Self-adaptive dynamic convolution CT (Computed Tomography) flaw detection method, device, equipment and medium

Abstract

The invention relates to a self-adaptive dynamic convolution CT flaw detection method and device, equipment and a medium. The method comprises the following steps: acquiring a to-be-recognized CT image of an electronic component; preprocessing the to-be-identified CT image to obtain a basic CT image; performing feature extraction and feature classification processing on the basic CT image through a convolutional neural network model of adaptive dynamic convolution to obtain a classification result; and performing defect detection on the to-be-identified CT image based on the classification result to obtain a target detection result. The method and the device can adapt to the condition that the wound defects in the lithium battery are different in size and varied, and are beneficial to improving the accuracy of lithium battery internal defect detection.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to an adaptive dynamic convolution CT flaw detection method, device, equipment and medium. Background Art

[0002] CT flaw detection technology for lithium batteries is a method that uses computer tomography (CT) technology to detect internal defects in lithium batteries. By acquiring multiple slice images of lithium batteries, CT technology can non-destructively observe and analyze the internal structure and defects of lithium batteries, thereby evaluating their quality and performance. This technology can help detect problems such as bubbles, foreign matter, internal short circuits, and abnormal electrode structures in lithium batteries, discover potential safety hazards in advance, and guide quality control during the production process. At the same time, CT flaw detection technology can also provide reliable data support for the design and improvement of lithium batteries.

[0003] The artificial intelligence-based lithium battery CT flaw detection technology is a method that combines artificial intelligence technology with computer tomography technology to detect defects and evaluate the quality of lithium batteries. Its main feature is to use deep learning algorithms to automatically learn and extract the characteristics of internal defects of lithium batteries through training on a large amount of lithium battery CT image data to achieve defect detection. However, the trauma defects inside lithium batteries vary in size and are varied. The current convolutional neural network (CNN) model based on a fixed-size convolution kernel cannot effectively adapt to these defects, resulting in low accuracy in the detection of internal defects in lithium batteries. Summary of the invention

[0004] The purpose of the embodiments of the present application is to propose an adaptive dynamic convolution CT flaw detection method, device, equipment and medium to adapt to the situation where the trauma defects inside the lithium battery vary in size and are varied, and to improve the accuracy of internal defect detection in the lithium battery.

[0005] In order to solve the above technical problems, the embodiment of the present application provides a CT flaw detection method with adaptive dynamic convolution, comprising:

[0006] Acquire a CT image of the electronic component to be identified;

[0007] Preprocessing the CT image to be identified to obtain a basic CT image;

[0008] Performing feature extraction and feature classification processing on the basic CT image through a convolutional neural network model of adaptive dynamic convolution to obtain a classification result;

[0009] Based on the classification result, defect detection is performed on the CT image to be identified to obtain a target detection result.

[0010] In order to solve the above technical problems, the embodiment of the present application provides a CT flaw detection device with adaptive dynamic convolution, comprising:

[0011] A CT image acquisition unit to be identified, used for acquiring a CT image of the electronic component to be identified;

[0012] An image preprocessing unit, used for preprocessing the CT image to be identified to obtain a basic CT image;

[0013] A classification result generating unit, used for performing feature extraction and feature classification processing on the basic CT image through a convolutional neural network model of adaptive dynamic convolution to obtain a classification result;

[0014] The detection result generating unit is used to perform defect detection on the CT image to be identified based on the classification result to obtain a target detection result.

[0015] In order to solve the above technical problems, a technical solution adopted by the present invention is: to provide a computer device, including one or more processors; a memory for storing one or more programs, so that the one or more processors can implement any one of the above-mentioned adaptive dynamic convolution CT flaw detection methods.

[0016] In order to solve the above technical problems, a technical solution adopted by the present invention is: a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the CT flaw detection method of adaptive dynamic convolution described in any one of the above is implemented.

[0017] The embodiment of the present invention provides a method, device, equipment and medium for CT flaw detection with adaptive dynamic convolution. The method includes: obtaining a CT image to be identified of an electronic component; preprocessing the CT image to be identified to obtain a basic CT image; performing feature extraction and feature classification processing on the basic CT image through a convolutional neural network model with adaptive dynamic convolution to obtain a classification result; performing defect detection on the CT image to be identified based on the classification result to obtain a target detection result. The embodiment of the present invention performs feature extraction and feature classification processing on the basic CT image through a convolutional neural network model with adaptive dynamic convolution, and performs defect detection on the CT image to be identified based on the classification result. It can adapt to the situation where the trauma defects inside the lithium battery are of different sizes and changeable, which is conducive to improving the accuracy of internal defect detection of lithium batteries. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the scheme in the present application, a brief introduction is given below to the drawings required for use in the description of the embodiments of the present application. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0019] Figure 1 It is a flow chart of the implementation process of the adaptive dynamic convolution CT flaw detection method provided in the embodiment of the present application;

[0020] Figure 2 It is a flowchart for implementing a sub-process in the CT flaw detection method of adaptive dynamic convolution provided in an embodiment of the present application;

[0021] Figure 3 It is a flowchart for implementing a sub-process in the CT flaw detection method of adaptive dynamic convolution provided in an embodiment of the present application;

[0022] Figure 4 It is a flowchart for implementing a sub-process in the CT flaw detection method of adaptive dynamic convolution provided in an embodiment of the present application;

[0023] Figure 5 is a flow chart of the implementation of the adaptive dynamic convolution CT flaw detection method process provided by another embodiment of the present application;

[0024] Figure 6 It is a schematic diagram of a CT flaw detection device with adaptive dynamic convolution provided in an embodiment of the present application;

[0025] Figure 7 It is a schematic diagram of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0026] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by technicians in the technical field of the present application; the terms used in the specification of the application herein are only for the purpose of describing specific embodiments and are not intended to limit the present application; the terms "including" and "having" and any variations thereof in the specification and claims of the present application and the above-mentioned drawings are intended to cover non-exclusive inclusions. The terms "first", "second", etc. in the specification and claims of the present application or the above-mentioned drawings are used to distinguish different objects, not to describe a specific order.

[0027] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0028] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings.

[0029] The present invention is described in detail below with reference to the accompanying drawings and embodiments.

[0030] It should be noted that the adaptive dynamic convolution CT flaw detection method provided in the embodiment of the present application is generally executed by a computer device, and accordingly, the adaptive dynamic convolution CT flaw detection device is generally configured in a computer device.

[0031] See also Figure 1 , Figure 1 A specific implementation of the CT flaw detection method using adaptive dynamic convolution is shown.

[0032] It should be noted that if there are substantially the same results, the method of the present invention is not limited to Figure 1 The process sequence shown is limited to the following steps:

[0033] S1: Acquire a CT image of the electronic component to be identified.

[0034] The embodiment of the present application provides an adaptive dynamic convolution CT flaw detection method for detecting internal defects of lithium batteries. The method can adapt to the situation where the internal defects of lithium batteries vary in size and are varied, which is beneficial to improving the accuracy of internal defect detection of lithium batteries.

[0035] Specifically, the present application adopts CT flaw detection technology, which is a method of detecting internal defects of lithium batteries using computer tomography (CT) technology. CT technology can non-destructively observe and analyze the internal structure and defects of lithium batteries by acquiring multiple slice images of lithium batteries, thereby evaluating their quality and performance. Therefore, in the embodiment of the present application, it is necessary to obtain a CT image of the electronic component to be identified, and the electronic component can be a lithium battery.

[0036] S2: Preprocessing the CT image to be identified to obtain a basic CT image.

[0037] Specifically, in order to improve the accuracy of lithium battery defect detection, the embodiment of the present application needs to preprocess the CT image to be identified to obtain a basic CT image, wherein the preprocessing includes grayscale normalization, denoising, and image enhancement.

[0038] See also Figure 2 , Figure 2 A specific implementation of step S2 is shown, which is described in detail as follows:

[0039] S21: performing grayscale normalization processing on the CT image to be identified so as to adjust the grayscale value of the CT image to be identified to a uniform range, thereby obtaining a grayscale CT image.

[0040] Specifically, the grayscale values ​​of the CT images to be identified are adjusted to a uniform range. The grayscale values ​​of the CT images to be identified can be scaled to between 0 and 1 or between -1 and 1 to reduce the illumination and contrast differences between different images.

[0041] S22: Using a preset filtering technology to remove noise in the grayscale CT image to obtain a denoised CT image.

[0042] S23: performing image enhancement processing on the denoised CT image to obtain the basic CT image.

[0043] Specifically, Gaussian filtering or median filtering is used to remove noise from grayscale CT images, improve the signal-to-noise ratio of the image, and make the defect features more obvious. Then, through contrast adjustment, histogram equalization and other technologies, the contrast and visibility of the area of ​​interest in the image are enhanced, making the defects easier to detect.

[0044] S3: Perform feature extraction and feature classification processing on the basic CT image through a convolutional neural network model with adaptive dynamic convolution to obtain a classification result.

[0045] Specifically, the convolutional neural network model in the embodiment of the present application is a detection model built on the basis of the VNET convolutional neural network. Among them, VNET is a deep learning model for image segmentation tasks, which extracts image features through convolution and pooling operations, and restores resolution and retains detail information through upsampling and jump connections. VNET is a deep learning model for three-dimensional image segmentation tasks. It is an improved version based on the U-Net model, specially designed for processing three-dimensional volume data, such as CT images, MRI images, etc. The main feature of the VNET model is the introduction of jump connections and upsampling operations to retain high-resolution detail information. Through the combination of encoders and decoders, the VNET model can effectively segment three-dimensional images and extract semantic information while retaining detail information. It is widely used in the field of medical imaging, including tasks such as tumor segmentation, organ segmentation, and lesion detection. In CT flaw detection of lithium batteries, the VNET model can be used to accurately segment defective areas of lithium batteries to help quality assessment and safety detection.

[0046] Specifically, the existing lithium battery defect detection method is based on a convolutional neural network model with a fixed-size convolution kernel, which cannot adapt to the situation where the trauma defects inside the lithium battery are of different sizes and vary greatly, resulting in low defect detection accuracy. The embodiment of the present application uses a convolutional neural network model with adaptive dynamic convolution to perform feature extraction and feature classification processing on the basic CT image to detect lithium battery defects. In each convolution kernel, the embodiment of the present application uses adaptive dynamic convolution instead of traditional convolution. This mechanism adaptively selects the appropriate convolution kernel size and shape according to the size and shape of the input image. This can better capture defects of different sizes and shapes.

[0047] Furthermore, the present application also provides a specific implementation method: dynamically adjusting the convolution kernel size of the convolutional neural network model through a preset formula;

[0048] The preset formula is:

[0049]

[0050] Among them, w m，n is the kernel weight, x is the input feature, 2*d+1 is the size of the receptive field, A and B are adjustable parameters, j and k represent the horizontal and vertical coordinates of the pixel position of the image, m and n represent the horizontal and vertical coordinates of the weight position of the convolution window, and y j,k is the convolution kernel size.

[0051] See also Figure 3 , Figure 3 A specific implementation of step S3 is shown, which is described in detail as follows:

[0052] S31: Input the basic CT image into the convolutional neural network model of adaptive dynamic convolution.

[0053] S32: gradually extracting features of the basic CT image through the multi-layer convolutional layers and pooling layers of the encoder to obtain encoder output features.

[0054] Specifically, the encoder consists of multiple convolutional layers and pooling layers to gradually extract features of the input image. Each pooling layer reduces the resolution of the image and increases the receptive field of the feature. The main goal of the encoder is to compress and abstract the features of the input image to extract higher-level semantic information.

[0055] In the embodiment of the present application, the specific process of the encoder in the electronic component defect detection system is a process of gradual abstraction and dimensionality reduction. Through the combination of convolutional layers, activation functions and pooling layers, the encoder can convert the input image into a series of feature maps containing rich semantic information. These feature maps are then used in subsequent decoders or classifiers to achieve defect detection and location.

[0056] S33: Perform feature decoding based on the encoder output features through multiple upsampling layers and deconvolution layers of the decoder to generate an image segmentation result.

[0057] See also Figure 4 , Figure 4 A specific implementation of step S33 is shown, which is described in detail as follows:

[0058] S331: gradually increase the resolution of the encoder output features through the multiple upsampling layers and the deconvolution layers of the decoder to obtain basic high-resolution features.

[0059] S332: Fusing the high-resolution features in the encoder with the basic high-resolution features through a skip connection to obtain a fused feature.

[0060] S333: Generate a segmentation result with the same resolution as the basic CT image based on the fusion feature and the basic high-resolution feature to obtain the image segmentation result.

[0061] Specifically, the decoder consists of multiple upsampling layers and deconvolution layers to restore the feature map output by the encoder to the resolution of the original image. The upsampling operation can enlarge the resolution of the feature map, while the skip connection connects the feature map of the corresponding layer in the encoder with the feature map of the corresponding layer in the decoder to preserve detail information and edge information.

[0062] In an embodiment of the present application, the decoder gradually increases the resolution of the feature map through multiple upsampling layers. The upsampling operation can be implemented in many ways, such as deconvolution, bilinear interpolation or trilinear interpolation. In an embodiment of the present application, deconvolution is used to implement upsampling because it can optimize the upsampling process through learning. In each upsampling layer, the size of the feature map will gradually increase to approach the resolution of the original image. The deconvolution layer is a key part in the upsampling process. It increases the size of the feature map through transposed convolution kernels and appropriate padding. The deconvolution layer not only increases the resolution of the feature map, but also further extracts and refines the features through convolution operations. Each upsampling layer or deconvolution layer in the decoder may be connected to the corresponding layer in the encoder through a jump connection. The jump connection introduces the high-resolution feature map in the encoder into the decoding process, and concatenates or adds it with the basic high-resolution features to generate fused features, so that more detail information and edge information can be retained. Finally, a segmentation result with the same resolution as the basic CT image is generated based on the fusion features and the basic high-resolution features to obtain the image segmentation result.

[0063] S34: performing classification based on the image segmentation result by the classifier to obtain the classification result.

[0064] Specifically, a classifier is used to classify defects in the image segmentation results to obtain classification results. The defect classification results include bubbles, foreign matter, internal short circuits, and abnormal electrode structures in lithium batteries. Commonly used classifiers include support vector machines (SVM) and multi-layer perceptrons (MLP).

[0065] See also Figure 5 , Figure 5 A specific implementation method before step S3 is shown, which is described in detail as follows:

[0066] S3A: Acquire an initial CT image of the electronic component, and perform preprocessing, image annotation, and data division on the initial CT image to generate a training data set, a verification data set, and a test data set.

[0067] S3B: The training data set is input into the initial convolutional neural network model of adaptive dynamic convolution, and forward propagation is performed based on the training data set through the convolution layer, pooling layer and fully connected layer to obtain the prediction result of the model.

[0068] S3C: Calculate the loss value between the prediction result and the label in the verification data set through a preset loss function.

[0069] S3D: Perform back propagation based on the loss value, and update the model parameters of the initial convolutional neural network model by the gradient descent method to re-perform model iterative training until the loss value reaches a preset threshold or the number of iterations reaches a preset value, thereby obtaining the convolutional neural network model.

[0070] Specifically, before implementing step S3, the convolutional neural network model needs to be trained. In an embodiment of the present application, a large number of initial CT images need to be collected, and the initial CT images need to be preprocessed and image annotated. Among them, image annotation refers to annotating each initial CT image with its corresponding defect type and location information as a supervisory signal for model training. Then the annotated image is divided into data, which can be divided into a training set, a validation set and a test set, and the division ratio can be 7:2:1 or a similar ratio. Then the training data set is input into the initial convolutional neural network model of adaptive dynamic convolution, and forward propagation is performed based on the training data set through the convolution layer, the pooling layer and the fully connected layer to obtain the prediction result of the model; then the loss value between the prediction result and the label in the validation data set is calculated by the preset loss function. Among them, the preset loss function can be any one of the cross entropy loss and the mean square error loss. Finally, back propagation is performed based on the loss value, and the model parameters of the initial convolutional neural network model are updated by the gradient descent method to re-train the model iteration until the loss value reaches the preset threshold or the number of iterations reaches the preset value, and the convolutional neural network model is obtained. Among them, the preset threshold and the preset value are set according to actual conditions and are not limited here.

[0071] S4: Perform defect detection on the CT image to be identified based on the classification result to obtain a target detection result.

[0072] Furthermore, the present application also provides a specific implementation of step S4: using a threshold method or a bounding box regression technique to perform defect detection on the CT image to be identified based on the classification result to locate the defect position and quantify the defect degree to obtain the target detection result.

[0073] Specifically, the specific process of defect detection using the threshold method is as follows: according to the classification results and prior knowledge, a suitable threshold is set. This threshold can be global or local, depending on the characteristics of the defect and the background of the image. Then each pixel (or feature) in the image is compared with the set threshold. Then, based on the comparison results, the image is divided into defect areas and normal areas to locate the defects. Finally, the degree of the defect is quantified according to the threshold comparison, which can be quantified into mild defects, moderate defects and severe defects.

[0074] The specific process of defect detection using bounding box regression technology is as follows: Based on the classification results, the approximate location of the defect is preliminarily determined. This can be achieved by using methods such as sliding windows and region proposal networks (RPN) during the classification process. Then for each preliminarily located defect, a bounding box is predicted to more accurately represent its location. This bounding box is usually represented by four parameters (such as center coordinates, width, and height). Finally, the predicted bounding box is adjusted through the bounding box regression model to make it closer to the actual defect boundary, thereby achieving defect location. Finally, the degree of defect can be quantified based on the size of the bounding box.

[0075] In the embodiment of the present application, a CT image to be identified of an electronic component is obtained; the CT image to be identified is preprocessed to obtain a basic CT image; the basic CT image is subjected to feature extraction and feature classification processing by a convolutional neural network model of adaptive dynamic convolution to obtain a classification result; defect detection is performed on the CT image to be identified based on the classification result to obtain a target detection result. The embodiment of the present invention extracts features and performs feature classification processing on the basic CT image by a convolutional neural network model of adaptive dynamic convolution, and performs defect detection on the CT image to be identified based on the classification result, which can adapt to the situation where the trauma defects inside the lithium battery are of different sizes and changeable, and is conducive to improving the accuracy of internal defect detection of lithium batteries.

[0076] Please refer to Figure 6 , as a response to the above Figure 1 The present application provides an embodiment of a CT flaw detection device with adaptive dynamic convolution, and the device embodiment is similar to Figure 1 Corresponding to the method embodiment shown, the apparatus can be specifically applied to various computer devices.

[0077] like Figure 6 As shown, the adaptive dynamic convolution CT flaw detection device of this embodiment includes: a CT image acquisition unit 51 to be identified, an image preprocessing unit 52, a classification result generation unit 53 and a detection result generation unit 54, wherein:

[0078] The to-be-identified CT image acquisition unit 51 is used to acquire the to-be-identified CT image of the electronic component;

[0079] An image preprocessing unit 52, used for preprocessing the CT image to be identified to obtain a basic CT image;

[0080] A classification result generating unit 53 is used to perform feature extraction and feature classification processing on the basic CT image through a convolutional neural network model of adaptive dynamic convolution to obtain a classification result;

[0081] The detection result generating unit 54 is used to perform defect detection on the CT image to be identified based on the classification result to obtain a target detection result.

[0082] Furthermore, the convolutional neural network model includes an encoder, a decoder and a classifier, and the classification result generating unit 53 includes:

[0083] An image input unit, used for inputting the basic CT image into the convolutional neural network model of adaptive dynamic convolution;

[0084] A feature extraction unit, used for gradually extracting features of the basic CT image through multiple convolutional layers and pooling layers of the encoder to obtain encoder output features;

[0085] A feature decoding unit, configured to perform feature decoding based on the encoder output features through a plurality of upsampling layers and a deconvolution layer of the decoder to generate an image segmentation result;

[0086] A classification unit is used to perform classification based on the image segmentation result through the classifier to obtain the classification result.

[0087] Furthermore, the feature decoding unit includes:

[0088] A resolution increasing unit, configured to gradually increase the resolution of the encoder output feature through a plurality of the upsampling layers and the deconvolution layers of the decoder to obtain a basic high-resolution feature;

[0089] A feature fusion unit, used to fuse the high-resolution features in the encoder with the basic high-resolution features through a skip connection to obtain a fused feature;

[0090] A segmentation result generating unit is used to generate a segmentation result with the same resolution as the basic CT image based on the fusion feature and the basic high-resolution feature to obtain the image segmentation result.

[0091] Furthermore, the adaptive dynamic convolution CT flaw detection device also includes:

[0092] A convolution kernel adjustment unit is used to dynamically adjust the convolution kernel size of the convolutional neural network model through a preset formula.

[0093] The preset formula is:

[0094]

[0095] Among them, w m，nis the kernel weight, x is the input feature, 2*d+1 is the size of the receptive field, A and B are adjustable parameters, j and k represent the horizontal and vertical coordinates of the pixel position of the image, m and n represent the horizontal and vertical coordinates of the weight position of the convolution window, and y j,k is the convolution kernel size.

[0096] Furthermore, before the classification result generating unit 53, the following steps are also included:

[0097] A data partitioning unit, used for acquiring an initial CT image of the electronic component, and performing preprocessing, image annotation and data partitioning on the initial CT image to generate a training data set, a verification data set and a test data set;

[0098] A prediction result generating unit, used for inputting the training data set into an initial convolutional neural network model of adaptive dynamic convolution, performing forward propagation based on the training data set through a convolutional layer, a pooling layer and a fully connected layer to obtain a prediction result of the model;

[0099] A loss calculation unit, used to calculate the loss value between the prediction result and the label in the verification data set by using a preset loss function;

[0100] The training completion unit is used to perform back propagation based on the loss value, update the model parameters of the initial convolutional neural network model by the gradient descent method, and re-perform model iterative training until the loss value reaches a preset threshold or the number of iterations reaches a preset value, thereby obtaining the convolutional neural network model.

[0101] Furthermore, the image preprocessing unit 52 includes:

[0102] A grayscale normalization unit, configured to perform grayscale normalization processing on the CT image to be identified so as to adjust the grayscale value of the CT image to be identified to a uniform range, thereby obtaining a grayscale CT image;

[0103] A noise removal unit, used to remove noise in the grayscale CT image by using a preset filtering technology to obtain a denoised CT image;

[0104] The image enhancement unit is used to perform image enhancement processing on the denoised CT image to obtain the basic CT image.

[0105] Furthermore, the detection result generating unit 54 includes:

[0106] The defect detection unit is used to perform defect detection on the CT image to be identified based on the classification result by using a threshold method or a bounding box regression technology to locate the position of the defect and quantify the degree of the defect to obtain the target detection result.

[0107] In the embodiment of the present application, a CT image to be identified of an electronic component is obtained; the CT image to be identified is preprocessed to obtain a basic CT image; the basic CT image is subjected to feature extraction and feature classification processing by a convolutional neural network model of adaptive dynamic convolution to obtain a classification result; defect detection is performed on the CT image to be identified based on the classification result to obtain a target detection result. The embodiment of the present invention extracts features and performs feature classification processing on the basic CT image by a convolutional neural network model of adaptive dynamic convolution, and performs defect detection on the CT image to be identified based on the classification result, which can adapt to the situation where the trauma defects inside the lithium battery are of different sizes and changeable, and is conducive to improving the accuracy of internal defect detection of lithium batteries.

[0108] To solve the above technical problems, the present application also provides a computer device. Figure 7 , Figure 7 This is a basic structural block diagram of the computer device in this embodiment.

[0109] The computer device 6 includes a memory 61, a processor 62, and a network interface 63 that are interconnected through a system bus. It should be noted that the figure only shows a computer device 6 having three components, a memory 61, a processor 62, and a network interface 63, but it should be understood that it is not required to implement all the components shown, and more or fewer components can be implemented instead. Among them, those skilled in the art can understand that the computer device here is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to a microprocessor, an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA), a digital processor (DSP), an embedded device, etc.

[0110] Computer devices can be computing devices such as desktop computers, notebooks, PDAs, and cloud servers. Computer devices can interact with users through keyboards, mice, remote controls, touch pads, or voice control devices.

[0111] The memory 61 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX memory, etc.), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, disk, optical disk, etc. In some embodiments, the memory 61 can be an internal storage unit of the computer device 6, such as a hard disk or memory of the computer device 6. In other embodiments, the memory 61 can also be an external storage device of the computer device 6, such as a plug-in hard disk equipped on the computer device 6, a smart memory card (SmartMedia Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. Of course, the memory 61 can also include both the internal storage unit of the computer device 6 and its external storage device. In this embodiment, the memory 61 is generally used to store the operating system and various application software installed on the computer device 6, such as the program code of the CT flaw detection method of adaptive dynamic convolution, etc. In addition, the memory 61 can also be used to temporarily store various types of data that have been output or are to be output.

[0112] In some embodiments, the processor 62 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chips. The processor 62 is generally used to control the overall operation of the computer device 6. In this embodiment, the processor 62 is used to run the program code stored in the memory 61 or process data, such as running the program code of the above-mentioned adaptive dynamic convolution CT flaw detection method to implement various embodiments of the adaptive dynamic convolution CT flaw detection method.

[0113] The network interface 63 may include a wireless network interface or a wired network interface, and the network interface 63 is generally used to establish a communication connection between the computer device 6 and other electronic devices.

[0114] The present application also provides another implementation, namely, providing a computer-readable storage medium, which stores a computer program, and the computer program can be executed by at least one processor to enable the at least one processor to perform the steps of the above-mentioned adaptive dynamic convolution CT flaw detection method.

[0115] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, a disk, or an optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to execute the methods of each embodiment of the present application.

[0116] Obviously, the embodiments described above are only some embodiments of the present application, rather than all embodiments. The preferred embodiments of the present application are given in the accompanying drawings, but they do not limit the patent scope of the present application. The present application can be implemented in many different forms. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present application more thorough and comprehensive. Although the present application is described in detail with reference to the aforementioned embodiments, for those skilled in the art, it is still possible to modify the technical solutions recorded in the aforementioned specific implementation methods, or to replace some of the technical features therein with equivalents. Any equivalent structure made using the contents of the specification and drawings of this application, directly or indirectly used in other related technical fields, is similarly within the scope of patent protection of this application.

Claims

1. A CT flaw detection method based on adaptive dynamic convolution, characterized in that: include: Acquire a CT image of the electronic component to be identified; Preprocessing the CT image to be identified to obtain a basic CT image; Performing feature extraction and feature classification processing on the basic CT image through a convolutional neural network model of adaptive dynamic convolution to obtain a classification result; Based on the classification result, defect detection is performed on the CT image to be identified to obtain a target detection result.

2. The CT flaw detection method of adaptive dynamic convolution according to claim 1 is characterized in that: The convolutional neural network model includes an encoder, a decoder and a classifier. The convolutional neural network model using adaptive dynamic convolution performs feature extraction and feature classification processing on the basic CT image to obtain a classification result, including: Inputting the basic CT image into the convolutional neural network model of adaptive dynamic convolution; Stepwise extracting features of the basic CT image through multiple convolutional layers and pooling layers of the encoder to obtain encoder output features; Performing feature decoding based on the encoder output features through multiple upsampling layers and deconvolution layers of the decoder to generate an image segmentation result; The classification result is obtained by performing classification based on the image segmentation result through the classifier.

3. The CT flaw detection method of adaptive dynamic convolution according to claim 2 is characterized in that: The step of performing feature decoding based on the encoder output features through the multiple upsampling layers and deconvolution layers of the decoder to generate an image segmentation result includes: Stepwise increasing the resolution of the encoder output features through the multiple upsampling layers and the deconvolution layers of the decoder to obtain basic high-resolution features; The high-resolution features in the encoder are fused with the basic high-resolution features through a skip connection to obtain a fused feature; A segmentation result having the same resolution as the basic CT image is generated based on the fusion feature and the basic high-resolution feature to obtain the image segmentation result.

4. The CT flaw detection method of adaptive dynamic convolution according to claim 1 is characterized in that: The method further comprises: Dynamically adjusting the convolution kernel size of the convolutional neural network model through a preset formula; The preset formula is: Among them, w m，n is the kernel weight, x is the input feature, 2*d+1 is the size of the receptive field, A and B are adjustable parameters, j and k represent the horizontal and vertical coordinates of the pixel position of the image, m and n represent the horizontal and vertical coordinates of the weight position of the convolution window, and y j,k is the convolution kernel size.

5. The CT flaw detection method of adaptive dynamic convolution according to claim 1, characterized in that: Before performing feature extraction and feature classification processing on the basic CT image by the convolutional neural network model using adaptive dynamic convolution to obtain the classification result, the method further includes: Acquire an initial CT image of the electronic component, and perform preprocessing, image annotation, and data division on the initial CT image to generate a training data set, a verification data set, and a test data set; Inputting the training data set into the initial convolutional neural network model of adaptive dynamic convolution, performing forward propagation based on the training data set through the convolution layer, the pooling layer and the fully connected layer to obtain the prediction result of the model; Calculate the loss value between the prediction result and the label in the verification data set by using a preset loss function; Back propagation is performed based on the loss value, and the model parameters of the initial convolutional neural network model are updated by the gradient descent method to re-perform model iterative training until the loss value reaches a preset threshold or the number of iterations reaches a preset value, thereby obtaining the convolutional neural network model.

6. The CT flaw detection method of adaptive dynamic convolution according to claim 1, characterized in that: The preprocessing of the to-be-identified CT image to obtain a basic CT image includes: By performing grayscale normalization processing on the CT image to be identified, the grayscale value of the CT image to be identified is adjusted to a uniform range, thereby obtaining a grayscale CT image; Using a preset filtering technology to remove noise in the grayscale CT image to obtain a denoised CT image; The denoised CT image is subjected to image enhancement processing to obtain the basic CT image.

7. The CT flaw detection method of adaptive dynamic convolution according to any one of claims 1 to 6, characterized in that: The performing defect detection on the to-be-identified CT image based on the classification result to obtain a target detection result includes: A threshold method or a bounding box regression technique is used to perform defect detection on the CT image to be identified based on the classification result to locate the position of the defect and quantify the degree of the defect, thereby obtaining the target detection result.

8. An adaptive dynamic convolution CT flaw detection device, characterized in that: include: A CT image acquisition unit to be identified, used for acquiring a CT image of the electronic component to be identified; An image preprocessing unit, used for preprocessing the CT image to be identified to obtain a basic CT image; A classification result generating unit, used for performing feature extraction and feature classification processing on the basic CT image through a convolutional neural network model of adaptive dynamic convolution to obtain a classification result; The detection result generating unit is used to perform defect detection on the CT image to be identified based on the classification result to obtain a target detection result.

9. A computer device, characterized in that: The method comprises a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the CT flaw detection method with adaptive dynamic convolution as described in any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the CT flaw detection method using adaptive dynamic convolution as described in any one of claims 1 to 7 is implemented.

Images