An insulator sparse angle CT reconstruction method and system based on error feedback

Abstract

The application discloses an insulator sparse-angle CT reconstruction method and system based on error feedback, which comprises the following steps: acquiring insulator sparse-angle projection data and generating an initial reconstruction image; constructing an image domain error feedback iterative reconstruction network composed of K cascaded correction stages; taking the initial image as the 0th stage result, calculating the cumulative correction amount of the current result and the initial image in each stage; extracting gray scale features through a gray scale branch, extracting gradient structure features through a structure branch, and fusing to generate a residual correction amount; updating the next stage result by using a learnable step length; and outputting a final reconstruction image after repeating K stages. The multi-stage error feedback mechanism gradually corrects sparse-angle artifacts, and the gray scale and structure dual-branch perception is combined, so that the structure and gray scale information of the insulator CT image is effectively restored under the condition of reducing the projection viewing angle, and the application is suitable for insulator industrial CT rapid detection.

Description

Technical Field

[0001] This invention belongs to the field of non-destructive testing of power equipment and industrial CT image reconstruction technology, and particularly relates to a CT reconstruction method and system for sparse angles of insulators based on error feedback. Background Technology

[0002] Insulators are crucial components in power systems, providing electrical insulation and mechanical support. Their internal structure directly impacts the operational safety of transmission and transformation equipment. Because insulators are typically closed or semi-closed structures, surface inspection methods are insufficient for directly observing internal defects. Industrial CT imaging, which can obtain tomographic images of the tested object without damaging the sample, is widely used for insulator internal structure analysis and defect detection.

[0003] In cone-beam CT systems, rotational scanning is typically used to acquire projection data from different angles, followed by analytical reconstruction algorithms such as FDK to obtain tomographic images. While full-view scanning provides sufficient projection information, it requires acquiring projection data from numerous angles, resulting in longer scan times and larger data volumes. In scenarios involving batch insulator inspection, rapid screening, or high-throughput laboratory testing, full-view scanning significantly impacts inspection efficiency.

[0004] Sparse-angle CT improves imaging efficiency by reducing the number of projection angles. Unlike finite-angle CT, sparse-angle CT typically still covers the entire 360° scanning angle domain, but with increased intervals between adjacent sampling angles and a reduced number of projections. However, due to insufficient angle sampling, sparse-angle reconstructed images are prone to problems such as stripe artifacts, increased noise, and blurred structural boundaries. These problems can severely affect the identification of internal pores and local defects in insulators.

[0005] Existing sparse angle CT reconstruction methods are mainly divided into model-driven and data-driven methods. Model-driven methods typically introduce total variation, dictionary learning, or other prior constraints to improve reconstruction quality iteratively. However, they are computationally expensive and sensitive to parameter selection, and can easily lead to smoothing of structural details when the number of viewpoints is small. Data-driven methods, especially deep learning image domain methods, take sparse angle FDK reconstructed images as input and directly output high-quality images through neural networks, resulting in faster inference speeds. However, most existing image domain methods use single-step end-to-end mapping, which does not adequately consider the gradual changes in reconstruction errors during the recovery process, and can easily retain artifacts or weaken structural details in local areas. In addition, these methods often fail to fully utilize the complementary advantages of grayscale and structural information, resulting in limited effectiveness in the reconstruction of objects with complex internal structures, such as insulators.

[0006] Therefore, a sparse angle CT reconstruction method that can progressively correct errors and simultaneously utilize grayscale and structural information is needed to better meet the requirements for insulator internal structure inspection. Summary of the Invention

[0007] To address the problems of artifact retention, unclear structural boundaries, and limited single-step mapping recovery capability in existing sparse angle CT reconstruction methods for insulators, this invention provides a sparse angle CT reconstruction method and system for insulators based on error feedback. The purpose of this invention is to improve the reconstruction quality of insulator CT images by progressively improving the initial sparse angle reconstructed image through multi-stage residual correction while reducing the number of projection viewpoints, and by utilizing both grayscale and structural information to guide residual correction.

[0008] In a first aspect, the present invention provides a CT reconstruction method for sparse angles of insulators based on error feedback, comprising:

[0009] Acquire sparse angle projection data of the insulator under test on the complete circular scanning trajectory, and use the analytical reconstruction algorithm to reconstruct the sparse angle projection data to obtain the initial sparse angle reconstruction image.

[0010] An image domain error feedback iterative reconstruction network is constructed. The image domain error feedback iterative reconstruction network consists of K cascaded correction stages. Each correction stage includes a cumulative correction calculation unit, a gray-level and structure dual-branch residual correction module, and a learnable step size parameter.

[0011] The cumulative correction amount is calculated based on the current reconstruction result and the initial sparse angle reconstruction image by the cumulative correction amount calculation unit to obtain the cumulative correction amount of the current correction stage. The initial sparse angle reconstruction image is taken as the reconstruction result of the 0th correction stage, and the current reconstruction result is the reconstruction result output by the current correction stage.

[0012] The current reconstruction result and the cumulative correction amount are input into the gray-scale and structure dual-branch residual correction module. The gray-scale and structure dual-branch residual correction module extracts gray-scale features through the gray-scale branch and extracts structural features through the structure branch. After fusing the gray-scale features and the structural features, it outputs the residual correction amount for the current correction stage.

[0013] Using the learnable step size parameter of the current correction stage, the current reconstruction result is updated to obtain the reconstruction result of the next stage. The correction is repeated until K correction stages are completed, and the reconstruction result of the Kth stage is output as the CT reconstruction result image of the insulator sparse angle.

[0014] Secondly, the present invention provides an insulator sparse angle CT reconstruction system based on error feedback, comprising:

[0015] The acquisition module is configured to acquire sparse angle projection data of the insulator under test on a complete circular scanning trajectory, and to reconstruct the sparse angle projection data using an analytical reconstruction algorithm to obtain an initial sparse angle reconstruction image.

[0016] The construction module is configured to construct an image domain error feedback iterative reconstruction network. The image domain error feedback iterative reconstruction network consists of K cascaded correction stages. Each correction stage includes a cumulative correction calculation unit, a gray-level and structure dual-branch residual correction module, and a learnable step size parameter.

[0017] The calculation module is configured to calculate the cumulative correction amount of the current correction stage based on the current reconstruction result and the initial sparse angle reconstruction image according to the cumulative correction amount calculation unit, wherein the initial sparse angle reconstruction image is taken as the reconstruction result of the 0th correction stage, and the current reconstruction result is the reconstruction result output by the current correction stage.

[0018] The output module is configured to input the current reconstruction result and the cumulative correction amount to the gray-scale and structure dual-branch residual correction module. The gray-scale and structure dual-branch residual correction module extracts gray-scale features through the gray-scale branch and extracts structural features through the structural branch, and outputs the residual correction amount of the current correction stage after fusing the gray-scale features and the structural features.

[0019] The reconstruction module is configured to update the current reconstruction result using the learnable step size parameter of the current correction stage, obtain the reconstruction result of the next stage, and repeat the correction until K correction stages are completed, and output the reconstruction result of the Kth stage as the CT reconstruction result image of the insulator sparse angle.

[0020] Thirdly, an electronic device is provided, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the steps of the error feedback-based insulator sparse angle CT reconstruction method according to any embodiment of the present invention.

[0021] Fourthly, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein when the program instructions are executed by a processor, the processor performs the steps of the error feedback-based insulator sparse angle CT reconstruction method according to any embodiment of the present invention.

[0022] The insulator sparse angle CT reconstruction method and system based on error feedback proposed in this application have the following beneficial effects:

[0023] The sparse angle CT image restoration process is divided into multiple cascaded correction stages. Each stage calculates the cumulative correction amount based on the difference between the current reconstruction result and the initial image, and predicts the residual correction amount accordingly. This error feedback mechanism enables the network to pay attention to areas that have not been fully corrected in previous stages, avoiding artifact residues or structural smoothing caused by single-step mapping.

[0024] Introducing cumulative correction as error feedback information enables the current correction stage to perceive the changes that have occurred from the initial image to the current result, thereby more effectively addressing areas that still exhibit degradation and improving the accuracy and stability of the correction.

[0025] A dual-branch residual correction module for grayscale and structure is adopted to extract grayscale information and structural information respectively, and the two types of information are fused for residual prediction. The grayscale branch focuses on the restoration of image grayscale distribution, while the structural branch focuses on edge and detail preservation. The two complement each other and help to maintain the clarity of the internal structure of the insulator while reducing artifacts.

[0026] By employing a learnable step size to control the correction magnitude at each stage, the network can automatically adjust the update intensity of each stage based on the training data, thereby improving the convergence stability and reconstruction quality of the multi-stage correction process. Attached Figure Description

[0027] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0028] Figure 1 A flowchart of an insulator sparse angle CT reconstruction method based on error feedback is provided as an embodiment of the present invention;

[0029] Figure 2 A structural block diagram of an insulator sparse angle CT reconstruction system based on error feedback is provided in an embodiment of the present invention;

[0030] Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0031] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0032] Please see Figure 1 The diagram shows a flowchart of an insulator sparse angle CT reconstruction method based on error feedback according to this application.

[0033] like Figure 1 As shown, the CT reconstruction method for sparse angles of insulators based on error feedback specifically includes the following steps:

[0034] Step S101: Obtain sparse angle projection data of the insulator under test on the complete circular scanning trajectory, and use the analytical reconstruction algorithm to reconstruct the sparse angle projection data to obtain the initial sparse angle reconstruction image.

[0035] In this step, the sparse angle projection data of the insulator under test is first acquired along a complete circular scanning trajectory. The object under test can be a porcelain insulator, composite insulator, glass insulator, or other electrical insulation components that require internal structural inspection. Sparse angle CT scanning still acquires projections along a complete 360° circular trajectory, but the number of projection angles is less than that of a full-view scan, and the interval between adjacent sampling angles is increased. For example, a full-view scan can acquire 360 ​​angles (interval of 1°), while a sparse angle scan only acquires 60 or 30 angles (interval of 6° or 12°). After the scan is completed, the sparse angle projection data is obtained.

[0036] The FDK algorithm is used to perform analytical reconstruction on the sparse angle projection data to obtain the initial sparse angle reconstruction image. The FDK algorithm is a commonly used analytical reconstruction method in cone-beam CT, characterized by its high computational efficiency. However, due to insufficient projections, the initial reconstructed image... It will contain degradation phenomena such as stripe artifacts, noise enhancement, and structural blurring, but its overall outline and main structure are still visible. This initial image serves as input to the subsequent image domain correction network.

[0037] Step S102: Construct an image domain error feedback iterative reconstruction network. The image domain error feedback iterative reconstruction network consists of K cascaded correction stages. Each correction stage includes a cumulative correction calculation unit, a gray-level and structure dual-branch residual correction module, and a learnable step size parameter.

[0038] In this step, the image domain error feedback iterative reconstruction network is trained using an error feedback structure-constrained loss function.

[0039] The error feedback structure constraint loss function is used to simultaneously constrain the difference between the reconstruction result after the stage update and the ground truth image, the consistency between the actual residual correction amount and the current stage's true residual error, and the reconstruction error of the structural region. The structural weights are jointly determined by the ground truth image gradient and the cumulative correction amount gradient.

[0040] The expression for the error feedback structure constraint loss function is:

[0041] ,

[0042] ,

[0043] In the formula, The error feedback structure constraint loss function, This represents the reconstruction result of the k-th correction stage. This represents the reconstruction result of the (k+1)th correction stage. For the final reconstruction result, For the true value image, This is the residual correction amount for the current stage. For learnable step size parameters, This is the stage correction amount that actually applies to the current reconstruction results. This represents the actual remaining error at the current stage. For structural weights, For gradient operators, and These represent the gradient magnitudes of the ground truth image and the cumulative correction, respectively. This indicates element-wise multiplication. The stage weights increase with each correction stage. , , and All are weighting coefficients. This is the structural similarity loss.

[0044] In one specific embodiment, the image domain error feedback iterative reconstruction network needs to be trained using a joint loss function, and the training process is as follows:

[0045] (1) Training data preparation

[0046] Full-view CT projection data of multiple insulator samples were acquired, and high-quality ground-value images were reconstructed using the FDK algorithm. Sparse angular projection data is generated from the full-view projection data by extracting data at equal intervals (e.g., extracting 1 / 6 of the viewpoint to obtain 60 viewpoints), and then the FDK algorithm is used to reconstruct the initial sparse angular reconstruction image. This forms a pair of training samples. .

[0047] (2) Forward propagation

[0048] Reconstruct the image from the initial sparse angle. The input is fed into the image domain error feedback iterative reconstruction network, and K stages of iterative correction are performed to obtain the final reconstruction result and the reconstruction results of each intermediate stage.

[0049] Through the above training, the gray-level feature extraction function, structural feature extraction function, residual prediction function, and learnable step size parameters at each stage in the image domain error feedback iterative reconstruction network are all optimized. After training, the image domain error feedback iterative reconstruction network can be used for fast inference reconstruction of new samples.

[0050] Step S103: Calculate the cumulative correction amount for the current correction stage based on the current reconstruction result and the initial sparse angle reconstruction image using the cumulative correction amount calculation unit. The initial sparse angle reconstruction image is taken as the reconstruction result of the 0th correction stage, and the current reconstruction result is the reconstruction result output by the current correction stage.

[0051] In this step, the cumulative correction amount Defined as the current reconstruction result Reconstructed image with initial sparse angle The difference.

[0052] Step S104: Input the current reconstruction result and cumulative correction amount into the gray-scale and structure dual-branch residual correction module. The gray-scale and structure dual-branch residual correction module extracts gray-scale features through the gray-scale branch and extracts structural features through the structural branch, and outputs the residual correction amount of the current correction stage after fusing the gray-scale features and the structural features.

[0053] In this step, the current reconstruction results will be... With cumulative correction The splicing is performed in the channel dimension to obtain the spliced ​​tensor. And through a gray-level feature extraction function composed of convolutional layers and residual blocks. Generate grayscale features ;

[0054] Gray-scale feature extraction function This can be further expressed as:

[0055] ,

[0056] in, Indicates the current reconstruction result With cumulative correction The input tensor obtained by concatenating the channels. Used to perform initial convolution mapping on the spliced ​​tensor to obtain initial grayscale features. It is used to further extract grayscale residual features based on the initial grayscale features, thereby enhancing grayscale error information. Used to map the enhanced features and output the grayscale features of the current stage. . , and The parameters in the data are all learned through the training process. Cumulative correction amount. It is directly introduced into the grayscale feature extraction process, enabling the grayscale branch to generate grayscale features by combining the current reconstruction results and existing error feedback information.

[0057] Using the horizontal difference operator and vertical difference operator Calculate the gradients of the current reconstruction result and the cumulative correction amount respectively, and construct the gradient input. And through structural feature extraction function Generate structural features ;

[0058] Structural feature extraction function This can be further expressed as:

[0059] ,

[0060] in, Used for gradient input Perform initial convolutional mapping to obtain initial structural features. It is used to further extract structural residual features based on the initial structural features, and enhance the information on edge and local structural changes. Used to map the enhanced structural features and output the structural features of the current stage. . , and The parameters in the training process are all learned.

[0061] grayscale features and structural features The data is fused and then predicted using the residual prediction function. Generate residual correction amount .

[0062] Residual prediction function It can be further defined as a learnable function consisting of feature fusion convolution, residual correction feature extraction, and output convolution. Its input is grayscale features. and structural features The two are first concatenated in the channel dimension, and then the residual correction amount for the current correction stage is generated through the residual prediction function. Its expression can be written as:

[0063] ,

[0064] in, This represents the fused feature obtained by splicing grayscale features and structural features in the channel dimension. Used for initial convolution mapping of fused features. Used to extract residual correction features, and further learn the error information that still needs to be compensated at the current stage. Used to map the enhanced features to the residual correction amount for the current stage. .

[0065] Step S105: Use the learnable step size parameter of the current correction stage to update the current reconstruction result, obtain the reconstruction result of the next stage, and repeat the correction until K correction stages are completed, and output the reconstruction result of the Kth stage as the CT reconstruction result image of the insulator sparse angle.

[0066] In this step, the learnable step size parameter is an adaptive step size constrained by both error feedback and structural information, expressed as:

[0067] ,

[0068] In the formula, For the first The learnable step size parameter for each correction stage To preset the upper limit of step size, For the Sigmoid function, This is the reconstruction result for the current correction phase. This is the residual correction amount predicted at the current stage. This is the cumulative correction amount. and These represent the gradient information of the current reconstruction result and the cumulative correction amount, respectively. To prevent extremely small constants with a denominator of zero, This represents the total number of calibration stages. , , , and For the first The learnable parameters corresponding to each correction stage This indicates the calibration stage number.

[0069] In summary, the method of this application acquires sparse angle projection data of insulators and generates an initial reconstructed image; constructs an image domain error feedback iterative reconstruction network consisting of K cascaded correction stages; uses the initial image as the result of stage 0, and calculates the cumulative correction amount between the current result and the initial image in each stage; extracts gray-level features through the gray-level branch and extracts gradient structure features through the structure branch, and fuses them to generate residual correction amount; updates the next stage result using a learnable step size; and outputs the final reconstructed image after repeating K stages. By gradually correcting sparse angle artifacts through a multi-stage error feedback mechanism and combining gray-level and structure dual-branch perception, the method effectively recovers the structural and gray-level information of insulator CT images under the condition of reducing the projection viewing angle, and is suitable for rapid industrial CT inspection of insulators.

[0070] Please see Figure 2 The diagram shows a structural block diagram of an insulator sparse angle CT reconstruction system based on error feedback according to this application.

[0071] like Figure 2 As shown, the insulator sparse angle CT reconstruction system 200 includes an acquisition module 210, a construction module 220, a calculation module 230, an output module 240, and a reconstruction module 250.

[0072] The acquisition module 210 is configured to acquire sparse angle projection data of the insulator under test on a complete circular scanning trajectory, and reconstruct the sparse angle projection data using an analytical reconstruction algorithm to obtain an initial sparse angle reconstructed image; the construction module 220 is configured to construct an image domain error feedback iterative reconstruction network, which consists of K cascaded correction stages, each correction stage including a cumulative correction calculation unit, a gray-level and structure dual-branch residual correction module, and a learnable step size parameter; the calculation module 230 is configured to calculate the cumulative correction amount of the current correction stage based on the current reconstruction result and the initial sparse angle reconstructed image according to the cumulative correction calculation unit, wherein the initial sparse angle reconstructed image is used as... The current reconstruction result is the reconstruction result output by the current correction stage, which is the reconstruction result of the 0th correction stage. The output module 240 is configured to input the current reconstruction result and the cumulative correction amount to the gray-scale and structure dual-branch residual correction module. The gray-scale and structure dual-branch residual correction module extracts gray-scale features through the gray-scale branch and extracts structural features through the structural branch. After fusing the gray-scale features and the structural features, it outputs the residual correction amount of the current correction stage. The reconstruction module 250 is configured to update the current reconstruction result using the learnable step size parameter of the current correction stage to obtain the reconstruction result of the next stage. The correction is repeated until K correction stages are completed, and the reconstruction result of the Kth stage is output as the CT reconstruction result image of the insulator sparse angle.

[0073] It should be understood that Figure 2 The modules and references described in the document Figure 1 The steps described in the text correspond to those in the method described above. Therefore, the operations, features, and corresponding technical effects described above also apply to the method described in the text. Figure 2 The various modules in the document will not be described in detail here.

[0074] In other embodiments, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein when the program instructions are executed by a processor, the processor performs the error feedback-based insulator sparse angle CT reconstruction method in any of the above method embodiments.

[0075] In one embodiment, the computer-readable storage medium of the present invention stores computer-executable instructions, which are configured as follows:

[0076] The sparse angle projection data of the insulator under test on the complete circular scanning trajectory is obtained, and the sparse angle projection data is reconstructed using an analytical reconstruction algorithm to obtain an initial sparse angle reconstruction image.

[0077] An image domain error feedback iterative reconstruction network is constructed, which consists of K cascaded correction stages. Each correction stage includes a cumulative correction calculation unit, a gray-level and structure dual-branch residual correction module, and a learnable step size parameter.

[0078] The cumulative correction amount is calculated based on the current reconstruction result and the initial sparse angle reconstruction image by the cumulative correction amount calculation unit to obtain the cumulative correction amount of the current correction stage. The initial sparse angle reconstruction image is taken as the reconstruction result of the 0th correction stage, and the current reconstruction result is the reconstruction result output by the current correction stage.

[0079] The current reconstruction result and cumulative correction amount are input into the gray-scale and structure dual-branch residual correction module. The gray-scale and structure dual-branch residual correction module extracts gray-scale features through the gray-scale branch and extracts structural features through the structure branch. After fusing the gray-scale features and the structural features, it outputs the residual correction amount for the current correction stage.

[0080] Using the learnable step size parameter of the current correction stage, the current reconstruction result is updated to obtain the reconstruction result of the next stage. The correction is repeated until K correction stages are completed, and the reconstruction result of the Kth stage is output as the CT reconstruction result image of the insulator sparse angle.

[0081] Computer-readable storage media may include a program storage area and a data storage area, wherein the program storage area may store an operating system and an application program required for at least one function; the data storage area may store data created based on the use of the error feedback-based insulator sparse angle CT reconstruction system, etc. Furthermore, the computer-readable storage medium may include high-speed random access memory, and may also include memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some embodiments, the computer-readable storage medium may optionally include memory remotely configured relative to a processor, which can be connected to the error feedback-based insulator sparse angle CT reconstruction system via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0082] Figure 3 This is a schematic diagram of the structure of the electronic device provided in the embodiment of the present invention, such as... Figure 3As shown, the device includes a processor 310 and a memory 320. The electronic device may also include an input device 330 and an output device 340. The processor 310, memory 320, input device 330, and output device 340 can be connected via a bus or other means. Figure 3 Taking a bus connection as an example, the memory 320 is the computer-readable storage medium described above. The processor 310 executes various server functions and data processing by running non-volatile software programs, instructions, and modules stored in the memory 320, thereby implementing the error feedback-based insulator sparse angle CT reconstruction method described in the above embodiment. The input device 330 can receive input digital or character information and generate key signal inputs related to user settings and function control of the error feedback-based insulator sparse angle CT reconstruction system. The output device 340 may include a display screen or other display device.

[0083] The aforementioned electronic device can execute the method provided in the embodiments of the present invention, and has the corresponding functional modules and beneficial effects for executing the method. Technical details not described in detail in this embodiment can be found in the method provided in the embodiments of the present invention.

[0084] In one implementation, the above-described electronic device is applied to an error feedback-based insulator sparse angle CT reconstruction system for a client, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to:

[0085] The sparse angle projection data of the insulator under test on the complete circular scanning trajectory is obtained, and the sparse angle projection data is reconstructed using an analytical reconstruction algorithm to obtain an initial sparse angle reconstruction image.

[0086] An image domain error feedback iterative reconstruction network is constructed, which consists of K cascaded correction stages. Each correction stage includes a cumulative correction calculation unit, a gray-level and structure dual-branch residual correction module, and a learnable step size parameter.

[0087] The cumulative correction amount is calculated based on the current reconstruction result and the initial sparse angle reconstruction image by the cumulative correction amount calculation unit to obtain the cumulative correction amount of the current correction stage. The initial sparse angle reconstruction image is taken as the reconstruction result of the 0th correction stage, and the current reconstruction result is the reconstruction result output by the current correction stage.

[0088] The current reconstruction result and cumulative correction amount are input into the gray-scale and structure dual-branch residual correction module. The gray-scale and structure dual-branch residual correction module extracts gray-scale features through the gray-scale branch and extracts structural features through the structure branch. After fusing the gray-scale features and the structural features, it outputs the residual correction amount for the current correction stage.

[0089] Using the learnable step size parameter of the current correction stage, the current reconstruction result is updated to obtain the reconstruction result of the next stage. The correction is repeated until K correction stages are completed, and the reconstruction result of the Kth stage is output as the CT reconstruction result image of the insulator sparse angle.

[0090] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of various embodiments or some parts of embodiments.

[0091] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A CT reconstruction method for sparse angles of insulators based on error feedback, characterized in that, include: Acquire sparse angle projection data of the insulator under test on the complete circular scanning trajectory, and use the analytical reconstruction algorithm to reconstruct the sparse angle projection data to obtain the initial sparse angle reconstruction image. An image domain error feedback iterative reconstruction network is constructed. The image domain error feedback iterative reconstruction network consists of K cascaded correction stages. Each correction stage includes a cumulative correction calculation unit, a gray-level and structure dual-branch residual correction module, and a learnable step size parameter. The cumulative correction amount is calculated based on the current reconstruction result and the initial sparse angle reconstruction image by the cumulative correction amount calculation unit to obtain the cumulative correction amount for the current correction stage. The initial sparse angle reconstruction image is taken as the reconstruction result of the 0th correction stage, and the current reconstruction result is the reconstruction result output by the current correction stage, specifically including: Current reconstruction results With cumulative correction The splicing is performed in the channel dimension to obtain the spliced ​​tensor. And through a gray-level feature extraction function composed of convolutional layers and residual blocks. Generate grayscale features ; Using the horizontal difference operator and vertical difference operator Calculate the gradients of the current reconstruction result and the cumulative correction amount respectively, and construct the gradient input. And through structural feature extraction function Generate structural features ; grayscale features and structural features The data is fused and then predicted using the residual prediction function. Generate residual correction amount ; The current reconstruction result and the cumulative correction amount are input into the gray-scale and structure dual-branch residual correction module. The gray-scale and structure dual-branch residual correction module extracts gray-scale features through the gray-scale branch and extracts structural features through the structure branch. After fusing the gray-scale features and the structural features, it outputs the residual correction amount for the current correction stage. Using the learnable step size parameter of the current correction stage, the current reconstruction result is updated to obtain the reconstruction result of the next stage. This correction is repeated until K correction stages are completed. The reconstruction result of the Kth stage is output as the CT reconstruction result image of the insulator's sparse angle. The learnable step size parameter is an adaptive step size constrained by both error feedback and structural information, and its expression is: ， In the formula, For the first The learnable step size parameter for each correction stage. To preset the upper limit of step size, For the Sigmoid function, This is the reconstruction result for the current correction phase. This is the residual correction amount predicted at the current stage. This is the cumulative correction amount. and These represent the gradient information of the current reconstruction result and the cumulative correction amount, respectively. To prevent extremely small constants with a denominator of zero, This represents the total number of calibration stages. , , , and For the first The learnable parameters corresponding to each correction stage This indicates the calibration stage number.

2. The insulator sparse angle CT reconstruction method based on error feedback according to claim 1, characterized in that, The image domain error feedback iterative reconstruction network is trained using an error feedback structure-constrained loss function. The error feedback structure constraint loss function is used to simultaneously constrain the difference between the reconstruction result after the stage update and the ground truth image, the consistency between the actual residual correction amount and the current stage's true residual error, and the reconstruction error of the structural region. The structural weights are jointly determined by the ground truth image gradient and the cumulative correction amount gradient.

3. The insulator sparse angle CT reconstruction method based on error feedback according to claim 2, characterized in that, The expression for the error feedback structure constraint loss function is as follows: ， ， In the formula, The error feedback structure constraint loss function, This represents the reconstruction result of the k-th correction stage. This represents the reconstruction result of the (k+1)th correction stage. For the final reconstruction result, For the true value image, This is the residual correction amount predicted at the current stage. For learnable step size parameters, This is the stage correction amount that actually applies to the current reconstruction results. This represents the actual remaining error at the current stage. For structural weights, For gradient operators, and These represent the gradient magnitudes of the ground truth image and the cumulative correction, respectively. This indicates element-wise multiplication. The stage weights increase with each correction stage. , , and All are weighting coefficients. This is the structural similarity loss.

4. A CT reconstruction system for sparse angles of insulators based on error feedback, characterized in that, include: The acquisition module is configured to acquire sparse angle projection data of the insulator under test on a complete circular scanning trajectory, and to reconstruct the sparse angle projection data using an analytical reconstruction algorithm to obtain an initial sparse angle reconstruction image. The construction module is configured to construct an image domain error feedback iterative reconstruction network. The image domain error feedback iterative reconstruction network consists of K cascaded correction stages. Each correction stage includes a cumulative correction calculation unit, a gray-level and structure dual-branch residual correction module, and a learnable step size parameter. The calculation module is configured to calculate the cumulative correction amount for the current correction stage based on the current reconstruction result and the initial sparse angle reconstruction image using the cumulative correction amount calculation unit. The initial sparse angle reconstruction image is used as the reconstruction result for the 0th correction stage, and the current reconstruction result is the reconstruction result output by the current correction stage. Specifically, this includes: Current reconstruction results With cumulative correction The splicing is performed in the channel dimension to obtain the spliced ​​tensor. And through a gray-level feature extraction function composed of convolutional layers and residual blocks. Generate grayscale features ; Using the horizontal difference operator and vertical difference operator Calculate the gradients of the current reconstruction result and the cumulative correction amount respectively, and construct the gradient input. And through structural feature extraction function Generate structural features ; grayscale features and structural features The data is fused and then predicted using the residual prediction function. Generate residual correction amount ; The output module is configured to input the current reconstruction result and the cumulative correction amount to the gray-scale and structure dual-branch residual correction module. The gray-scale and structure dual-branch residual correction module extracts gray-scale features through the gray-scale branch and extracts structural features through the structural branch, and outputs the residual correction amount of the current correction stage after fusing the gray-scale features and the structural features. The reconstruction module is configured to update the current reconstruction result using the learnable step size parameter of the current correction stage, obtain the reconstruction result of the next stage, and repeat the correction until K correction stages are completed. The reconstruction result of the Kth stage is then output as the CT reconstruction result image of the insulator's sparse angle. The learnable step size parameter is an adaptive step size constrained by both error feedback and structural information, and its expression is: ， In the formula, For the first The learnable step size parameter for each correction stage. To preset the upper limit of step size, For the Sigmoid function, This is the reconstruction result for the current correction phase. This is the residual correction amount predicted at the current stage. This is the cumulative correction amount. and These represent the gradient information of the current reconstruction result and the cumulative correction amount, respectively. To prevent extremely small constants with a denominator of zero, This represents the total number of calibration stages. , , , and For the first The learnable parameters corresponding to each correction stage This indicates the calibration stage number.

5. An electronic device, characterized in that, include: At least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 3.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method according to any one of claims 1 to 3.

Images