminating the need for manual inspection, saving labor, and having higher inspection efficiency and accuracy.

[0077] In some embodiments, in step S403, determining the inspection result of the spinneret based on the state information of each pore to be inspected includes: determining the pore pass rate, which is the ratio of the number of pores to be inspected whose state information matches the preset state to the total number of pores to be inspected, based on the state information of each pore to be inspected; and determining the inspection result of the spinneret based on the pore pass rate and the pass rate threshold.

[0078] The preset state may be the standard state of the pore to be inspected, that is, the shape of the pore to be inspected without clogging or cracking. From the state information of each pore to be inspected, the number of pores to be inspected that match the preset state, that is, the number of pores to be inspected whose shape has not changed, can be determined. By dividing this number by the total number of pores to be inspected, the pass rate of the pores can be obtained.

[0079] The pass rate threshold can be set according to the actual situation, such as 1 or 0.98. Specifically, it can be set according to the situation. If the pass rate threshold is 1, it means that no shape change has occurred in all pores to be inspected, indicating that the spinneret is qualified. Otherwise, it means that the spinneret is a defective product and needs to be cleaned or replaced.

[0080] In some embodiments, method 400 further includes controlling a robotic arm in a spinneret inspection apparatus to move the spinneret to a first area where qualified spinnerets are placed when the inspection result meets a preset passing condition.

[0081] Referring to FIG. 1, a spinneret inspection apparatus may be provided with a robotic arm 150, and the robotic arm 150 may have a general structure that enables highly flexible movement.

[0082] When the inspection result meets the preset passing condition, the pore pass rate may be equal to or higher than a passing rate threshold. Under this condition, the robotic arm 150 can also transport the spinneret to a first area that can be a storage area for qualified spinnerets.

[0083] By transporting the spinneret with the robotic arm, it is possible to perform sorting of the spinnerets while realizing spinneret inspection, and further simplify labor costs.

[0084] In some embodiments, method 400 When the inspection result does not meet the preset passing condition, controlling the robotic arm in the spinneret inspection apparatus to move the spinneret to a second area where unqualified spinnerets are placed, and sending a first notification indicating that the spinneret is unqualified, where the first notification includes the spinneret number.

[0085] When the inspection result does not meet the preset passing condition, the robotic arm 150 can transport the spinneret to a second area that is a storage area for unqualified spinnerets, which facilitates the operator to timely clean that part.

[0086] Also, during the transportation to the second area, the first notification can be sent to the operator to notify the existence of unqualified spinnerets in the second area. Further, the first notification may include information such as the spinneret number, which can facilitate the management of the spinnerets.

[0087] In some embodiments, referring to FIG. 1, the spinneret inspection device includes a first image sensor 161 and a second image sensor 162 respectively disposed on both sides of the axis of the imaging member 130.

[0088] In step S401, obtaining the spinneret projection image on the imaging member includes controlling the first image sensor to capture the first projection image on the imaging member, and controlling the second image sensor to capture the second projection image on the imaging member, where the first projection image and the second projection image are the same spinneret projection image, and determining the spinneret projection image based on the first projection image and the second projection image.

[0089] In this embodiment, the spinneret inspection device may include two image sensors respectively disposed on the left and right sides of the imaging member 130, so that the projection image of the spinneret on the imaging member 130 can be captured from different angles.

[0090] Obtaining the spinneret projection image on the imaging member can be achieved by obtaining the first projection image on the imaging member with the first image sensor and obtaining the second projection image on the imaging member with the second image sensor. The first projection image and the second projection image are images of projections of the same spinneret at different angles.

[0091] And since the spinneret projection image can be obtained based on the first projection image and the second projection image, the accuracy of the spinneret projection image can be improved based on the projection images at different angles.

[0092] In some embodiments, determining the spinneret projection image based on the first projection image and the second projection image includes performing perspective correction on the first projection image to obtain a first corrected image, and performing perspective correction on the second projection image to obtain a second corrected image. Including obtaining a spinneret projection image by fusing a first corrected image and a second corrected image.

[0093] Since the first projection image and the second projection image are respectively located on both sides of the axis of the imaging member, and certain deformation may occur due to the viewing angle, it can be understood that in order to obtain the first corrected image and the second corrected image, the projection images are respectively processed by perspective correction first. The first corrected image and the second corrected image may have the same viewing angle as the orthographic projection image of the spinneret on the imaging member 130.

[0094] Next, the first corrected image and the second corrected image can be fused by, for example, overlapping the edges of the projections presented by both, which can improve the sharpness and accuracy of the spinneret projection image and help improve the accuracy of the inspection result.

[0095] FIG. 5 is a schematic block diagram showing a training device for an image recognition model according to an embodiment of the present disclosure. Referring to FIG. 5, a training device 500 for an image recognition model provided by an embodiment of the present disclosure includes A processing unit 501 that processes a spinneret sample image based on a first image recognition model to obtain label state information of each pore in the spinneret sample image. Here, the spinneret sample image is a projection image formed on the imaging member after the light emitted from the light source sequentially passes through each pore of the spinneret sample and is enlarged by the enlarging member, and the label state information of each pore is used to represent the shape state of the corresponding pore, the processing unit 501, A training unit 502 that trains a preset image recognition model based on the spinneret sample image and the label state information of each pore to obtain a target image recognition model. The target image recognition model is used to process the spinneret projection image to obtain the state information of each pore in the spinneret, the training unit 502.

[0096] In some embodiments, the processing unit 501 further Input the spinneret sample image into the first sub-model in the first image recognition model to obtain the first information set for each pore, where the first information set includes the first state information of each pore, the first position information of each pore, and the first confidence level corresponding to the first state information of each pore, and Input the spinneret sample image into the second sub-model in the first image recognition model to obtain the second information set for each pore, where the second information set includes the second state information of each pore, the second position information of each pore, and the second confidence level corresponding to the second state information of each pore, and It is used to determine the labeled state information of each pore based on the first information set and the second information set.

[0097] In some embodiments, the processing unit 501 further Determine the subset of state information for each pore based on the first position information in the first information set and the second position information in the second information set, where the subset of state information includes the first state information and the second state information of the same pore, and For the subset of state information of the first pore in each pore, when the first state information of the first pore and the second state information of the first pore do not match, determine the overall confidence level of the first pore based on the first confidence level corresponding to the first state information and the second confidence level corresponding to the second state information, and Determine the labeled state information of the first pore based on the overall confidence level of the first pore and the confidence threshold, and It is used to determine the labeled state information of each pore based on at least the labeled state information of the first pore.

[0098] In some embodiments, the training unit 502 further Process the spinneret sample image based on a preset image recognition model to obtain the predicted state information for each pore, and Determine the first loss function based on the predicted state information of each pore and the labeled state information of each pore, and It is to obtain a second loss function by adding a temperature scalar to the first loss function, where the temperature scalar balances the sensitivity of a preset image inspection model to differences between pores in different shape states, and It is used to adjust the parameters of a preset image recognition model based on the second loss function to obtain a target image recognition model.

[0099] In some embodiments, the training unit 502 further Based on the temperature scalar, to determine a dynamic temperature function for representing the change relationship of the temperature scalar with time, and It is used to obtain a second loss function by adding the dynamic temperature function to the first loss function. In some embodiments, the training unit 502 further It is used to determine the first loss function based on the predicted state information of each pore, the labeled state information of each pore, and the loss function of the first image recognition model.

[0100] FIG. 6 is a schematic block diagram showing a spinneret inspection device according to another embodiment of the present disclosure. Referring to FIG. 6, a spinneret inspection device 600 provided by an embodiment of the present disclosure includes a gripping member 110 for gripping a spinneret, a light source 120 and an imaging member 130 respectively disposed on both sides of the gripping member 110, and an enlarging member 140 disposed between the imaging member 130 and the gripping member 110. The light emitted from the light source 120 passes through each inspection target pore 210 of the spinneret 200 and is enlarged by the enlarging member 140, and then forms a spinneret projection image on the imaging member 130. The spinneret inspection device 600 is used in a spinneret inspection device including the enlarging member 140, An acquisition unit 601 for acquiring a spinneret projection image on the imaging member, A prediction unit for processing the spinneret projection image using a target image recognition model to obtain state information of each inspection target pore in the spinneret. The target image recognition model is trained based on any one of the training methods in the above embodiments. The prediction unit 602 It includes a determination unit 603 for determining the inspection result of the spinneret based on the state information of each pore to be inspected.

[0101] In some embodiments, the determination unit 603 is further used to control the robot arm in the spinneret inspection device to move the spinneret to the first area where the qualified spinnerets are placed when the inspection result meets the preset qualified conditions.

[0102] In some embodiments, the determination unit 603 is further used to control the robot arm in the spinneret inspection device to move the spinneret to the second area where the unqualified spinnerets are placed when the inspection result does not meet the preset qualified conditions, and used to send a first notification indicating that the spinneret is unqualified, where the first notification includes the number of the spinneret.

[0103] In some embodiments, the spinneret inspection device includes a first imaging element and a second imaging element respectively arranged on both sides of the axis of the imaging member. The acquisition unit 601 is further used to control the first imaging element to capture the first projection image on the imaging member, and used to control the second imaging element to capture the second projection image on the imaging member, where the first projection image and the second projection image are the same spinneret projection image, and used to determine the spinneret projection image based on the first projection image and the second projection image.

[0104] In some embodiments, the acquisition unit 601 is further used to perform perspective correction on the first projection image to obtain a first corrected image, and used to perform perspective correction on the second projection image to obtain a second corrected image, and used to fuse the first corrected image and the second corrected image to obtain the spinneret projection image.

[0105] In some embodiments, the determination unit 603 is further configured to determine a pore qualification rate, which is the ratio of the number of pores to be inspected whose state information matches a preset state to the total number of pores to be inspected, based on the state information of each pore to be inspected; and determine an inspection result of the spinneret based on the pore qualification rate and a qualification rate threshold.

[0106] For the specific functions and exemplary descriptions of each module and sub-module of the apparatus in the embodiments of the present disclosure, reference may be made to the related descriptions of the corresponding steps in the embodiments of the above method, which will not be repeated here.

[0107] The electronic device provided by the embodiments of the present disclosure includes at least one processor; and a memory communicatively connected to the at least one processor. The memory stores instructions executable by the at least one processor, and when the instructions are executed by the at least one processor, the instructions cause the at least one processor to execute any one of the methods in the embodiments of the present disclosure.

[0108] In the embodiments of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions for causing a computer to execute any one of the methods in the embodiments of the present disclosure is provided.

[0109] In the embodiments of the present disclosure, a program is provided, and when the program is executed by a processor, the program implements any one of the methods in the embodiments of the present disclosure.

[0110] FIG. 7 is a structural block diagram of an electronic device according to an embodiment of the present disclosure. As shown in FIG. 7, the electronic device includes a memory 710 and a processor 720, and a computer program executable by the processor 720 is stored in the memory 710. The number of the memory 710 and the processor 720 can be one or more. The memory 710 can store one or more computer programs, and when the one or more computer programs are executed by the electronic device, the electronic device is caused to execute the method provided by the above method embodiment. The electronic device can further include the following. A communication interface 730 is used to communicate with an external device and perform data interaction and transmission.

[0111] When the memory 710, the processor 720, and the communication interface 730 are independently implemented, the memory 710, the processor 720, and the communication interface 730 are connected to each other via a bus and can communicate with each other. The bus can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, an EISA (Extended Industry Standard Architecture) bus, or the like. The bus can be classified into an address bus, a data bus, a control bus, and the like. For ease of explanation, only a single thick line is shown in FIG. 7, but it does not indicate only a single bus or a single type of bus.

[0112] Optionally, in a specific implementation form, when the memory 710, the processor 720, and the communication interface 730 are integrated on one chip, the memory 710, the processor 720, and the communication interface 730 can communicate with each other via an internal interface.

[0113] The above-mentioned processor may be a Central Processing Unit (CPU), and it should be understood that it may also be other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field Programmable Gate Arrays (FPGAs) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware assemblies, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc. Note that the processor can be a processor that supports the Advanced RISC Machines (ARM) architecture.

[0114] Furthermore, optionally, the memory may include a read-only memory and a random access memory, and may further include a non-volatile random access memory. The memory can be either a volatile memory or a non-volatile memory, or may include both a volatile memory and a non-volatile memory. Here, the non-volatile memory can include ROM (Read-Only Memory), PROM (Programmable ROM), EPROM (Erasable PROM), EEPROM (Electrically EPROM), or flash memory. The volatile memory can include a random access memory (Random Access Memory, RAM) that functions as an external cache. By way of example and not limitation, many forms of RAM are available. For example, static random access memory (Static RAM, SRAM), dynamic random access memory (Dynamic Random Access Memory, DRAM), synchronous DRAM (Synchronous DRAM, SDRAM), double data rate SDRAM (Double Data Date SDRAM, DDR SDRAM), enhanced SDRAM (Enhanced SDRAM, ESDRAM), synchlink DRAM (Synchlink DRAM, SLDRAM), and direct RAMBUS RAM (Direct RAMBUS RAM, DR RAM).

[0115] In the above embodiments, it may be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented by software, all or part of it may be implemented in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer instructions are loaded and executed on a computer, the process or function according to the embodiments of the present disclosure is generated in whole or in part. The computer may be a general-purpose computer, a dedicated computer, a computer network, or other programmable devices. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wire (such as coaxial cable, optical fiber, Digital Subscriber Line (DSL)) or wireless (such as infrared, Bluetooth®, microwave, etc.). The computer-readable storage medium may be any available medium accessible by a computer, or a data storage device including a server, a data center, etc. integrated with one or more available media. The available medium may be a magnetic medium (such as a floppy (registered trademark) disk, a hard disk, a magnetic tape), an optical medium (such as a Digital Versatile Disc (DVD)), or a semiconductor medium (such as a Solid State Disk (SSD)). It should be noted that the computer-readable storage medium referred to in the present disclosure may be a non-volatile storage medium, in other words, a non-transitory storage medium.

[0116] Those skilled in the art can understand that all or some of the steps for implementing the above embodiments may be implemented by hardware, or may be implemented by instructing relevant hardware through a program, and the program may be stored in a computer-readable storage medium, and the above storage medium may be a read-only memory, a magnetic disk, an optical disk, etc.

[0117] In the description of the embodiments of the present disclosure, the descriptions of reference terms such as "one embodiment", "some embodiments", "example", "specific example", or "some examples" mean that the specific features, structures, materials, or features described in relation to the embodiment or example are included in at least one embodiment or example of the present disclosure. And the specific features, structures, materials, or features described can be combined in any one or more embodiments or examples in an appropriate manner. Furthermore, those skilled in the art may combine different embodiments or examples described in the present disclosure and the features of different embodiments or examples within a non-contradictory range.

[0118] In the description of the embodiments of the present disclosure, " / " represents the meaning of "or" unless otherwise described. For example, A / B may represent either A or B. The "and / or" in the present disclosure only explains the relationship of related objects and indicates that there may be three types of relationships. For example, A and / or B can indicate the following. There are three situations where A exists alone, A and B exist simultaneously, and B exists alone.

[0119] In the description of the embodiments of the present disclosure, the terms "first" and "second" are used only for the purpose of description and should not be construed as indicating or implying relative importance, nor should they be construed as implying the number of technical features shown. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of such features. In the description of the embodiments of the present disclosure, "a plurality" means two or more unless otherwise described.

[0120] The above are only exemplary embodiments of the present disclosure and do not limit the present disclosure. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and scope of the principles of the present disclosure should all be included within the protection scope of the present disclosure.

Claims

1. A method for training an image recognition model, comprising: Processing a spinneret sample image based on a first image recognition model to obtain label state information of each pore in the spinneret sample image, where the spinneret sample image is a projection image formed on an imaging member after light emitted from a light source passes through each pore of the spinneret sample and is magnified by a magnifying member, and the label state information of each pore is used to represent the shape state of each corresponding pore; and training a preset image recognition model based on the spinneret sample image and the label status information of each of the holes to obtain a target image recognition model, the target image recognition model being used to process the spinneret projection image to obtain status information of each of the holes in the spinneret. How to train an image recognition model.

2. processing a spinneret sample image based on the first image recognition model to obtain label state information for each hole in the spinneret sample image; inputting the spinneret sample image into a first sub-model in the first image recognition model to obtain a first information set for each of the holes, the first information set including first status information for each of the holes, first position information for each of the holes, and a first confidence level corresponding to the first status information for each of the holes; inputting the spinneret sample image into a second sub-model in the first image recognition model to obtain a second set of information for each of the holes, the second set of information including second state information for each of the holes, second position information for each of the holes, and a second confidence level corresponding to the second state information for each of the holes; determining label status information for each of the pores based on the first set of information and the second set of information. The method of training an image recognition model according to claim 1 .

3. determining label state information for each of the pores based on the first set of information and the second set of information, determining a state information subset for each of the pores based on first location information in the first information set and second location information in the second information set, the state information subset including first state information and second state information for the same pore; For a first pore state information subset in each of the pores, when the first state information of the first pore does not match the second state information of the first pore, determining an overall reliability of the first pore based on a first reliability corresponding to the first state information and a second reliability corresponding to the second state information; determining label status information of the first pore based on a total reliability of the first pore and a reliability threshold; determining label status information for each of the pores based on label status information for at least the first pore; The method for training an image recognition model according to claim 2.

4. Training a preset image recognition model based on the spinneret sample image and the label state information of each of the holes to obtain a target image recognition model; processing the spinneret sample image based on a pre-established image recognition model to obtain predicted state information for each hole; determining a first loss function based on the predicted state information of each of the pores and the label state information of each of the pores; adding a temperature scalar to the first loss function to obtain a second loss function, the temperature scalar balancing the sensitivity of a pre-defined image inspection model to differences between pores of different geometric states; and adjusting parameters of the preset image recognition model based on the second loss function to obtain the target image recognition model. The method of training an image recognition model according to claim 1 .

5. Adding a temperature scalar to the first loss function to obtain a second loss function, determining a dynamic temperature function based on a temperature scalar to represent a relationship of change of the temperature scalar over time; and adding the dynamic temperature function to the first loss function to obtain the second loss function. The method for training an image recognition model according to claim 4.

6. Determining a first loss function based on the predicted state information of each of the pores and the label state information of each of the pores, determining a first loss function based on predicted state information of each of the pores, label state information of each of the pores, and a loss function of the first image recognition model; The method for training an image recognition model according to claim 4.

7. A spinneret inspection method for use in a spinneret inspection device comprising: a gripping member for gripping a spinneret; a light source and an imaging member disposed on both sides of the gripping member, respectively; and a magnifying member disposed between the imaging member and the gripping member, wherein light emitted from the light source passes through each inspection target hole of the spinneret and is magnified by the magnifying member, and then forms a spinneret projection image on the imaging member, acquiring a spinneret projection image on said imaging member; Processing the spinneret projection image using a target image recognition model to obtain status information of each inspected orifice in the spinneret, the target image recognition model being trained based on the training method according to any one of claims 1 to 6; and and determining an inspection result of the spinneret based on the status information of each of the inspected holes. Spinneret inspection methods.

8. The spinneret inspection method includes: Further comprising controlling a robot arm in the spinneret inspection device to move the spinneret to a first area for placing a passing spinneret when the inspection result satisfies a preset passing condition; The method of claim 7 .

9. The spinneret inspection method includes: Controlling a robot arm in the spinneret inspection device to move the spinneret to a second area for placing rejected spinnerets when the inspection result does not satisfy a preset pass condition; sending a first notification indicating that the spinneret is rejected, the first notification including a number of the spinneret. The method of claim 7 .

10. The spinneret inspection device includes a first imaging element and a second imaging element disposed on either side of an axis of the imaging member, Acquiring a spinneret projection image on the imaging member comprises: controlling the first imaging element to capture a first projected image on the imaging member; controlling the second imaging element to capture a second projected image on the imaging member, the first projected image and the second projected image being the same spinneret projected image; determining the spinneret projection image based on the first projection image and the second projection image; The method of claim 7 .

11. Determining the spinneret projection image based on the first projection image and the second projection image includes: performing perspective correction on the first projected image to obtain a first corrected image; performing perspective correction on the second projected image to obtain a second corrected image; and fusing the first corrected image and the second corrected image to obtain the spinneret projection image. The method of claim 10.

12. Determining an inspection result of the spinneret based on the status information of each of the inspected holes, determining a pore pass rate based on the state information of each of the inspection target pores, the rate being the number of inspection target pores whose state information matches a preset state to the total number of inspection target pores; determining an inspection result of the spinneret based on the pore pass rate and a pass rate threshold; The method of claim 7 .

13. A device for training an image recognition model, comprising: a processing unit for processing a spinneret sample image based on a first image recognition model to obtain label state information of each pore in the spinneret sample image, where the spinneret sample image is a projection image formed on an imaging member after light emitted from a light source passes through each pore of the spinneret sample in sequence and is magnified by a magnifying member, and the label state information of each pore is used to represent the shape state of each corresponding pore; a training unit for training a preset image recognition model based on the spinneret sample image and the label state information of each of the holes to obtain a target image recognition model, the target image recognition model being used to process the spinneret projection image to obtain the state information of each of the holes in the spinneret; A training device for image recognition models.

14. A spinneret inspection device for use in a spinneret inspection apparatus, comprising: a gripping member for gripping a spinneret; a light source and an imaging member disposed on both sides of the gripping member, respectively; and a magnifying member disposed between the imaging member and the gripping member, wherein light emitted from the light source passes through each inspection target hole of the spinneret and is magnified by the magnifying member, and then forms a spinneret projection image on the imaging member, an acquisition unit for acquiring a spinneret projection image on the imaging member; a prediction unit for processing the spinneret projection image using a target image recognition model to obtain status information of each inspected orifice in the spinneret, the target image recognition model being trained according to the training method according to any one of claims 1 to 6; and and a determination unit for determining an inspection result of the spinneret based on the state information of each of the inspection target holes. Spinneret inspection device.

15. At least one processor; a memory in communication with the at least one processor; The memory stores instructions executable by the at least one processor, the instructions, when executed by the at least one processor, causing the at least one processor to perform the method of claim 1. Electronic devices.

16. A non-transitory computer readable storage medium for storing instructions that cause a computer to perform the method of claim 1.

17. A program for implementing the method of claim 1 when executed by a processor in a computer.

18. At least one processor; a memory in communication with the at least one processor; The memory stores instructions executable by the at least one processor, the instructions, when executed by the at least one processor, causing the at least one processor to perform the method of claim 7. Electronic devices.

19. A non-transitory computer readable storage medium for storing instructions that cause a computer to perform the method of claim 7.

20. A program for implementing the method according to claim 7 when executed by a processor in a computer.

Images