Defect detection method, device and equipment for ceramic module and storage medium

Abstract

The invention discloses a defect detection method, device and equipment for a ceramic module and a storage medium. The method comprises the following steps: acquiring an original module image corresponding to a target ceramic module to be subjected to defect detection; according to the original module image and a target colloid labeling model obtained through pre-training, a colloid region image corresponding to the target ceramic module is determined, the target colloid labeling model is obtained through training according to a U-shaped neural network model in advance, and the target colloid labeling model uses a convolution block attention mechanism module; and according to the module template image of the target ceramic module and the colloid region image, determining whether the target ceramic module has a colloid penetration defect. According to the technical scheme, whether the ceramic module has the colloid penetration problem or not can be accurately determined, and the reasoning process of the model is remarkably accelerated while high calculation precision is kept.

Description

Technical Field

[0001] The present invention relates to the technical field of image defect detection, and particularly to a method, device, equipment and storage medium for defect detection of a ceramic module. Background Art

[0002] A ceramic module is a functional module or component made of ceramic materials, which is widely used in the fields of electronics, communication, energy, medical treatment, etc. Due to its unique physical and chemical properties (such as high hardness, high temperature resistance, corrosion resistance, good insulation, etc.), ceramic materials play an important role in modular design.

[0003] The ceramic module mainly consists of a high-density main body fired from alumina powder and a low-density colloid loaded during the firing process. In the defect detection of the ceramic module, whether there is a penetration problem with the glass glue is the main problem in the defect detection of the ceramic module.

[0004] When using X-ray images for defect detection of ceramic modules, the energy obtained at each point of the ceramic module is different, and the imaging is also different. In the traditional defect detection process, the local average gray difference algorithm is usually used for detection, that is, the image after X imaging is smoothed by a mean window to obtain a mean image, and each pixel point of the original image is compared with the pixel point on the mean image, and the pixel points with gray scale values higher than a certain range are determined as the colloid area. However, when the mean window is large, some ceramic bodies with higher gray scale values will be identified as colloids; if the mean window is small, the difference between the mean image and the original image is not large, and no effective gray difference can be obtained. Therefore, there is an urgent need for a measurement method that is robust to uneven gray levels. Summary of the Invention

[0005] The present invention provides a method, device, equipment and storage medium for defect detection of a ceramic module, so as to accurately determine whether there is a penetration problem with the colloid in the ceramic module, and while maintaining a high calculation accuracy, significantly accelerating the inference process of the model.

[0006] According to one aspect of the present invention, a method for defect detection of a ceramic module is provided. The method includes:

[0007] Obtaining an original module image corresponding to a target ceramic module to be defect-detected;

[0008] Determining a colloid area image corresponding to the target ceramic module according to the original module image and a target colloid annotation model obtained by pre-training, wherein the target colloid annotation model is pre-trained according to a U-shaped neural network model, and the target colloid annotation model uses a convolutional block attention mechanism module;

[0009] Determine whether there is a colloid penetration defect in the target ceramic module according to the module template image and the colloid area image of the target ceramic module.

[0010] According to another aspect of the present invention, there is provided a defect detection device for a ceramic module. The device includes:

[0011] An original module image acquisition module, configured to acquire an original module image corresponding to a target ceramic module to be defect-detected;

[0012] A colloid area image annotation module, configured to determine a colloid area image corresponding to the target ceramic module according to the original module image and a target colloid annotation model obtained by pre-training, wherein the target colloid annotation model is pre-trained according to a U-shaped neural network model, and the target colloid annotation model uses a convolutional block attention mechanism module;

[0013] A colloid penetration defect determination module, configured to determine whether there is a colloid penetration defect in the target ceramic module according to the module template image and the colloid area image of the target ceramic module.

[0014] According to another aspect of the present invention, there is provided an electronic device, the electronic device includes:

[0015] At least one processor; and

[0016] A memory communicatively connected to the at least one processor; wherein,

[0017] The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the defect detection method of the ceramic module according to any embodiment of the present invention.

[0018] According to another aspect of the present invention, there is provided a computer-readable storage medium, the computer-readable storage medium stores computer instructions, and the computer instructions are used to implement the defect detection method of the ceramic module according to any embodiment of the present invention when executed by a processor.

[0019] In the technical solution of the embodiment of the present invention, an original module image corresponding to a target ceramic module to be defect-detected is obtained. According to the original module image and a target colloid annotation model obtained by pre-training, a colloid region image corresponding to the target ceramic module is determined, wherein the target colloid annotation model is pre-trained according to a U-shaped neural network model, and the target colloid annotation model uses a convolutional block attention mechanism module. Through the convolutional block attention mechanism module, the screening ability of network features for the colloid region is enhanced. Furthermore, according to the module template image of the target ceramic module and the colloid region image, it is possible to accurately determine whether there is a colloid penetration defect in the target ceramic module.

[0020] It should be understood that the content described in this part is not intended to identify the key or important features of the embodiments of the present invention, nor is it used to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0022] Figure 1 is a flowchart of a method for defect detection of a ceramic module according to Embodiment 1 of the present invention;

[0023] Figure 2 is the original module image of the target ceramic module under X-ray;

[0024] Figure 3 are the original module image, colloid region image, and module template image of the target ceramic module;

[0025] Figure 4 is a flowchart of a method for defect detection of a ceramic module according to Embodiment 2 of the present invention;

[0026] Figure 5 is a flowchart of a method for defect detection of a ceramic module according to Embodiment 3 of the present invention;

[0027] Figure 6 is a structural diagram of a device for defect detection of a ceramic module according to Embodiment 4 of the present invention;

[0028] Figure 7 is a schematic structural diagram of an electronic device for implementing the method for defect detection of a ceramic module in the embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0029] In order to enable those skilled in the art to better understand the solution of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0030] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily need to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments of the present invention described here can be implemented in an order other than those illustrated or described here. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0031] Embodiment 1

[0032] Figure 1 It is a flowchart of a method for detecting defects of a ceramic module provided in Embodiment 1 of the present invention. This embodiment is applicable to detecting whether there is a situation of colloid penetration in the ceramic module. This method can be executed by a defect detection device of the ceramic module. The defect detection device of the ceramic module can be implemented in the form of hardware and / or software, and the defect detection device of the ceramic module can be configured in an electronic device. As Figure 1 shown, the method includes:

[0033] S101. Obtain the original module image corresponding to the target ceramic module to be defect-detected.

[0034] Among them, the target ceramic module generally refers to a ceramic component or module that needs to be subjected to quality inspection during industrial production or manufacturing. The original module image can be an X-ray image of the target ceramic module.

[0035] The imaging process of X-ray is that a beam of conical X-energy rays is emitted by an X-ray source and transmitted to the target ceramic module after a certain distance. The energy received by the target ceramic module is converted into the final original module image through signal processing. During the transmission process, due to the different densities of the penetrated objects, the energy emitted by the X-ray is attenuated to varying degrees. Therefore, the energy obtained at each point on the target ceramic module is different, and the imaging also varies. Materials with higher density and greater thickness absorb more X-rays and appear as brighter areas in the image; materials with lower density absorb less X-rays and appear as darker areas in the image.

[0036] Figure 2 This is the original module image of the target ceramic module under X-ray. As Figure 2 shown, when actually using X-ray images for defect detection of the target ceramic module, after the X-ray source energy is absorbed by objects with different densities, the main body made of alumina powder in the ceramic module is a high-density material with low brightness, and the loaded colloid during the firing process is a low-density object with high brightness.

[0037] S102. Determine the colloid region image corresponding to the target ceramic module according to the original module image and the target colloid annotation model obtained by pre-training.

[0038] It should be noted that the target colloid annotation model is pre-trained according to the U-shaped neural network model, and the target colloid annotation model uses a convolutional block attention mechanism module.

[0039] The target colloid annotation model belongs to the U-shaped neural network model. In the field of artificial intelligence segmentation, what the U-shaped network has in common with other networks is that it continuously extracts shallow features and deep features of different dimensions through layer-by-layer convolution. At the same time, the U-shaped network uses skip connections to connect the shallow network with the deep features, enabling the network to obtain more feature information at the final segmentation layer. This is also the reason why the U-shaped network has significant effects.

[0040] The training process of the target colloid annotation model can be as follows: input the training sample image into a preset U-shaped neural network model for colloid region annotation, and determine the training error of the preset U-shaped neural network model according to the colloid annotation result output by the preset U-shaped neural network model and the standard colloid region. Then, adjust the network parameters in the preset U-shaped neural network model according to the training error. When the preset convergence condition is met, it is determined that the training of the preset U-shaped neural network model is completed, and the preset U-shaped neural network model with the training completed is determined as the target colloid annotation model.

[0041] The target colloid annotation model uses the Convolutional Block Attention Module (CBAM). The Convolutional Block Attention Module is an attention mechanism for Convolutional Neural Networks (CNNs), aiming to enhance the network's ability to extract important features. CBAM combines Channel Attention Module (CAM) and Spatial Attention Module (SAM) to adaptively adjust the weights of feature maps, thereby improving the performance of the model. Its main purpose is to suppress irrelevant feature information.

[0042] The channel attention mechanism operates as follows: Global Average Pooling (GAP) and Global Max Pooling (GMP) are respectively performed on the original module image to obtain two 1D vectors. These two vectors are respectively input into a shared Multi-Layer Perceptron (MLP) to generate their respective corresponding channel attention weights. The two weights are added and activated through the Sigmoid function to obtain the final channel attention map. The channel attention map is multiplied by the original module image to obtain the weighted initial feature map.

[0043] Channel attention mechanism formula:

[0044] Mc(F) = σ(MLP(GAP(F)) + MLP(GMP(F)));

[0045] Where F is the input feature map; GAP and GMP are Global Average Pooling and Global Max Pooling respectively; MLP is the Multi-Layer Perceptron; σ is the Sigmoid activation function.

[0046] The spatial attention mechanism operates as follows: Global Average Pooling and Global Max Pooling are performed on the initial feature map in the channel dimension to obtain two 2D feature maps. These two feature maps are concatenated and input into a convolutional layer to generate spatial attention weights. Through the Sigmoid function activation, the final spatial attention map is obtained. The spatial attention map is multiplied by the initial feature map to obtain the colloid region image corresponding to the target ceramic module.

[0047] Spatial attention mechanism formula:

[0048] Ms(F) = σ(f 7×7 ([GAP(F); GMP(F)]));

[0049] Where F is the input feature map; GAP and GMP are Global Average Pooling and Global Max Pooling respectively; f 7×7is a 7x7 convolutional layer; σ is the Sigmoid activation function.

[0050] Specifically, the original module image first passes through the channel attention module in the target colloid annotation model. The channel attention map output by the channel attention module is multiplied by the original module image to obtain a weighted initial feature map. The initial feature map then passes through the spatial attention module, and the spatial attention map output by the spatial attention module is multiplied by the initial feature map to obtain the colloid region image corresponding to the target ceramic module.

[0051] In segmentation networks such as U-Net and DeepLab, CBAM can improve the ability to capture detailed features. Through the channel and spatial attention mechanisms, CBAM can adaptively enhance important features and suppress irrelevant features. In the task of image classification and annotation, CBAM can significantly improve the accuracy and robustness of the model.

[0052] S103. Determine whether the target ceramic module has a colloid penetration defect based on the module template image and the colloid region image of the target ceramic module.

[0053] Among them, the module template image may refer to the standard glue injection area image of the target ceramic module.

[0054] Specifically, the module template image is compared with the colloid region image, so as to determine whether the target ceramic module has a colloid penetration defect, and the penetration defect level can also be determined according to the pre-established standard.

[0055] Figure 3 are the original module image, the colloid region image, and the module template image of the target ceramic module. As Figure 3 shown, after obtaining the colloid region image and the module template image, the module template image of the standard product component is used to match the colloid region image. The area within the module template image is considered to be the component colloid, and the remaining colloid is considered to be the redundant residual colloid. Finally, a connected component analysis is performed on the residual colloid. If the residual colloid causes the connection of the component region, it is considered that the target ceramic module has a penetration defect, otherwise it is considered that the target ceramic module has no penetration defect.

[0056] The technical solution of the embodiment of the present invention is to obtain the original module image corresponding to the target ceramic module to be defect-detected. According to the original module image and the pre-trained target colloid annotation model, the colloid region image corresponding to the target ceramic module is determined, wherein the target colloid annotation model is pre-trained according to the U-type neural network model, and the target colloid annotation model uses a convolutional block attention mechanism module. Through the convolutional block attention mechanism module, the screening ability of the network features for the colloid region is enhanced, and then according to the module template image of the target ceramic module and the colloid region image, it can be accurately determined whether the target ceramic module has a colloid penetration defect.

[0057] Embodiment 2

[0058] Figure 4 This is a flow chart of a defect detection method for a ceramic module provided by the second embodiment of the present invention. Based on the above embodiments, this embodiment introduces a target teacher-student learning model to reduce the computational complexity of the target colloid annotation model. Figure 4 As shown, the method includes:

[0059] S201, obtaining an original module image corresponding to a target ceramic module to be defect-detected.

[0060] S202: construct a target teacher-student learning model based on the target colloid labeling model.

[0061] Among them, the target teacher-student learning model includes a target teacher model and a target student model.

[0062] The target colloid annotation model is a U-type neural network model. Although the U-type neural network model can achieve high-precision pixel segmentation, it also increases the amount of calculation and parameters. In addition, there is also a problem that some convolution kernels extract similar features in the network feature extraction, which limits its application in the industrial field to a certain extent. In this regard, from the perspective of model quantization, it is necessary to propose a colloid segmentation network model based on a lightweight network.

[0063] In order to reduce the model inference time, the technical solution of the present invention accelerates the model by using knowledge distillation of the teacher-student learning framework. The teacher-student learning framework is a machine learning method that is usually used for model compression, knowledge transfer, and model optimization. The core idea is to guide the learning process of a simple, lightweight model (called the "student model") through a complex, high-performance model (called the "teacher model"). The student model achieves performance close to or even beyond the teacher model by learning the output or intermediate features of the teacher model, while having lower computational overhead and faster inference speed.

[0064] It should be noted that the target teacher-student learning model may refer to a model combination obtained by training based on the teacher-student learning framework, which includes a target teacher model and a target student model. Specifically, based on the target colloid annotation model, a target teacher model is constructed.

[0065] S203. Determine the colloid region image corresponding to the target ceramic module according to the original module image and the target student model.

[0066] In the technical solution of the embodiment of the present invention, the trained target student model can achieve the annotation ability of the target colloid annotation model, but the target student model requires less computing resources and has a faster inference speed. By replacing the target colloid annotation model with the target student model, the colloid region image corresponding to the target ceramic module can be determined faster.

[0067] Exemplarily, the determining the colloid region image corresponding to the target ceramic module according to the original module image and the target student model includes: inputting the original module image into the target student model for colloid annotation, and determining the colloid region image corresponding to the target ceramic module based on the output of the target student model.

[0068] S204. Determine whether the target ceramic module has a colloid penetration defect according to the module template image and the colloid region image of the target ceramic module.

[0069] In the technical solution of the embodiment of the present invention, by constructing a target teacher-student learning model according to the target colloid annotation model, the target teacher-student learning model includes a target teacher model and a target student model. According to the original module image and the target student model, the colloid region image corresponding to the target ceramic module is determined. By virtue of the characteristics of the target student model such as lightweight, simple structure and fewer parameters, the feature representation of the target colloid annotation model is imitated to annotate the colloid region image, and at the same time, high computing performance can be maintained.

[0070] Embodiment III

[0071] Figure 5 It is a flowchart of a method for detecting defects of a ceramic module provided in Embodiment III of the present invention. On the basis of the above embodiments, the process of constructing the target teacher-student learning model is further refined. As Figure 5 shown, the method includes:

[0072] S301. Obtain the original module image corresponding to the target ceramic module to be defect-detected.

[0073] S302. Determine the target colloid annotation model as the target teacher model in the teacher-student learning framework.

[0074] S303. Perform a reduction adjustment process on the target teacher model to obtain a student model framework.

[0075] Specifically, directly determine the target colloid annotation model as the target teacher model. Reduce and adjust the model structure in the target teacher model, and then a student model framework can be obtained.

[0076] Exemplarily, the step of performing a reduction adjustment process on the target teacher model to obtain a student model framework includes: based on a preset reduction ratio, perform a reduction process on the network layer of the target teacher model to obtain a reduced model framework; replace the standard convolutional layer in the reduced model framework with a depthwise separable convolutional layer to obtain a student model framework.

[0077] Among them, the preset reduction ratio can be set according to the actual situation. In the present invention, preferably, the preset reduction ratio is 50%.

[0078] The depthwise separable convolutional layer is a lightweight convolutional operation, which is widely used in deep learning models, especially on mobile and embedded devices, to reduce the number of model parameters and computational volume while maintaining high performance. It is an efficient alternative to the standard convolution and mainly consists of two parts: depthwise convolution (DW) and pointwise convolution (PW).

[0079] The depthwise convolution requires the depth of the convolutional kernel to be 1. When a convolutional kernel performs a convolution operation, it only operates on one channel of the input feature. At the same time, the pointwise convolution requires the size of the convolutional kernel to be 1*1, and its convolutional kernel depth is the same as the depth of the input feature matrix. The combination of the two makes the computational volume of the operation M*N become the level of M+N.

[0080] The specific computational reduction amount is as follows: Let D F represent the size of the input feature matrix, D K represent the size of the convolutional kernel, M represent the depth of the input feature matrix (the depth of the convolutional kernel), and N represent the depth of the output feature matrix (the number of convolutional kernels).

[0081] In the case where the stride is 1, the number of parameters of the standard convolution is: D k *D K *M*N;

[0082] The computational volume of the standard convolution is: D k *D k *M*MN*D F *D F ;

[0083] The number of parameters of the depthwise separable convolution is: Dk *D k *1*M + 1*1*M*N = D k *D k *M + M*N;

[0084] The computational cost of depthwise separable convolution is: D k *D k *1*M*D F *D F *1 + 1*1*M*D F *D F *N = D k *D k *M*D F *D F +M*N*D F *D F ;

[0085] From the comparison of the computational costs and the number of parameters of the two:

[0086] From the above formula, after using depthwise separable convolution, the theoretical number of parameters and the computational cost can both be reduced to In actual use, since the convolutional kernel size usually takes the value of 3, applying depthwise separable convolution can reduce it by about 88%.

[0087] Specifically, on the basis that the target teacher model has been determined, the student model framework of the target student model is to reduce the preset reduction ratio on the network layer of the target teacher model and use depthwise separable convolution to accelerate network inference. In addition, from the imaging diagram, there is an obvious boundary between the foreground and the background, and the network can reduce the high-dimensional feature extractor, so the 1024-layer convolutional kernel at the bottom is deleted, and thus the student model framework of the target student model can be obtained.

[0088] S304. Based on the training sample images, train the student model framework to obtain the target student model.

[0089] Specifically, train the student model framework according to the training sample images, and thus the target student model can be obtained.

[0090] Exemplarily, the training the student model framework based on the training sample images to obtain the target student model includes:

[0091] Determine the target annotation error of the student model framework according to the standard annotation image corresponding to the training sample images and the student model framework; adjust the network parameters in the student model framework according to the target annotation error; when the end training condition is reached, determine the trained student model framework as the target student model.

[0092] Among them, the target annotation error may refer to the annotation error of the student model framework during the training process.

[0093] Specifically, input the training sample image into the student model framework for colloid annotation, and the error between the output result of the student model framework and the standard annotation image can be determined as the target annotation error. Adjust the network parameters in the student model framework according to the target annotation error. When the training condition of the student model framework is reached (such as reaching the preset number of training times or the output marking result meets the expected requirements), the trained student model framework is determined as the target student model.

[0094] Exemplarily, determining the target annotation error of the student model framework according to the standard annotation image corresponding to the training sample image and the student model framework includes:

[0095] Input the training sample image into the target teacher model and the student model framework respectively for colloid annotation, and obtain the first annotation image output by the target teacher model and the second annotation image output by the student model framework;

[0096] Determine the first annotation error according to the first annotation image and the second annotation image;

[0097] Determine the second annotation error according to the standard annotation image corresponding to the training sample image and the second annotation image;

[0098] Determine the target annotation error according to the first annotation error, the second annotation error and the error ratio weight.

[0099] Among them, the first annotation image may refer to the annotation result output by the target teacher model, and the second annotation image may refer to the annotation result output by the student model framework. The first annotation error may refer to the error between the first annotation image and the second annotation image. The second annotation error may refer to the error between the standard annotation image and the second annotation image. The error ratio weight can be determined according to the actual situation. Preferably, the error ratio weight in the present invention can be 50%.

[0100] Specifically, after obtaining the first annotation error and the second annotation error, perform weighted summation on the first annotation error and the second annotation error according to the error ratio weight, and then the target annotation error can be obtained.

[0101] The technical solution of the embodiment of the present invention guides the learning of the target student model through the output of the target teacher model. The target student model not only learns the true label (i.e., the second annotation error), but also learns the annotation features of the target teacher model (i.e., the first annotation error), realizes knowledge transfer and model optimization, and reduces the dependence on a large amount of labeled data.

[0102] S305. Determine the colloid region image corresponding to the target ceramic module according to the original module image and the target student model.

[0103] S306. Determine whether there is a colloid penetration defect in the target ceramic module according to the module template image and the colloid region image of the target ceramic module.

[0104] In the technical solution of the embodiment of the present invention, by determining the target colloid annotation model as the target teacher model in the teacher-student learning framework, performing a reduction adjustment process on the target teacher model to obtain a student model framework, and training the student model framework based on training sample images to obtain a target student model, so that the student model can significantly reduce the number of parameters and the amount of calculation and improve the inference speed by reducing the network layer and using depthwise separable convolution.

[0105] Embodiment 4

[0106] Figure 6 It is a schematic structural diagram of a defect detection device for a ceramic module provided in Embodiment 4 of the present invention. As Figure 6 shown, the device includes:

[0107] An original module image acquisition module 401, configured to acquire an original module image corresponding to a target ceramic module to be defect-detected;

[0108] A colloid region image annotation module 402, configured to determine a colloid region image corresponding to the target ceramic module according to the original module image and a pre-trained target colloid annotation model, wherein the target colloid annotation model is pre-trained according to a U-shaped neural network model, and the target colloid annotation model uses a convolutional block attention mechanism module;

[0109] A colloid penetration defect determination module 403, configured to determine whether there is a colloid penetration defect in the target ceramic module according to the module template image and the colloid region image of the target ceramic module.

[0110] In the technical solution of the embodiment of the present invention, by acquiring an original module image corresponding to a target ceramic module to be defect-detected, determining a colloid region image corresponding to the target ceramic module according to the original module image and a pre-trained target colloid annotation model, wherein the target colloid annotation model is pre-trained according to a U-shaped neural network model, and the target colloid annotation model uses a convolutional block attention mechanism module. Through the convolutional block attention mechanism module, the screening ability of network features for the colloid region is enhanced, and thus, according to the module template image and the colloid region image of the target ceramic module, it can be accurately determined whether there is a colloid penetration defect in the target ceramic module.

[0111] Optionally, the colloid region image annotation module 402 includes:

[0112] A teacher-student learning model determination sub-module, configured to construct a target teacher-student learning model according to the target colloid annotation model, where the target teacher-student learning model includes a target teacher model and a target student model;

[0113] A colloid region image annotation sub-module, configured to determine a colloid region image corresponding to the target ceramic module according to the original module image and the target student model.

[0114] Optionally, the teacher-student learning model determination sub-module includes:

[0115] A teacher model determination unit, configured to determine the target colloid annotation model as the target teacher model in the teacher-student learning framework;

[0116] A student framework determination unit, configured to perform a reduction adjustment process on the target teacher model to obtain a student model framework;

[0117] A student model determination unit, configured to perform model training on the student model framework based on training sample images to obtain a target student model.

[0118] Optionally, the student framework determination unit is specifically configured to:

[0119] Perform a reduction process on the network layer of the target teacher model based on a preset reduction ratio to obtain a reduced model framework;

[0120] Replace the standard convolutional layer in the reduced model framework with a depthwise separable convolutional layer to obtain a student model framework.

[0121] Optionally, the student model determination unit includes:

[0122] An annotation error determination sub-unit, configured to determine a target annotation error of the student model framework according to the student model framework, the training sample images, and the standard annotation images corresponding to the training sample images;

[0123] A network parameter adjustment sub-unit, configured to adjust network parameters in the student model framework according to the target annotation error;

[0124] A student model determination sub-unit, configured to determine the trained student model framework as the target student model when the end training condition is reached.

[0125] Optionally, the annotation error determination sub-unit is specifically configured to:

[0126] Input the training sample images into the target teacher model and the student model framework respectively for colloid annotation, to obtain the first annotation image output by the target teacher model and the second annotation image output by the student model framework;

[0127] Determine the first annotation error according to the first annotation image and the second annotation image;

[0128] Determine the second annotation error according to the standard annotation image corresponding to the training sample image and the second annotation image;

[0129] Determine the target annotation error according to the first annotation error, the second annotation error and the error ratio weight.

[0130] Optionally, the colloid area image annotation sub-module is specifically used for:

[0131] Input the original module image into the target student model for colloid annotation, and determine the colloid area image corresponding to the target ceramic module based on the output of the target student model. The defect detection device for the ceramic module provided by the embodiments of the present invention can execute the defect detection method for the ceramic module provided by any embodiment of the present invention, and has the corresponding functional modules and beneficial effects for executing the method.

[0132] Embodiment Five

[0133] Figure 7 Fig. shows a schematic structural diagram of an electronic device 10 that can be used to implement the embodiments of the present invention. The electronic device is intended to represent various forms of digital computers, such as, laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as, personal digital processing, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present invention described and / or claimed herein.

[0134] As Figure 7As shown, the electronic device 10 includes at least one processor 11 and a memory communicatively connected to the at least one processor 11, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc. Among them, the memory stores a computer program executable by the at least one processor. The processor 11 can execute various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 into the random access memory (RAM) 13. In the RAM 13, various programs and data required for the operation of the electronic device 10 can also be stored. The processor 11, the ROM 12, and the RAM 13 are connected to each other through a bus 14. The input / output (I / O) interface 15 is also connected to the bus 14.

[0135] Multiple components in the electronic device 10 are connected to the I / O interface 15, including: an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a disk, an optical disc, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0136] The processor 11 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the processor 11 include but are not limited to a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The processor 11 executes the various methods and processes described above, such as the defect detection method of the ceramic module.

[0137] In some embodiments, the defect detection method of the ceramic module can be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as the storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or installed onto the electronic device 10 via the ROM 12 and / or the communication unit 19. When the computer program is loaded into the RAM 13 and executed by the processor 11, one or more steps of the defect detection method of the ceramic module described above can be executed. Alternatively, in other embodiments, the processor 11 can be configured to execute the defect detection method of the ceramic module in any other appropriate way (for example, by means of firmware).

[0138] The various embodiments of the systems and techniques described above in this specification can be implemented in digital electronic circuitry, integrated circuit systems, field programmable gate arrays (FPGA), application specific integrated circuits (ASIC), application specific standard products (ASSP), systems on chip (SOC), complex programmable logic devices (CPLD), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include: being implemented in one or more computer programs that are executable and / or interpretable on a programmable system including at least one programmable processor, which may be a special-purpose or general-purpose programmable processor that receives data and instructions from, and transmits data and instructions to, a storage system, at least one input device, and at least one output device.

[0139] The computer program for implementing the method of the present invention can be written in any combination of one or more programming languages. These computer programs can be provided to a processor of a general purpose computer, special purpose computer, or other programmable data processing apparatus, such that the computer programs, when executed by the processor, cause the functions / operations specified in the flowchart and / or block diagram to be implemented. The computer program can be executed entirely on the machine, partly on the machine, as a stand-alone software package partly on the machine and partly on a remote machine or entirely on the remote machine or server.

[0140] In the context of the present invention, a computer-readable storage medium can be a tangible medium that can contain or store a computer program for use by or in connection with an instruction execution system, apparatus, or device. The computer-readable storage medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. Alternatively, the computer-readable storage medium can be a machine-readable signal medium. More specific examples of the machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0141] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and a pointing device (e.g., a mouse or a trackball) through which the user can provide input to the electronic device. Other kinds of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0142] The systems and techniques described herein can be implemented in a computing system including backend components (e.g., as a data server), or a computing system including middleware components (e.g., an application server), or a computing system including frontend components (e.g., a user computer having a graphical user interface or a web browser through which the user can interact with an implementation of the systems and techniques described herein), or a computing system including any combination of such backend components, middleware components, or frontend components. The components of the system can be interconnected to each other by digital data communication in any form or medium (e.g., a communication network). Examples of communication networks include: local area network (LAN), wide area network (WAN), blockchain network, and the Internet.

[0143] The computing system can include a client and a server. The client and the server are generally far from each other and usually interact through a communication network. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or a cloud host, which is a host product in the cloud computing service system, solving the defects of difficult management and weak business scalability existing in traditional physical hosts and VPS services.

[0144] It should be understood that various forms of the processes shown above can be used, with steps reordered, added, or deleted. For example, the steps recited in the present invention can be executed in parallel, sequentially, or in a different order, as long as the desired results of the technical solution of the present invention can be achieved, and no limitation is made herein.

[0145] The above specific embodiments do not constitute a limitation on the protection scope of the present invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A defect detection method for a ceramic module, characterized in that: include: Acquire an original module image corresponding to a target ceramic module to be defect-detected; Determine the colloid region image corresponding to the target ceramic module according to the original module image and the pre-trained target colloid annotation model, wherein the target colloid annotation model is pre-trained according to a U-type neural network model, and the target colloid annotation model uses a convolutional block attention mechanism module; Whether a colloid penetration defect occurs in the target ceramic module is determined according to the module template image of the target ceramic module and the colloid region image.

2. The method according to claim 1, characterized in that: The method of determining the colloid region image corresponding to the target ceramic module according to the original module image and the pre-trained target colloid annotation model includes: According to the target colloid annotation model, a target teacher-student learning model is constructed, wherein the target teacher-student learning model includes a target teacher model and a target student model; According to the original module image and the target student model, a colloid region image corresponding to the target ceramic module is determined.

3. The method according to claim 2, characterized in that The method of constructing a target teacher-student learning model according to the target colloid labeling model includes: determining the target colloid annotation model as a target teacher model in a teacher-student learning framework; The target teacher model is reduced and adjusted to obtain a student model framework; Based on the training sample images, the student model framework is trained to obtain a target student model.

4. The method according to claim 3, characterized in that The target teacher model is reduced and adjusted to obtain a student model framework, including: Based on a preset reduction ratio, the network layer of the target teacher model is reduced to obtain a reduced model framework; The standard convolutional layer in the reduced model framework is replaced with a depth-wise separable convolutional layer to obtain a student model framework.

5. The method according to claim 3, characterized in that: The method of training the student model framework based on the training sample image to obtain a target student model includes: Determining a target annotation error of the student model framework according to the student model framework, the training sample image, and a standard annotation image corresponding to the training sample image; Adjusting network parameters in the student model framework according to the target annotation error; When the training end condition is reached, the trained student model framework is determined as the target student model.

6. The method according to claim 5, characterized in that The step of determining a target annotation error of the student model framework according to the standard annotation image corresponding to the training sample image and the student model framework includes: Inputting the training sample images into the target teacher model and the student model framework for colloid annotation respectively, obtaining a first annotated image output by the target teacher model and a second annotated image output by the student model framework; determining a first annotation error according to the first annotated image and the second annotated image; Determining a second annotation error according to the standard annotated image corresponding to the training sample image and the second annotated image; A target labeling error is determined according to the first labeling error, the second labeling error, and an error matching weight.

7. The method according to claim 2, characterized in that: Determining the colloid region image corresponding to the target ceramic module according to the original module image and the target student model includes: The original module image is input into the target student model for colloid annotation, and based on the output of the target student model, a colloid region image corresponding to the target ceramic module is determined.

8. A defect detection device for a ceramic module, characterized in that: include: An original module image acquisition module is used to acquire an original module image corresponding to a target ceramic module to be defect-detected; A colloid region image annotation module, used to determine the colloid region image corresponding to the target ceramic module according to the original module image and a pre-trained target colloid annotation model, wherein the target colloid annotation model is pre-trained according to a U-type neural network model, and the target colloid annotation model uses a convolutional block attention mechanism module; The colloid penetration defect determination module is used to determine whether the target ceramic module has a colloid penetration defect according to the module template image of the target ceramic module and the colloid region image.

9. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the defect detection method for a ceramic module according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the defect detection method for a ceramic module according to any one of claims 1 to 7 when executed.

Images