Tire X-ray detection method and system based on improved generative adversarial network

Abstract

An embodiment of the present invention provides a tire X-ray inspection method and system based on an improved generative adversarial network, belonging to the field of image recognition technology. The method comprises: acquiring a tire image to be inspected; performing at least two first feature extraction operations and one second feature extraction operation on the tire image to obtain a predicted image of the tire; calculating a loss function between the predicted image and the tire image; determining whether the loss function is greater than or equal to a preset threshold; if the loss function is determined to be less than the threshold, sequentially performing five rounds of depthwise separable convolution operations, batch normalization operations, and activation operations on the predicted image and the tire image to obtain a sixth feature map; calculating a target function based on the feature map output by the fourth round of depthwise separable convolution operations, batch normalization operations, and activation operations and the sixth feature map output by the fifth round; calculating an anomaly score for the tire based on the target function; and determining whether the tire has an anomaly based on the anomaly score.

Description

Technical Field

[0001] The present invention relates to the technical field of image recognition, and in particular to a tire X-ray detection method and system based on an improved generative adversarial network. Background Art

[0002] Tires play an irreplaceable role in transportation. Based on their structure, they can be divided into radial tires and bias tires. Radial tires offer numerous advantages, such as safety, durability, environmental friendliness, comfort, and superior driving performance, earning widespread recognition from users. However, the complex production process for radial tires requires high precision, making quality issues inevitable. These issues can impact the tire's service life and even threaten the user's life. Therefore, each tire undergoes nondestructive testing before leaving the factory.

[0003] Currently, tire defect detection typically involves using an X-ray inspection machine to image the tire. Workers then observe the resulting X-ray images and identify and classify defects based on the shape and grayscale characteristics of the defects. However, this approach has numerous drawbacks: manual visual inspection is subjective, and results are easily influenced by the worker's dedication and knowledge. Furthermore, workers are prone to errors and missed detections due to the heavy workload and intensity of their daily work. While some existing image recognition network models have been developed, these models are not effective in achieving good recognition results. Summary of the Invention

[0004] The purpose of the embodiments of the present invention is to provide a tire X-ray detection method and system based on an improved adversarial generative network, which can improve the recognition rate of tire X-ray images.

[0005] To achieve the above objectives, an embodiment of the present invention provides a tire X-ray detection method based on an improved generative adversarial network, comprising:

[0006] Acquire a tire image to be inspected;

[0007] performing at least two first feature extraction operations and one second feature extraction operation on the tire image to obtain a predicted image of the tire;

[0008] Calculating a loss function of the predicted image and the tire image;

[0009] Determine whether the loss function is greater than or equal to a preset threshold;

[0010] If it is determined that the loss function is less than the threshold, sequentially performing five rounds of depthwise separable convolution operations, batch normalization operations, and activation operations on the predicted image and the tire image to obtain a sixth feature map;

[0011] Calculate the objective function based on the feature maps output by the fourth round of depth-wise separable convolution, batch normalization, and activation operations, and the sixth feature map output by the fifth round;

[0012] Calculating an abnormality score of the tire according to the objective function;

[0013] It is determined whether the tire has an abnormality according to the abnormality score.

[0014] Optionally, the first feature extraction operation includes:

[0015] performing two rounds of depthwise separable convolution operations, batch normalization operations, and activation operations on the tire image to obtain a first feature map;

[0016] Performing a hybrid attention mechanism operation on the first feature map to obtain a second feature map;

[0017] Performing a skip connection operation on the second feature map to obtain a third feature map;

[0018] performing an upsampling operation on a sixth feature map output by the next first feature extraction operation to obtain a fourth feature map;

[0019] Superimposing the third characteristic map and the fourth characteristic map to obtain a fifth characteristic map;

[0020] A depthwise separable convolution operation, a batch normalization operation, and an activation operation are performed on the fifth feature map to obtain a sixth feature map.

[0021] A convolution operation is performed on the fifth feature map to obtain a predicted image.

[0022] Optionally, the non-first first feature extraction operation includes:

[0023] performing a maximum pooling operation on the first feature map of the previous first feature extraction operation to obtain a tire image for this first feature extraction operation;

[0024] performing two rounds of depthwise separable convolution operations, batch normalization operations, and activation operations on the tire image to obtain a first feature map;

[0025] Performing a hybrid attention mechanism operation on the first feature map to obtain a second feature map;

[0026] Performing a skip connection operation on the second feature map to obtain a third feature map;

[0027] performing an upsampling operation on a sixth feature map output by the next first feature extraction operation to obtain a fourth feature map;

[0028] Superimposing the third characteristic map and the fourth characteristic map to obtain a fifth characteristic map;

[0029] A depthwise separable convolution operation, a batch normalization operation, and an activation operation are performed on the fifth feature map to obtain a sixth feature map.

[0030] Optionally, the last first feature extraction operation includes:

[0031] Performing two rounds of depth-wise separable convolution operations, batch normalization operations, and activation operations on the first feature map of the previous first feature extraction operation to obtain the current first feature map;

[0032] performing two rounds of depthwise separable convolution operations, batch normalization operations, and activation operations on the tire image to obtain a first feature map;

[0033] Performing a hybrid attention mechanism operation on the first feature map to obtain a second feature map;

[0034] Performing a skip connection operation on the second feature map to obtain a third feature map;

[0035] performing an upsampling operation on the second feature map output by the next first feature extraction operation to obtain a fourth feature map;

[0036] Superimposing the third characteristic map and the fourth characteristic map to obtain a fifth characteristic map;

[0037] A depthwise separable convolution operation, a batch normalization operation, and an activation operation are performed on the fifth feature map to obtain a sixth feature map.

[0038] Optionally, the second feature extraction operation includes:

[0039] performing a maximum pooling operation on the first feature map of the previous first feature extraction operation to obtain the current tire image;

[0040] performing two rounds of depthwise separable convolution operations, batch normalization operations, and activation operations on the tire image to obtain a first feature map;

[0041] Perform a hybrid attention mechanism operation on the first feature map to obtain a second feature map.

[0042] Optionally, calculating the loss function of the predicted image and the tire image includes:

[0043] According to formula (1), the loss function Lcon is calculated.

[0044]

[0045] in, is the Manhattan distance, x is the tire image, and G(x) is the predicted image.

[0046] Optionally, the objective function is calculated based on the feature maps output by the fourth round of depthwise separable convolution operations, batch normalization operations, and activation operations, and the sixth feature map output by the fifth round:

[0047] The objective function is calculated according to formula (2) to formula (4),

[0048]

[0049] , (3)

[0050] L=λ con L con +λ lat L lat +λ adv L adv , (4)

[0051] Among them, L con 、L lat 、L adv are loss functions, is the Manhattan distance, D(x) is the potential feature of the tire image x, D(G(x)) is the potential feature of the predicted image G(x), λ con ,λ lat ,λ adv They are loss functions L con 、L lat 、L adv The weight of , L is the objective function.

[0052] Optionally, calculating the tire abnormality score according to the objective function includes:

[0053] The anomaly score is calculated according to formula (5),

[0054] A(x)=λR(x)+(1-λ)L(x), (5)

[0055] Wherein, A(x) is the anomaly score, R(x) is the difference between the test image and the corresponding predicted image, L(x) is the difference between the latent representation of the tire image and the latent representation of the predicted image, and λ is a weighting parameter.

[0056] Optionally, calculating the tire abnormality score according to the objective function includes:

[0057]

[0058] in, is the normalized result of the anomaly score, min(A) is the minimum value of the anomaly score, and max(A) is the maximum value of the anomaly score.

[0059] On the other hand, the present invention further provides a tire X-ray inspection system based on an improved generative adversarial network, wherein the system includes a processor configured to execute any of the above methods.

[0060] Through the above technical solution, the tire X-ray detection method and system based on the improved adversarial generative network provided by the present invention achieves high-precision detection of key areas by designing a adversarial generative network model compared to conventional networks in the prior art because the network introduces a hybrid attention mechanism and depthwise separable convolution, thereby improving the detection accuracy of X-ray images.

[0061] Other features and advantages of the embodiments of the present invention will be described in detail in the subsequent detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] The accompanying drawings are used to provide a further understanding of the embodiments of the present invention and constitute a part of the specification. Together with the following detailed description, they are used to explain the embodiments of the present invention, but do not constitute a limitation of the embodiments of the present invention. In the accompanying drawings:

[0063] Figure 1 is a flowchart of a tire X-ray detection method based on an improved generative adversarial network according to one embodiment of the present invention;

[0064] Figure 2 is a structural block diagram of a network structure corresponding to a tire X-ray detection method based on an improved generative adversarial network according to an embodiment of the present invention;

[0065] Figure 3 is a schematic diagram for supplementary explanation of the network structure corresponding to the tire X-ray detection method based on the improved generative adversarial network according to one embodiment of the present invention;

[0066] Figure 4 is a flowchart of a method for performing a first feature extraction operation according to an embodiment of the present invention;

[0067] Figure 5 is a flowchart of a method for a non-first first feature extraction operation according to one embodiment of the present invention;

[0068] Figure 6 is a flowchart of a method for the last first feature extraction operation according to one embodiment of the present invention;

[0069] Figure 7 is a flowchart of a method for a second feature extraction operation according to one embodiment of the present invention. DETAILED DESCRIPTION

[0070] The following describes the specific implementation of the embodiment of the present invention in detail with reference to the accompanying drawings. It should be understood that the specific implementation described herein is only used to illustrate and explain the embodiment of the present invention and is not used to limit the embodiment of the present invention.

[0071] like Figure 1 FIG. 1 is a flow chart of a tire X-ray detection method based on an improved adversarial generative network according to an embodiment of the present invention. Figure 1 In the present invention, the detection method may include:

[0072] In step S10, an image of a tire to be inspected is obtained;

[0073] In step S11, performing at least two first feature extraction operations and one second feature extraction operation on the tire image to obtain a predicted image of the tire;

[0074] In step S12, a loss function of the predicted image and the tire image is calculated;

[0075] In step S13, it is determined whether the loss function is greater than or equal to a preset threshold;

[0076] In step S14, if the loss function is determined to be less than the threshold, five rounds of depthwise separable convolution, batch normalization, and activation are sequentially performed on the predicted image and the tire image to obtain a sixth feature map.

[0077] In step S15, the objective function is calculated based on the feature map output by the fourth round of depth-wise separable convolution operation, batch normalization operation, and activation operation and the sixth feature map output by the fifth round;

[0078] In step S16, the tire abnormality score is calculated according to the objective function;

[0079] In step S17 , it is determined whether the tire has an abnormality based on the abnormality score.

[0080] With Gai Ru Figure 1The network structure corresponding to the method shown can be as follows Figure 2 and Figure 3 As shown. Figures 1 to 3 In the embodiment, step S10 can be used to obtain a tire image to be detected. The tire image can be an X-ray image of the tire. Step S11 is used to perform at least two first feature extraction operations and one second feature extraction operation on the tire image to obtain a predicted image of the tire. Specifically, in the embodiment Figure 2 In the first feature extraction operation, the first feature extraction operation may include: Figure 4 The steps shown in Figure 4 In the first feature extraction operation, the first feature extraction operation may include:

[0081] In step S20, two rounds of depthwise separable convolution operations, batch normalization operations, and activation operations are sequentially performed on the tire image to obtain a first feature map;

[0082] In step S21, a hybrid attention mechanism operation is performed on the first feature map to obtain a second feature map;

[0083] In step S22, a skip connection operation is performed on the second feature map to obtain a third feature map;

[0084] In step S23, an upsampling operation is performed on the sixth feature map output by the next first feature extraction operation to obtain a fourth feature map;

[0085] In step S24, the third feature map and the fourth feature map are superimposed to obtain a fifth feature map;

[0086] In step S25, a depth-wise separable convolution operation, a batch normalization operation, and an activation operation are performed on the fifth feature map to obtain a sixth feature map.

[0087] In step S26, a convolution operation is performed on the fifth feature map to obtain a predicted image.

[0088] The non-first feature extraction operation may include: Figure 5 The steps shown in Figure 5 In the non-first feature extraction operation, the non-first first feature extraction operation may include:

[0089] In step S30, a maximum pooling operation is performed on the first feature map of the previous first feature extraction operation to obtain a tire image for this first feature extraction operation;

[0090] In step S31, two rounds of depthwise separable convolution operations, batch normalization operations, and activation operations are sequentially performed on the tire image to obtain a first feature map;

[0091] In step S32, a hybrid attention mechanism operation is performed on the first feature map to obtain a second feature map;

[0092] In step S33, a skip connection operation is performed on the second feature map to obtain a third feature map;

[0093] In step S34, an upsampling operation is performed on the sixth feature map output by the next first feature extraction operation to obtain a fourth feature map;

[0094] In step S35, the third feature map and the fourth feature map are superimposed to obtain a fifth feature map;

[0095] In step S36, a depth-wise separable convolution operation, a batch normalization operation, and an activation operation are performed on the fifth feature map to obtain a sixth feature map.

[0096] The last first feature extraction operation can include Figure 6 The steps shown in Figure 6 In the embodiment, the last first feature extraction operation may include:

[0097] In step S40, two rounds of depth-wise separable convolution operations, batch normalization operations, and activation operations are sequentially performed on the first feature map of the previous first feature extraction operation to obtain the first feature map of this time;

[0098] In step S41, two rounds of depthwise separable convolution operations, batch normalization operations, and activation operations are sequentially performed on the tire image to obtain a first feature map;

[0099] In step S42, a hybrid attention mechanism operation is performed on the first feature map to obtain a second feature map;

[0100] In step S43, a skip connection operation is performed on the second feature map to obtain a third feature map;

[0101] In step S44, an upsampling operation is performed on the second feature map output by the next first feature extraction operation to obtain a fourth feature map;

[0102] In step S45, the third feature map and the fourth feature map are superimposed to obtain a fifth feature map;

[0103] In step S46, a depthwise separable convolution operation, a batch normalization operation, and an activation operation are performed on the fifth feature map to obtain a sixth feature map.

[0104] The second feature extraction operation may include Figure 7 The steps shown in Figure 7 In the embodiment of the present invention, the second feature extraction operation may include:

[0105] In step S50, a maximum pooling operation is performed on the first feature map of the previous first feature extraction operation to obtain the current tire image;

[0106] In step S51, two rounds of depthwise separable convolution operations, batch normalization operations, and activation operations are sequentially performed on the tire image to obtain a first feature map;

[0107] In step S52, a hybrid attention mechanism operation is performed on the first feature map to obtain a second feature map.

[0108] In this Figure 2 and Figure 3 In the illustrated network structure, the first feature extraction operation (generator) actually consists of a U-shaped encoder-decoder structure. In the encoder of the generator, the feature map of each layer first passes through two DSCs (depthwise separable convolutions) to double the number of channels within the feature map. Then, it passes through a CBAM (hybrid attention mechanism) to obtain a CBAM feature map, which is then halved in size by applying max pooling. The CBAM feature map of each layer is skip-connected with the feature map of the corresponding layer in the decoder to obtain a new feature map with attention information. In the decoder of the generator, the feature map of each layer first passes through two DSCs to halve the number of channels within the feature map. Then, it is up-sampled to double its size. Finally, the generator outputs the generated image through a 1×1 convolution. This approach can focus more on relevant areas in the image. Figure 1 On the right side is the discriminator, which uses the same network structure as DCGAN to extract the latent representation of the input image and distinguish whether the input image is a real image or a generated image.

[0109] Among them, due to the introduction of deep-wise-separable convolution (DSC), the feature extraction operation process strikes a balance between model complexity and learning performance. Although a network with fewer parameters is less likely to overfit to the training set. However, reducing the number of network parameters may also lead to oversimplification of the network, thereby reducing learning performance. By introducing the DSC, the traditional convolution operation can be divided into per-channel convolution and per-point convolution, reducing the network complexity (number of parameters) while maintaining network performance.

[0110] The hybrid attention mechanism introduces a special order by connecting the channel attention module and the spatial attention module in the feature map. The channel attention module models the importance of each feature channel and then enhances or suppresses different channels for different images. The spatial attention module aims to locate the region of interest in the image and obtain the image weight distribution map. In the proposed method, CBAM is integrated into each encoder layer of the generator to obtain the attention information of the corresponding layer. The feature map of each layer is then passed through the channel module and the spatial attention module in sequence to obtain the channel and spatial attention information embedded in the feature map.

[0111] Steps S12 to S14 are used to calculate the loss function between the predicted image generated after the feature extraction operation and the initially input tire image, thereby determining whether the predicted image has lost the original features in the tire image. When step S14 determines that the value of the loss function is greater than or equal to the preset threshold, the predicted image is already distorted, and therefore a new tire image needs to be re-read. Conversely, when step S14 determines that the value of the loss function is less than the threshold, it indicates that the currently generated predicted image can well retain the original features in the tire image, thereby being able to well reflect the abnormality score corresponding to the tire image in the subsequent identification process. Although the loss function can be in a variety of forms known to those skilled in the art, in a preferred example of the present invention, the loss function can be, for example, formula (1),

[0112]

[0113] in, is the Manhattan distance, x is the tire image, and G(x) is the predicted image.

[0114] Steps S15 to S17 are mainly used to calculate the abnormality score based on the generated prediction image and tire image. Specifically, although the specific form of the objective function can be various forms known to those skilled in the art, in a preferred embodiment of the present invention, the objective function can be Formula (2) to Formula (4)

[0115]

[0116] , (3)

[0117] L=λ con L con +λ lat L lat +λ adv L adv , (4)

[0118] Among them, L con 、Llat 、L adv are loss functions, is the Manhattan distance, D(x) is the potential feature of the tire image x, D(G(x)) is the potential feature of the predicted image G(x), λ con ,λ lat ,λ adv They are loss functions L con 、L lat 、L adv The weight of L is the objective function. In a preferred embodiment of the present invention, λ con =1,λ lat =1,λ adv =40. In addition, during data training, the Adam function can be used as the optimization function, and the initial learning rate of the optimization function is lr = 2×10 -3 The optimization function can be updated using a lambda decay method, and the momentum of the lambda decay is β1 = 0.5, β2 = 0.999.

[0119] Similarly, the calculation of the anomaly score can be performed in various forms known to those skilled in the art. However, in a preferred embodiment of the present invention, the anomaly score can be calculated according to formula (5).

[0120] A(x)=λR(x)+(1-λ)L(x), (5)

[0121] Wherein, A(x) is the anomaly score, R(x) is the difference between the test image and the corresponding predicted image, L(x) is the difference between the latent representation of the tire image and the latent representation of the predicted image, and λ is a weighting parameter. In a preferred example of the present invention, λ can be set to 0.2.

[0122] Furthermore, in the case of a large data set, in order to avoid numerical overflow, the anomaly score can be further standardized. Specifically, the anomaly score can be processed according to formula (6):

[0123]

[0124] in, is the normalized result of the anomaly score, min(A) is the minimum value of the anomaly score, and max(A) is the maximum value of the anomaly score.

[0125] On the other hand, the present invention further provides a tire X-ray inspection system based on an improved generative adversarial network, wherein the system includes a processor configured to execute any of the above methods.

[0126] Through the above technical solution, the tire X-ray detection method and system based on the improved adversarial generative network provided by the present invention achieves high-precision detection of key areas by designing a adversarial generative network model compared to conventional networks in the prior art because the network introduces a hybrid attention mechanism and depthwise separable convolution, thereby improving the detection accuracy of X-ray images.

[0127] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0128] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the steps in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0129] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0130] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0131] In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0132] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0133] Computer-readable media includes permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. The information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include transitory computer-readable media (transitory media), such as modulated data signals and carrier waves.

[0134] It should also be noted that the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, commodity, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, an element defined by the phrase "comprises a ..." does not exclude the presence of other identical elements in the process, method, commodity, or apparatus that includes the element.

[0135] The above are merely embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should all be included within the scope of the claims of the present application.

Claims

1. A tire X-ray detection method based on an improved adversarial generative network, characterized in that: The method comprises: Acquire a tire image to be inspected; performing at least two first feature extraction operations and one second feature extraction operation on the tire image to obtain a predicted image of the tire; Calculating a loss function of the predicted image and the tire image; Determine whether the loss function is greater than or equal to a preset threshold; If it is determined that the loss function is less than the threshold, sequentially performing five rounds of depthwise separable convolution operations, batch normalization operations, and activation operations on the predicted image and the tire image to obtain a sixth feature map; Calculate the objective function based on the feature maps output by the fourth round of depth-wise separable convolution, batch normalization, and activation operations, and the sixth feature map output by the fifth round; Calculating an abnormality score of the tire according to the objective function; determining whether the tire has an abnormality according to the abnormality score; The objective function is calculated based on the feature maps output by the fourth round of depth-wise separable convolution, batch normalization, and activation operations, and the sixth feature map output by the fifth round: The objective function is calculated according to formula (2) to formula (4), ，（2） ，（3） ，（4） in, 、 、 are loss functions, is the Manhattan distance, For the tire image potential characteristics, The predicted image potential characteristics, 、 、 The loss functions are 、 、 The weight of is the objective function.

2. The detection method according to claim 1, wherein The first feature extraction operation includes: performing two rounds of depthwise separable convolution operations, batch normalization operations, and activation operations on the tire image to obtain a first feature map; Performing a hybrid attention mechanism operation on the first feature map to obtain a second feature map; Performing a skip connection operation on the second feature map to obtain a third feature map; performing an upsampling operation on a sixth feature map output by the next first feature extraction operation to obtain a fourth feature map; Superimposing the third characteristic map and the fourth characteristic map to obtain a fifth characteristic map; Performing a depthwise separable convolution operation, a batch normalization operation, and an activation operation on the fifth feature map to obtain a sixth feature map; A convolution operation is performed on the fifth feature map to obtain a predicted image.

3. The detection method according to claim 1, wherein The non-first first feature extraction operation includes: performing a maximum pooling operation on the first feature map of the previous first feature extraction operation to obtain a tire image for this first feature extraction operation; performing two rounds of depthwise separable convolution operations, batch normalization operations, and activation operations on the tire image to obtain a first feature map; Performing a hybrid attention mechanism operation on the first feature map to obtain a second feature map; Performing a skip connection operation on the second feature map to obtain a third feature map; performing an upsampling operation on a sixth feature map output by the next first feature extraction operation to obtain a fourth feature map; Superimposing the third characteristic map and the fourth characteristic map to obtain a fifth characteristic map; A depthwise separable convolution operation, a batch normalization operation, and an activation operation are performed on the fifth feature map to obtain a sixth feature map.

4. The method according to claim 1, wherein The last operation of extracting the first feature includes: Performing two rounds of depth-wise separable convolution operations, batch normalization operations, and activation operations on the first feature map of the previous first feature extraction operation to obtain the current first feature map; performing two rounds of depthwise separable convolution operations, batch normalization operations, and activation operations on the tire image to obtain a first feature map; Performing a hybrid attention mechanism operation on the first feature map to obtain a second feature map; Performing a skip connection operation on the second feature map to obtain a third feature map; performing an upsampling operation on the second feature map output by the next first feature extraction operation to obtain a fourth feature map; Superimposing the third characteristic map and the fourth characteristic map to obtain a fifth characteristic map; A depthwise separable convolution operation, a batch normalization operation, and an activation operation are performed on the fifth feature map to obtain a sixth feature map.

5. The method according to claim 1, wherein The second feature extraction operation includes: performing a maximum pooling operation on the first feature map of the previous first feature extraction operation to obtain the current tire image; performing two rounds of depthwise separable convolution operations, batch normalization operations, and activation operations on the tire image to obtain a first feature map; Perform a hybrid attention mechanism operation on the first feature map to obtain a second feature map.

6. The method according to claim 1, characterized in that Calculating the loss function of the predicted image and the tire image includes: Calculate the loss function according to formula (1) , ，（1） in, is the Manhattan distance, is the tire image, is the predicted image.

7. The method according to claim 1, characterized in that Calculating the tire abnormality score according to the objective function includes: The anomaly score is calculated according to formula (5), ，（5） in, Score the anomaly, is the difference between the test image and the corresponding predicted image, is the difference between the latent representation of the tire image and the latent representation of the predicted image, is the weighting parameter.

8. The method according to claim 1, characterized in that Calculating the tire abnormality score according to the objective function includes: ，（6） in, is the normalized result of the abnormality score, is the minimum value of the anomaly score, is the maximum value of the anomaly score.

9. A tire X-ray inspection system based on an improved adversarial generative network, characterized in that: The system comprises a processor configured to execute the method according to any one of claims 1 to 8.

Images