A projection image super-resolution reconstruction method, an imaging system and a storage medium

Abstract

The embodiment of the application relates to the technical field of super-resolution reconstruction, and particularly relates to a projection image super-resolution reconstruction method, an imaging system and a storage medium, the method is applied to the imaging system, and the imaging system comprises a ray source and a detector. The embodiment of the application obtains a standard projection image by carrying out air removal and normalization processing on an initial projection image, then inputs the standard projection image into a super-resolution neural network model to obtain a super-resolution image, improves the noise removal capability of the super-resolution neural network model, effectively filters out noise, can reduce the influence of super-resolution on image details, and thus, geometric parameters of a reconstructed super-resolution image are calculated based on the super-resolution image, the super-resolution image is reconstructed according to the geometric parameters, a super-resolution reconstructed image is obtained, the reconstructed super-resolution reconstructed image restores more, more detailed and real image details, reduces the artifact problem of the reconstructed image, and has excellent super-resolution effect.

Description

Technical Field

[0001] The present invention relates to the field of super-resolution reconstruction technology, and in particular to a super-resolution reconstruction method, imaging system and storage medium for projected images. Background Technology

[0002] Among related technologies, using neural networks for image super-resolution is a relatively mature approach, but it is primarily used for processing natural images. Currently, super-resolution of CT reconstructed images involves training a neural network model and using that model to improve the image resolution. However, this method struggles to restore the image details lost in low-resolution CT reconstructed images.

[0003] It's understandable that CT reconstructed images, being 3D slices calculated from the original projected images, differ in image characteristics from natural images. The difference between low-resolution and high-resolution CT reconstructed images is significantly different from the difference between natural images. Some subtle details may disappear entirely in low-resolution CT reconstructed images, while traces of these details may remain in low-resolution natural images. Simply applying super-resolution methods from natural images to CT reconstructed images yields mediocre results, failing to restore lost details and resulting in a poor user experience. Summary of the Invention

[0004] In view of this, one objective of the present invention is to provide a super-resolution reconstruction method, imaging system and storage medium for projected images, so as to improve the situation where the super-resolution effect of CT reconstructed images is generally poor and the user experience is bad.

[0005] To address the aforementioned technical problems, the embodiments of the present invention provide the following technical solutions:

[0006] In a first aspect, embodiments of the present invention provide a super-resolution reconstruction method for projected images, applied to an imaging system, the imaging system including a radiation source and a detector, the method comprising:

[0007] Acquire the initial projection image to be super-reconstructed by the detector, which includes the projection pattern formed by the projection of the target object;

[0008] The initial projection image is subjected to air removal and normalization processing to obtain the standard projection image;

[0009] The standard projected image is input into the super-resolution neural network model to obtain the super-resolution image. The super-resolution neural network model is a pre-built neural network model trained with sample image data to meet preset conditions and is configured to perform super-resolution on the projected image.

[0010] Based on the super-resolution image, geometric parameters for reconstructing the super-resolution image are calculated. These geometric parameters include the first distance between the target object and the X-ray source and the second distance between the detector and the X-ray source.

[0011] The super-resolution image is reconstructed based on geometric parameters to obtain the target image, which is the super-resolution reconstructed image of the initial projected image.

[0012] In some embodiments, a standard projected image is input into a super-resolution neural network model to obtain a super-resolution image, including:

[0013] Primary features are extracted from the standard projection image to obtain a candidate feature matrix, which includes the primary features of the standard projection image.

[0014] High-level features are extracted from the candidate feature matrix to obtain the reference feature matrix, which includes high-level features of the standard projected image.

[0015] Upsampling is performed on the high-level features in the reference feature matrix to obtain the reference image;

[0016] A super-resolution image is obtained based on the reference image and the standard projection image.

[0017] In some embodiments, the super-resolution neural network model includes a primary feature extraction layer to extract primary features from the standard projected image and obtain a candidate feature matrix, including:

[0018] Primary features in the standard projection image are extracted using multiple convolutional kernels of the primary feature extraction layer, resulting in multiple sets of candidate features, which are the primary features of the standard projection image.

[0019] Multiple sets of candidate features are concatenated to obtain a candidate feature matrix.

[0020] In some embodiments, the super-resolution neural network model includes multiple cascaded high-level feature extraction layers to extract high-level features from the candidate feature matrix, resulting in a reference feature matrix, including:

[0021] The candidate feature matrix is ​​input into the first-level high-level feature extraction layer to extract high-level features, resulting in the first-level output feature matrix of the first-level high-level feature extraction layer.

[0022] The first-level output feature matrix is ​​input into the next-level advanced feature extraction layer to extract advanced features, thus obtaining the next-level output feature matrix of the next-level advanced feature extraction layer.

[0023] The remaining advanced feature extraction layers are used to repeat the process: the output feature matrix of the next level is input into the advanced feature extraction layer of the next level to extract advanced features, and the output feature matrix of the next level of the advanced feature extraction layer is obtained, until the last advanced feature extraction layer extracts the advanced features of the reference feature matrix, and the reference feature matrix is ​​the output feature matrix of the advanced feature extraction layer above the last advanced feature extraction layer.

[0024] In some embodiments, upsampling is performed on high-level features in the reference feature matrix to obtain a reference image, including:

[0025] Determine the scaling factor, which is the magnification of the standard projected image for super-resolution;

[0026] Based on the scaling factor, the high-level features in the reference feature matrix are upsampled to obtain the reference image.

[0027] In some embodiments, a super-resolution image is obtained based on a reference image and a standard projected image, including:

[0028] An upsampling operation is performed on the standard projection image to obtain a reference image, the size of which is the same as that of the reference image.

[0029] The super-resolution image is obtained by fusing the baseline image and the reference image.

[0030] In some embodiments, the initial projected image is subjected to air removal and normalization processing to obtain a standard projected image, including:

[0031] Acquire an air projection image. The air projection image is the projection image taken by the detector when no object is placed between the radiation source and the detector. The shooting conditions when the detector takes the air projection image are the same as those when it takes the initial projection image. The shooting conditions include light intensity and second distance.

[0032] Based on the air projection image and the initial projection image, candidate projection images are obtained;

[0033] The grayscale values ​​of pixels in the candidate projection image are normalized to obtain the standard projection image.

[0034] In some embodiments, candidate projection images are obtained based on the air projection image and the initial projection image, including:

[0035] The grayscale value of the pixel at the candidate position in the air projection image is added to the first preset value to obtain the first grayscale value of the pixel at the candidate position.

[0036] The gray value of the pixel at the candidate position in the initial projection image is added to the second preset value to obtain the second gray value of the pixel at the candidate position. The candidate position is any position in the air projection image and the initial projection image.

[0037] Based on the first gray value and the second gray value, the target gray value of the pixel at the candidate position is calculated;

[0038] Based on the coordinates of each position in the air projection image or the initial projection image, the target gray values ​​of pixels at all positions are combined to obtain the candidate projection image.

[0039] In a second aspect, embodiments of the present invention provide an imaging system, comprising:

[0040] A controller, and a radiation source and a detector respectively communicated with the controller;

[0041] The controller includes:

[0042] A processor and a memory communicatively connected to the processor;

[0043] The memory stores computer program instructions executable by the processor, which, when executed by the processor, cause the controller to perform any of the projection image super-resolution reconstruction methods proposed in the first aspect.

[0044] In a third aspect, embodiments of the present invention provide a computer-readable storage medium storing processor-executable computer program instructions, which, when executed by a processor, cause the computer to perform any of the projection image super-resolution reconstruction methods proposed in the first aspect.

[0045] The embodiments of the present invention have the following beneficial effects: Unlike related technologies, the projection image super-resolution reconstruction method provided in the embodiments of the present invention is applied to an imaging system, which includes a radiation source and a detector. The method includes: acquiring an initial projection image to be super-resolution reconstructed by the detector, wherein the initial projection image includes a projection pattern formed by the projection of a target object; performing air removal and normalization processing on the initial projection image to obtain a standard projection image; inputting the standard projection image into a super-resolution neural network model to obtain a super-resolution image, wherein the super-resolution neural network model is a model obtained by training a pre-constructed neural network model using sample image data to meet preset conditions, and is configured to perform super-resolution on the projection image; calculating geometric parameters for reconstructing the super-resolution image based on the super-resolution image, wherein the geometric parameters include a first distance between the target object and the radiation source and a second distance between the detector and the radiation source; and reconstructing the super-resolution image based on the geometric parameters to obtain a target image, wherein the target image is a super-resolution reconstructed image of the initial projection image.

[0046] This invention provides a standard projection image obtained by removing air and normalizing the initial projection image. The standard projection image is then input into a super-resolution neural network model to obtain a super-resolution image, thereby improving the noise removal capability of the super-resolution neural network model and effectively filtering out noise. This reduces the impact of super-resolution on image details. Based on the super-resolution image, the geometric parameters of the reconstructed super-resolution image are calculated, and the super-resolution image is reconstructed according to the geometric parameters to obtain a super-resolution reconstructed image. This results in the reconstructed super-resolution image restoring more detailed and realistic image details, reducing artifact problems in the reconstructed image, and exhibiting excellent super-resolution performance. Attached Figure Description

[0047] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the prior art or embodiments will be briefly introduced below. Obviously, the drawings described below only show some embodiments of the present invention and should not be considered as limiting the scope of protection. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0048] Figure 1 These are schematic diagrams illustrating application scenarios of the super-resolution reconstruction method for projected images in some embodiments of the present invention;

[0049] Figure 2 These are schematic diagrams of the imaging system provided in some embodiments of the present invention;

[0050] Figure 3 This is a schematic diagram of the structure of the controller in an imaging system provided in some embodiments of the present invention;

[0051] Figure 4 This is a flowchart illustrating the projection image super-resolution reconstruction method provided in some embodiments of the present invention;

[0052] Figure 5a This is a schematic diagram of the initial projected image in some embodiments of the present invention;

[0053] Figure 5b This is a schematic diagram of an air projection image in some embodiments of the present invention;

[0054] Figure 5c This is a schematic diagram of candidate projection images in some embodiments of the present invention;

[0055] Figure 6 This is a schematic diagram of the initial projection image captured by the detector in some embodiments of the present invention;

[0056] Figure 7 yes Figure 6 A schematic diagram of the super-resolution image obtained after super-resolution of the initial projected image shown in the embodiment;

[0057] Figure 8 yes Figure 7 A schematic diagram of the target image obtained after super-resolution image reconstruction, as shown in the embodiment;

[0058] Figure 9 This is a schematic diagram illustrating the process of obtaining a super-resolution image from a standard projection image in some embodiments of the present invention;

[0059] Figure 10 This is a schematic diagram of how a super-resolution neural network model processes candidate feature matrices to obtain reference feature matrices in some embodiments of the present invention. Detailed Implementation

[0060] To make the objectives and advantages of the embodiments of the present invention more readily understood, 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 a part of the embodiments of the present invention, and not all of them. The detailed description of the embodiments of the present invention in the accompanying drawings is not intended to limit the scope of protection claimed by the present invention, but only to illustrate selected embodiments of the present invention. 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.

[0061] It should be noted that, unless there is a conflict, the various technical features involved in the embodiments of the present invention described below can be combined with each other, and all are within the protection scope of the present invention. Furthermore, although functional modules are divided in the device or structural schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. In addition, the terms "first," "second," "third," and other similar expressions used herein do not limit the data or execution order, but are only for illustrative purposes and to distinguish identical or similar items with substantially the same function and effect, and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features.

[0062] Unless otherwise defined, the technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the invention. It should be understood that the term "and / or" as used herein includes any and all combinations of one or more of the listed items.

[0063] Among related technologies, using neural networks for image super-resolution is a relatively mature approach, but it is primarily used for processing natural images. Currently, super-resolution of CT reconstructed images involves training a neural network model and using that model to improve the image resolution. However, this method struggles to restore the image details lost in low-resolution CT reconstructed images.

[0064] Understandably, since CT reconstructed images are 3D slices calculated from the original projected images, their image features differ from natural images. The difference between low-resolution and high-resolution CT reconstructed images is significantly different from the difference between natural images. Some subtle details may disappear entirely in low-resolution CT reconstructed images, while some traces will remain in low-resolution natural images. The inventors found that simply applying super-resolution methods from natural images to CT reconstructed images yields only mediocre results and cannot restore lost image details, resulting in a poor user experience.

[0065] In the super-resolution reconstruction method for projection images provided in this invention, a standard projection image is obtained by removing air and normalizing the initial projection image captured by the detector. The standard projection image is then input into a super-resolution neural network model to obtain a super-resolution image. This improves the noise removal capability of the super-resolution neural network model, effectively filters out noise, and reduces the impact of super-resolution on image details. Based on the super-resolution image, the geometric parameters for reconstructing the super-resolution image are calculated. The super-resolution image is then reconstructed according to the geometric parameters to obtain a super-resolution reconstructed image. This results in the reconstructed super-resolution image restoring more detailed and realistic image details, reducing artifact problems in the reconstructed image, and exhibiting excellent super-resolution performance.

[0066] Please see Figure 1 , Figure 1 The illustration shows a schematic diagram of an application scenario of the projection image super-resolution reconstruction method provided in some embodiments of the present invention.

[0067] See Figure 1 As shown, the application scenario includes an imaging system 100, which includes a controller 110, an X-ray source 120, and a detector 130. The X-ray source 120 and the detector 130 are respectively connected to the controller 110 via a network. It should be understood that examples of networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0068] For example, the radiation source 120 emits radiation towards the target object 200, which is located within the detection field of the detector 130. The detector 130 captures a projected image of the target object 200 to obtain an initial projected image. The initial projected image is the image of the target object 200 formed on the plane of the detector 130. The purpose of this embodiment is to super-resolution reconstruct the initial projected image. The target object 200 can be a circuit board, electronic component, or any other suitable type of object.

[0069] In this embodiment of the invention, the controller 110 obtains the initial projection image captured by the detector 130 through the network, or the detector 130 actively sends the initial projection image captured by it to the controller 110 through the network. In this way, the controller 110 obtains the initial projection image to be super-reconstructed.

[0070] It is understandable that since super-resolution enhances images, image enhancement inevitably leads to noise amplification, especially salt-and-pepper noise. Salt-and-pepper noise has a significant impact on the reconstructed image, potentially resulting in unrealistic image details. Therefore, this embodiment of the invention significantly improves the super-resolution neural network model's ability to remove salt-and-pepper noise and reduces its impact on image details by performing air removal and normalization processes on the projected image before inputting the processed image into the super-resolution neural network model. Specifically, after obtaining the initial projected image, air removal and normalization processes are performed on it to obtain a standard projected image. The standard projected image is then input into the super-resolution neural network model, which performs super-resolution processing on the standard projected image to obtain the super-resolution image.

[0071] It is worth noting that the super-resolution neural network model in this embodiment of the invention is a model obtained by training a pre-constructed neural network model with sample image data until it meets the preset convergence conditions (e.g., the super-resolution loss is less than or equal to the preset loss value), and configuring the final parameters of the model, and configuring the super-resolution neural network model to perform super-resolution on the projected image.

[0072] Subsequently, geometric parameters for reconstructing the super-resolution image are calculated based on the super-resolution image. These geometric parameters include, but are not limited to, the first distance between the target object 200 and the X-ray source 120, the second distance between the detector 130 and the X-ray source 120, and the positional relationship (e.g., pose and relative motion state) between the target object 200 and the X-ray source 120 and the detector 130. Finally, the super-resolution image is reconstructed based on the geometric parameters to obtain the target image, which is the super-resolution reconstructed image of the initial projected image (i.e., the super-resolution reconstructed image).

[0073] It should be understood that, Figure 1The illustrated application scenarios are merely illustrative of one instance of super-resolution and reconstruction of projected images by the imaging system 100 in some embodiments of the present invention, and do not limit the structure, type, or number of imaging systems in other application scenarios or embodiments. For example, in other application scenarios or embodiments, the imaging system may also include... Figure 1 The structure shown has more or fewer components, or has the same as Figure 1 The diagram shows different configurations of the structure.

[0074] To facilitate understanding of the super-resolution reconstruction method for projected images provided in the embodiments of the present invention, the imaging system provided in the embodiments of the present invention will first be described in detail.

[0075] Please see Figure 2 , Figure 2 A schematic diagram of the imaging system in some embodiments of the present invention is shown.

[0076] like Figure 2 As shown, the imaging system 100 includes a controller 110 and a radiation source 120 and a detector 130 that are respectively connected to the controller 110 in communication. The radiation source 120 is used to emit radiation, the detector 130 is used to capture projected images, and the controller is used to coordinate and control the radiation source 120 and the detector 130 to work together to complete the super-resolution and reconstruction of the projected images.

[0077] It is worth noting that the X-ray source 120 and detector 130 can be placed in any suitable location, as long as they are suitable for emitting X-rays to the target object and for capturing projected images. The X-ray source 120 and detector 130 can be any suitable type of device, equipment, or component.

[0078] Please see Figure 3 , Figure 3 The schematic diagram illustrates the structure of the controller in the imaging system provided by some embodiments of the present invention.

[0079] See Figure 3 As shown, the controller 110 includes at least one processor 111 and a memory 112 that are communicatively connected. Figure 3 Taking a bus system 113 and a processor 111 as an example, the various components in the controller 110 are coupled together through the bus system 113, which is used to realize the connection and communication between the various components. It is easy to understand that the bus system 113 may include, in addition to the data bus, a power bus, a control bus, and a status signal bus, etc. However, for the sake of clarity and brevity, in... Figure 3 The general labels all buses as Bus System 113. Understandably, Figure 3The structures shown in the embodiments are merely illustrative and do not limit the structure of the controller described above. For example, the controller may also include components that are more... Figure 3 The structure shown has more or fewer components, or has the same as Figure 3 The diagram shows different configurations of the structure.

[0080] The processor 111 provides computational and control capabilities to support the controller 110 in executing corresponding business logic and functions. For example, it supports the controller 110 in executing any of the projection image super-resolution reconstruction methods provided in this embodiment of the invention, or in executing the steps of any possible implementation of any of the projection image super-resolution reconstruction methods provided in this embodiment of the invention. Those skilled in the art will understand that the processor 111 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0081] The memory 112, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs, non-transitory computer-executable programs, instructions, and modules, such as the program, instructions, and modules corresponding to the projection image super-resolution reconstruction method in the embodiments of the present invention. In some embodiments, the memory 112 may include a program storage area and a data storage area. The program storage area may store the operating system and application programs required for at least one function, and the data storage area may store data created according to the use of the processor 111. The processor 111 executes various functional applications and data processing of the controller 110 by running the non-transitory software programs, instructions, and modules stored in the memory 112, thereby implementing any projection image super-resolution reconstruction method provided in the embodiments of the present invention, or executing the steps in any possible implementation of any projection image super-resolution reconstruction method provided in the embodiments of the present invention. In some embodiments, the memory 112 may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, the memory 112 may also include memory remotely located relative to the processor 111, and these remotely located memories may be connected to the processor 111 through a communication network. It is understood that examples of the aforementioned communication networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0082] As can be understood from the above, the implementing entity of any projection image super-resolution reconstruction method provided in the embodiments of the present invention can be any suitable type of controller with certain computing and control capabilities, such as the controller 110 of the imaging system 100 described above. In some feasible implementations, any projection image super-resolution reconstruction method provided in the embodiments of the present invention can be implemented by a processor executing computer program instructions stored in memory.

[0083] The super-resolution reconstruction method for projected images provided by the present invention will be described in detail below with reference to exemplary applications and implementations of the imaging system provided in the embodiments of the present invention.

[0084] Please see Figure 4 , Figure 4 The schematic diagram illustrates a flowchart of a projection image super-resolution reconstruction method provided in some embodiments of the present invention.

[0085] Understandably, the projection image super-resolution reconstruction method provided in this embodiment of the invention can be applied to the aforementioned controller (e.g., the controller 110 of the imaging system 100). Specifically, the execution entity of the projection image super-resolution reconstruction method is the controller of the imaging system.

[0086] For example, such as Figure 4As shown, the super-resolution reconstruction method for projected images provided in this embodiment of the invention includes, but is not limited to, the following steps S41-S45:

[0087] S41: Acquire the initial projected image to be super-reconstructed, captured by the detector.

[0088] In this embodiment, the image captured by the detector is an initial projection image, which includes the projection pattern formed on the detector after the target object is projected by the X-ray source. In other words, the initial projection image includes the projection pattern formed by the target object. This embodiment of the invention directly obtains the initial projection image to be super-reconstructed from the detector.

[0089] S42: Perform air removal and normalization processing on the initial projected image to obtain the standard projected image.

[0090] Here, "removing air" refers to calculating the grayscale value of each pixel in both the initial projected image and the air-projected image to obtain the processed grayscale value of each pixel in the initial projected image. "Normalization" refers to normalizing the grayscale values ​​of each pixel in the initial projected image to a uniform range to obtain the normalized projected image.

[0091] It can be understood that the air projection image is the projection image taken by the detector when no object is placed between the radiation source and the detector. The shooting conditions when the detector takes the air projection image are the same as those when it takes the initial projection image. The shooting conditions include the light intensity and the second distance, that is, under the same light intensity and the second distance between the detector and the radiation source, the detector takes and acquires the air projection image and the initial projection image.

[0092] In some optional embodiments, the initial projected image is subjected to air removal and normalization processing to obtain a standard projected image, specifically including but not limited to the following steps S421-S423:

[0093] S421: Acquire an aerial projection image.

[0094] S422: Obtain candidate projection images based on the air projection image and the initial projection image.

[0095] For example, in this embodiment of the invention, the air projection image captured by the detector is directly obtained, and the grayscale value of the initial projection image and the air projection image is calculated pixel by pixel. For example, the average grayscale value of the pixels at corresponding positions in the initial projection image and the air projection image is calculated to obtain the reference grayscale value of the pixels at the corresponding positions. The reference grayscale values ​​of the pixels at all positions are combined together to obtain the candidate projection image.

[0096] For example, please see Figures 5a to 5c , Figure 5aThe initial projection images of some embodiments of the present invention are shown. Figure 5b The following are aerial projection images from some embodiments of the present invention. Figure 5c The illustration schematically shows candidate projection images in some embodiments of the present invention. The initial projection image, the aerial projection image, and the candidate projection images all include a reference position and a reference position. The reference position includes pixel positions 1 to 26, and the reference position is any position in the initial projection image, the aerial projection image, and the candidate projection image other than the reference position. For Figures 5a to 5c The air removal process involves averaging the pixels at corresponding positions in both the air projection image and the initial projection image. However, since the gray values ​​of pixels at the reference position are the same in both the air projection image and the initial projection image, while the gray values ​​of pixels at the reference position are different, the average gray value of the reference position is simply the reference position itself. Therefore, only the average gray values ​​of pixels at the reference position (i.e., pixel positions 1 to 26) in both the air projection image and the initial projection image need to be calculated to obtain the reference gray values ​​of the pixels at the reference position. The reference gray values ​​of the pixels at the reference position are then combined with the gray values ​​of the pixels at the reference position to obtain the result as shown below. Figure 5c The candidate projection image shown.

[0097] It is easy to understand that, in addition to calculating the average gray value of the pixels at the corresponding positions in the initial projected image and the air projected image, other calculation methods can be used to calculate the reference gray value of the pixels at the corresponding positions. Any suitable calculation method can be selected according to actual needs, and the embodiments of the present invention do not impose any limitations on this.

[0098] For example, in some embodiments, a candidate projection image is obtained based on the air projection image and the initial projection image, specifically including but not limited to the following steps S4221-S4224:

[0099] S4221: Add the gray value of the pixel at the candidate position in the air projection image to the first preset value to obtain the first gray value of the pixel at the candidate position.

[0100] S4222: Add the gray value of the pixel at the candidate position in the initial projected image to the second preset value to obtain the second gray value of the pixel at the candidate position.

[0101] S4223: Calculate the target gray value of the pixel at the candidate position based on the first gray value and the second gray value.

[0102] In this embodiment of the invention, the candidate position is any position in the aerial projection image and the initial projection image, for example, the candidate position is... Figure 5a Pixel position 4 in the initial projected image shown.

[0103] For example, the formula for calculating the target grayscale value of a pixel in this embodiment of the invention is as follows: , Let be the first grayscale value of the pixel at the candidate location. The second grayscale value of the pixel at the candidate location. Let be the target grayscale value of the pixel at the candidate position. After calculating the first grayscale value and the second grayscale value of the pixel at the candidate position in the air projection image and the initial projection image respectively, substitute the first grayscale value and the second grayscale value into the above calculation formula to calculate the target grayscale value of the pixel at the candidate position.

[0104] S4224: Based on the coordinates of each position in the air projection image or the initial projection image, combine the target gray values ​​of pixels at all positions to obtain a candidate projection image.

[0105] For example, based on the coordinates of each position in the air projection image or the initial projection image, the target gray values ​​of all pixels at all positions are combined in order of position coordinates to obtain the candidate projection image after air removal processing.

[0106] S423: Normalize the gray values ​​of pixels in the candidate projection image to obtain the standard projection image.

[0107] Specifically, determine a uniform range of pixel grayscale values, for example The gray values ​​of all pixels in the candidate projection image are normalized to a unified range, resulting in a standard projection image where the gray values ​​of the pixels are located in the unified range.

[0108] S43: Input the standard projection image into the super-resolution neural network model to obtain the super-resolution image.

[0109] In this embodiment, the super-resolution neural network model is a model obtained by training a pre-built neural network model with sample image data until it meets preset conditions. Based on the characteristics of the image super-resolution task, a neural network model is pre-built and designed. A sufficient number of sample image data are used to train the pre-built neural network model. That is, the loss is calculated based on the super-resolution image output by the model and the real super-resolution projected image during the training process. The model parameters are adjusted according to the loss, and the neural network model is iteratively trained until the neural network model meets the preset convergence condition, thus obtaining the trained super-resolution neural network model. The super-resolution neural network model is then configured to perform super-resolution on the projected image, that is, the super-resolution neural network model is configured to perform super-resolution on the projected image to improve the resolution of the projected image.

[0110] Understandably, the above content only briefly describes the super-resolution neural network model obtained by training a pre-built neural network model using sample image data to meet preset conditions. In the actual training process, it is necessary to perform multi-dimensional and comprehensive training, parameter tuning, and calculation on the neural network model to finally obtain the super-resolution neural network model.

[0111] In some embodiments, a high signal-to-noise ratio projected image is obtained through long exposure of the detector, and the grayscale value of the projected image is linearly scaled according to the exposure time. The projected image is denoised once using the BM3D (Block-Matching and 3D Filtering) algorithm, and then subjected to air removal processing. The denoised and air-removed projected image is used as high-resolution label image data. Specifically, the high-resolution label image data is bicubic downsampled to obtain low-resolution training image data. The sample image data includes both high-resolution label image data and low-resolution training image data.

[0112] For example, a standard projection image is input into a pre-trained super-resolution neural network model, and the super-resolution neural network model is used to super-resolution the standard projection image to obtain a super-resolution image (i.e., a projection image with higher resolution).

[0113] S44: Calculate the geometric parameters of the reconstructed super-resolution image based on the super-resolution image.

[0114] In this embodiment of the invention, the geometric parameters include a first distance between the target object and the radiation source and a second distance between the detector and the radiation source.

[0115] Specifically, based on the super-resolution image, geometric parameters related to the reconstructed super-resolution image are extracted or calculated. These geometric parameters are used to accurately describe the geometric positional relationships (such as pose and distance) between imaging system devices and are an important foundation for subsequent reconstruction calculations.

[0116] Several markers with known 3D coordinates are placed on the surface of the target object. The pixel coordinates of the markers on the super-resolution image are detected using super-resolution image. The camera matrix / projection matrix P is obtained by fitting using the Direct Linear Transform (DLT) algorithm or the least squares method. The camera matrix / projection matrix P is decomposed to obtain the detector intrinsic parameters (including pixel size, principal point, focal length approximation, etc.) and extrinsic parameters (source position and coordinate system rotation / tilt, translation offset). The first distance and the second distance are calculated based on the decomposed pixel size, principal point and transformation relationship. Finally, the first distance, the second distance, the detector principal point and pixel size, the detector tilt, offset and rotation, and the camera matrix / projection matrix P are extracted.

[0117] S45: Reconstruct the super-resolution image based on geometric parameters to obtain the target image.

[0118] In this embodiment, the target image is a super-resolution reconstructed image of the initial projection image, that is, a high-quality image obtained after correcting the geometric distortion, sampling and blur of the initial projection image, super-resolution and reconstruction, with higher clarity and resolution.

[0119] For example, a mapping from pixel coordinates to the detector plane is established using geometric parameters, mapping the pixel coordinates of the super-resolution image to continuous coordinates on the detector plane. For instance, for each pixel in the super-resolution image, its sampling position on the original sampling grid is calculated, and inverse resampling is performed using a higher-order interpolation algorithm (such as bicubic or Lanczos algorithm). The area scaling factor is calculated simultaneously during pixel mapping, and the pixel intensity / illuminance is calibrated and physically normalized to finally obtain the reconstructed super-resolution image (i.e., the target image).

[0120] For example, please see Figures 6 to 8 , Figure 6 The target object is schematically shown at a distance of 30 from the detector. The initial projection image captured by the time detector. Figure 7 It shows the Figure 6 The super-resolution image obtained after super-resolution of the initial projected image by a preset magnification (e.g., 2x) is shown. Figure 8 It shows the Figure 7 The target image is obtained after reconstructing the super-resolution image shown. It can be seen that... Figure 8 The target image shown has higher clarity and resolution, restoring more detailed and realistic image details, with excellent super-resolution effect.

[0121] This invention provides a standard projection image obtained by removing air and normalizing the initial projection image. The standard projection image is then input into a super-resolution neural network model to obtain a super-resolution image, thereby improving the noise removal capability of the super-resolution neural network model and effectively filtering out noise. This reduces the impact of super-resolution on image details. Based on the super-resolution image, the geometric parameters of the reconstructed super-resolution image are calculated, and the super-resolution image is reconstructed according to the geometric parameters to obtain a super-resolution reconstructed image. This results in the reconstructed super-resolution image restoring more detailed and realistic image details, reducing artifact problems in the reconstructed image, and exhibiting excellent super-resolution performance.

[0122] In some optional embodiments, a standard projection image is input into a super-resolution neural network model to obtain a super-resolution image, specifically including but not limited to the following steps S431-S434:

[0123] S431: Extract primary features from the standard projection image to obtain the candidate feature matrix.

[0124] For example, please refer to Figure 9 The super-resolution neural network model includes a first convolutional layer, a NAF module layer, a second convolutional layer, and a pixel reconstruction layer. After inputting a standard projection image into the super-resolution neural network model, the model uses the first convolutional layer to extract primary features from the standard projection image, such as local texture, edges, and low-level structures (preliminary separation of high-frequency and low-frequency components), to obtain a candidate feature matrix. The candidate feature matrix includes the primary features of the standard projection image. It should be understood that, in this embodiment, the first convolutional layer includes 16 pixels with a size of [missing information]. The convolution kernel can be a shallow convolutional network (3-6 layers) or a convolutional block with residuals.

[0125] For example, in some embodiments, primary features are extracted from the standard projected image to obtain a candidate feature matrix, specifically including but not limited to the following steps S4311-S4312:

[0126] S4311: Utilize multiple convolutional kernels of the primary feature extraction layer to extract primary features from the standard projection image, obtaining multiple sets of candidate features.

[0127] S4312: Concatenate multiple sets of candidate features to obtain a candidate feature matrix.

[0128] In this embodiment, the super-resolution neural network model includes a primary feature extraction layer, which is... Figure 9 The first convolutional layer shown, the primary feature extraction layer, includes multiple layers of size [missing information]. The convolution kernel.

[0129] For example, the super-resolution neural network model uses multiple convolutional kernels of the primary feature extraction layer to extract primary features from the standard projection image, and obtains multiple sets of candidate features. The candidate features are the primary features of the standard projection image (such as local texture, edges and low-level structures, etc.), and each convolutional kernel corresponds to a set of candidate features.

[0130] Specifically, the candidate features extracted by all convolutional kernels (i.e., multiple sets of candidate features) are concatenated and fused to obtain a candidate feature matrix. For example, in some embodiments, the primary feature extraction layer includes 16 kernels with a size of [missing information]. The convolutional kernels are used to extract a set of candidate features from each of the 16 convolutional layers. Then, the candidate features extracted by the 16 convolutional layers are concatenated to obtain a candidate feature matrix.

[0131] S432: Extract high-level features from the candidate feature matrix to obtain the reference feature matrix.

[0132] For example, please refer to Figure 9The super-resolution neural network model utilizes the NAF module layer to extract high-level features from the candidate feature matrix, learns more abstract structures (such as shape, repeating patterns, and long-range dependencies), generates more reliable high-frequency information and contextual understanding, and extracts different receptive fields (e.g., small / medium / large convolutional kernels or dilated convolutions) in parallel. These are then concatenated or weighted fusions to obtain a reference feature matrix, which includes high-level features from the standard projected image. It should be understood that in this embodiment, the NAF module layer includes four NAF-Block modules, which are the core modules in the NAFNet model.

[0133] For example, in some embodiments, high-level features are extracted from the candidate feature matrix to obtain a reference feature matrix, specifically including but not limited to the following steps S4321-S4323:

[0134] S4321: Input the candidate feature matrix into the first-level high-level feature extraction layer to extract high-level features, and obtain the first-level output feature matrix of the first-level high-level feature extraction layer.

[0135] S4322: Input the first-level output feature matrix into the next-level advanced feature extraction layer to extract advanced features, and obtain the next-level output feature matrix of the next-level advanced feature extraction layer.

[0136] S4323: Repeat the process using the remaining advanced feature extraction layers: input the output feature matrix of the next level into the advanced feature extraction layer to extract advanced features, and obtain the output feature matrix of the next level of the advanced feature extraction layer, until the last advanced feature extraction layer extracts the advanced features of the baseline feature matrix, and obtains the reference feature matrix.

[0137] In this embodiment, the baseline feature matrix is ​​the output feature matrix of the previous high-level feature extraction layer above the last high-level feature extraction layer.

[0138] In this embodiment, the super-resolution neural network model includes multiple cascaded high-level feature extraction layers, wherein the high-level feature extraction layers are... Figure 9 The NAF-Block module shown is the core module in the NAFNet model.

[0139] Understandably, NAFNet is an image restoration model proposed by Megvii Research Institute, with the NAF-Block module as its core module. In this embodiment of the invention, the NAF-Block module includes a lightweight convolution module (Mobileconvolution module (MB-Conv)) and a feedforward network. It should be understood that the lightweight convolution module includes a Channel Attention (SCA) layer, a Simple Gate (SG) layer, and convolution operations and a LayerNorm layer, etc. The feedforward module includes a Simple Gate layer, as well as convolution operations and a LayerNorm layer, etc. The Simple Gate layer is used to replace common nonlinear activation functions (ReLU, GELU, etc.), and its advantage is that it saves the computation of nonlinear functions without affecting the network performance, resulting in faster computation speed.

[0140] Specifically, the channel attention layer compresses the spatial dimension of the feature map to 1 using global average pooling, and then performs a dot product operation with the original feature map to achieve inter-channel interaction. Simple gating splits the weighted feature map into two parts along the channel dimension and multiplies them to obtain the gated feature map. Then, depthwise convolution and pointwise convolution are performed on the gated feature map to extract local features and adjust the channel dimensions. Finally, a LayerNorm layer makes the model training more stable and outputs more accurate features.

[0141] For example, please see Figure 10The system comprises multiple cascaded high-level feature extraction layers, including a first-level NAF-Block module, a second-level NAF-Block module, a third-level NAF-Block module (the layer preceding the last high-level feature extraction layer), and a fourth-level NAF-Block module (the last high-level feature extraction layer). The candidate feature matrix is ​​input into the first-level high-level feature extraction layer (the first-level NAF-Block module) to extract high-level features, resulting in the first-level output feature matrix. This first-level output feature matrix is ​​then input into the next-level high-level feature extraction layer (the second-level NAF-Block module) to extract further high-level features, resulting in the next-level output feature matrix (the second-level output feature matrix). The process is repeated using the remaining high-level feature extraction layers: the output feature matrix of the next level is input into the next-level high-level feature extraction layer to extract high-level features, resulting in the next-level output feature matrix of the next-level high-level feature extraction layer, until the last high-level feature extraction layer extracts the high-level features of the baseline feature matrix, resulting in the reference feature matrix. That is: the output feature matrix of the second level is input into the third-level NAF-Block module to extract high-level features, resulting in the third-level output feature matrix of the third-level NAF-Block module (i.e., the baseline feature matrix), and the third-level output feature matrix is ​​input into the fourth-level NAF-Block module to extract high-level features, resulting in the fourth-level output feature matrix of the fourth-level NAF-Block module (i.e., the reference feature matrix).

[0142] S433: Upsample the high-level features in the reference feature matrix to obtain the reference image.

[0143] For example, please refer to Figure 9 The super-resolution neural network model uses a second convolutional layer to pass the reference feature matrix to a pixel reassembly layer. The pixel reassembly layer upsamples the high-level features in the reference feature matrix. The upsampled result is then added to the interpolated low-level input via residual connections, preserving low-frequency information and learning only high-frequency residuals to obtain the final reference image. Bicubic upsampling or the Lanczos algorithm can be used for upsampling to map semantically rich high-level features to a high-resolution image, resulting in the reference image. The pixel reassembly layer employs pixel shuffle technology for upsampling. Pixel shuffle technology is computationally more efficient, requires fewer parameters, and effectively avoids the checkerboard artifacts that may be introduced by traditional upsampling methods (such as transposed convolution and deconvolution).

[0144] In some embodiments, an upsampling operation is performed on the high-level features in the reference feature matrix to obtain a reference image, specifically including but not limited to the following steps S4331-S4332:

[0145] S4331: Determine the scaling factor.

[0146] S4332: Based on the scaling factor, perform upsampling on the high-level features in the reference feature matrix to obtain the reference image.

[0147] In this embodiment, the scaling factor is the magnification factor of the standard projected image during super-resolution, that is, the magnification factor of the standard projected image. The scaling factor can be 2, 4, 8, etc., which means that the standard projected image is magnified by 2 times, 4 times, 8 times, etc. Specifically, according to the scaling factor, any suitable algorithm is selected to perform upsampling operations on the high-level features in the reference feature matrix to obtain the reference image.

[0148] S434: Obtain the super-resolution image based on the reference image and the standard projection image.

[0149] For example, please refer to Figure 9 The standard projection image is scaled and cropped to obtain a reference image with the same size as the reference image. Then, the super-resolution neural network model adds and fuses the reference image and the reference image pixel by pixel to obtain the super-resolution image.

[0150] In some embodiments, a super-resolution image is obtained based on a reference image and a standard projected image, specifically including but not limited to the following steps S4341-S4342:

[0151] S4341: Upsample the standard projected image to obtain a reference image.

[0152] S4342: The reference image and the base image are fused to obtain a super-resolution image.

[0153] In this embodiment, after obtaining the reference image (generated from upsampled high-level features), to ensure that the output super-resolution image has both high-resolution details and maintains consistency with the projection intensity distribution of the standard projection image, it is necessary to upsample the standard projection image. The image obtained by upsampling the standard projection image is then fused with the reference image to obtain the final super-resolution image. The reference image provides high-frequency detail information, while the baseline image provides low-frequency projection intensity and overall structure. The fusion of the two yields a super-resolution image that is both detailed and realistic.

[0154] Specifically, an interpolation algorithm (such as bilinear, bicubic, Lanczos, etc.) is used to upsample the standard projection image at a preset magnification factor (such as 2x, 4x, etc.) to obtain a reference image. The reference image has the same size as the reference image, i.e., the same resolution, and is used for subsequent pixel-level fusion with the reference image. The reference image retains the low-frequency components and grayscale distribution of the original projection, serving as a brightness and energy benchmark during fusion to prevent physical deviations or artifacts caused by high-frequency enhancement.

[0155] Specifically, the upsampled baseline image and reference image are added and fused pixel by pixel to obtain a super-resolution image. The fused super-resolution image has high-resolution detailed structure while maintaining the grayscale consistency and energy conservation properties of the original projection, which is used for subsequent geometric reconstruction of the super-resolution image.

[0156] In this embodiment, the overall grayscale and structural consistency of the standard projection image can be preserved through the above-described upsampling and image fusion process, while introducing high-frequency feature details from the reference image. This significantly improves image clarity, prevents blurring or edge loss caused by simple upsampling, avoids false image detail problems caused by directly using the reference image output by the super-resolution neural network model, and ensures the physical reliability of the image results.

[0157] In summary, the embodiments of the present invention obtain a standard projection image by removing air and normalizing the initial projection image. The standard projection image is then input into a super-resolution neural network model to obtain a super-resolution image, thereby improving the noise removal capability of the super-resolution neural network model and effectively filtering out noise. This reduces the impact of super-resolution on image details. Based on the super-resolution image, the geometric parameters of the reconstructed super-resolution image are calculated, and the super-resolution image is reconstructed according to the geometric parameters to obtain a super-resolution reconstructed image. This allows the reconstructed super-resolution image to restore more detailed and realistic image details, reducing artifact problems in the reconstructed image and achieving excellent super-resolution results.

[0158] This invention provides a computer-readable storage medium storing processor-executable computer program instructions. When executed by a processor, the computer program instructions cause the computer to perform the projection image super-resolution reconstruction method provided in this invention, or to perform the steps in any possible implementation of the projection image super-resolution reconstruction method provided in this invention.

[0159] In some embodiments, the storage medium may be a flash memory, a hard disk, an optical disk, a register, a magnetic surface memory, a removable disk, a CD-ROM, a random access memory (RAM), a read-only memory (ROM), an electrically programmable ROM, and an electrically erasable programmable ROM, or any other form of storage medium known in the art, or various devices including one or any combination of the above storage media.

[0160] In some embodiments, computer program instructions may take the form of programs, software, software modules, scripts, or code, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and may be deployed in any form, including as stand-alone programs or as modules, components, subroutines, or other units suitable for use in a computing environment.

[0161] As an example, computer program instructions may, but do not necessarily, correspond to files in a file system, and may be stored as part of a file that holds other programs or data, for example, in one or more scripts in an HTML (Hypertext Markup Language) document, or in a single file dedicated to the program in question, or in multiple collaborative files (e.g., a file that stores one or more modules, subroutines, or code sections).

[0162] As an example, computer program instructions can be deployed to execute on a single computing device (including devices such as smart terminals and servers), or on multiple computing devices located in one location, or on multiple computing devices distributed across multiple locations and interconnected via a communication network. It is readily understood that all or part of the steps of the methods described in the embodiments of the present invention above can be implemented directly using electronic hardware or processor-executable computer program instructions, or a combination of both.

[0163] Those skilled in the art will understand that the embodiments provided by this invention are merely illustrative. The order in which the steps in the methods of the embodiments are written does not imply a strict execution order and does not constitute any limitation on the implementation process. The order can be adjusted, merged, and deleted according to actual needs. Modules or sub-modules, units or sub-units in the apparatus or system of the embodiments can be merged, divided, and deleted according to actual needs. For example, the division of units is only a logical functional division, and there may be other division methods in actual implementation. For another example, multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed.

[0164] Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a general-purpose hardware platform, and of course, it can also be implemented using hardware. Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This computer program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. It should be understood that the storage medium can be flash memory, hard disk, optical disk, register, magnetic surface memory, removable disk, CD-ROM, random access memory (RAM), read-only memory (ROM), electrically programmable ROM, and electrically erasable programmable ROM, etc.

[0165] It should be noted that the above embodiments are for illustrating the technical concept and features of the present invention, and are intended to enable those skilled in the art to understand the content of the present invention and implement it accordingly. They should not be construed as limiting the scope of protection of the present invention. Those skilled in the art can understand that all or part of the processes of the above embodiments can be implemented by modifying the technical solutions described in the embodiments of the present invention, or by making equivalent substitutions for some of the technical features. It is understood that these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and should be considered as equivalent changes and modifications made based on the embodiments of the present invention, all of which should fall within the scope of the claims of the present invention.

Claims

1. A method of reconstructing a projected image super-resolution, applied to an imaging system, characterized in that, The imaging system comprises a ray source and a detector, and the method comprises: acquiring an initial projection image of a target object to be super-resolution reconstructed by the detector, the initial projection image comprising a projection pattern projected by the target object; performing air removal and normalization on the initial projection image based on an air projection image to obtain a standard projection image, the air projection image being a projection image captured by the detector when no object is arranged between the ray source and the detector, the capturing condition of the air projection image being the same as that of the initial projection image, the capturing condition comprising illumination intensity and a second distance between the detector and the ray source; inputting the standard projection image into a super-resolution neural network model to obtain a super-resolution image, the super-resolution neural network model being a model obtained by preconstructing a neural network model using sample image data and training the neural network model until a preset condition is met, the super-resolution neural network model being configured to perform super-resolution on a projection image; calculating geometric parameters related to the super-resolution image based on the super-resolution image, the geometric parameters comprising a first distance between the target object and the ray source and the second distance between the detector and the ray source; reconstructing the super-resolution image based on the geometric parameters to obtain a target image, the target image being a super-resolution reconstructed image of the initial projection image.

2. The method of claim 1, wherein, The inputting of the standard projection image into the super-resolution neural network model to obtain the super-resolution image comprises: extracting primary features in the standard projection image to obtain a candidate feature matrix, the candidate feature matrix comprising the primary features of the standard projection image; extracting high-level features in the candidate feature matrix to obtain a reference feature matrix, the reference feature matrix comprising the high-level features of the standard projection image; performing an up-sampling operation on the high-level features in the reference feature matrix to obtain a reference image; obtaining the super-resolution image based on the reference image and the standard projection image.

3. The method of claim 2, wherein, The super-resolution neural network model comprises a primary feature extraction layer, and the extracting of the primary features in the standard projection image to obtain the candidate feature matrix comprises: extracting the primary features in the standard projection image using a plurality of convolution kernels of the primary feature extraction layer to obtain a plurality of groups of candidate features, the candidate features being the primary features of the standard projection image; splicing the plurality of groups of candidate features to obtain the candidate feature matrix.

4. The method of claim 2, wherein, The super-resolution neural network model comprises a plurality of cascaded high-level feature extraction layers, and the extracting of the high-level features in the candidate feature matrix to obtain the reference feature matrix comprises: inputting the candidate feature matrix into a first high-level feature extraction layer to extract high-level features to obtain a first output feature matrix of the first high-level feature extraction layer; inputting the first output feature matrix into a next high-level feature extraction layer to extract high-level features to obtain a next output feature matrix of the next high-level feature extraction layer; repeatedly performing, using the remaining high-level feature extraction layers, the inputting of the next output feature matrix into the next high-level feature extraction layer to extract high-level features to obtain a next output feature matrix of the next high-level feature extraction layer until the last high-level feature extraction layer extracts high-level features of a reference feature matrix to obtain the reference feature matrix, the reference feature matrix being an output feature matrix of a previous high-level feature extraction layer of the last high-level feature extraction layer.

5. The method according to any one of claims 2-4, characterized in that, The up-sampling operation is performed on the high-level features in the reference feature matrix to obtain a reference image, and the up-sampling operation comprises: determining a scale factor, the scale factor being a magnification of the standard projection image for super-resolution; performing the up-sampling operation on the high-level features in the reference feature matrix according to the scale factor to obtain the reference image.

6. The method according to any one of claims 2-4, characterized in that, The super-resolution image is obtained based on the reference image and the standard projection image, and the obtaining comprises: performing the up-sampling operation on the standard projection image to obtain a benchmark image, the benchmark image having the same size as the reference image; fusing the benchmark image and the reference image to obtain the super-resolution image.

7. The method of claim 1, wherein, The standard projection image is obtained by performing the de-airing and normalization processing on the initial projection image, and the obtaining comprises: obtaining an air projection image; obtaining a candidate projection image based on the air projection image and the initial projection image; normalizing the gray value of the pixel in the candidate projection image to obtain the standard projection image.

8. The method of claim 7, wherein, The candidate projection image is obtained based on the air projection image and the initial projection image, and the obtaining comprises: adding the gray value of the pixel at the candidate position in the air projection image to a first preset value to obtain a first gray value of the pixel at the candidate position; adding the gray value of the pixel at the candidate position in the initial projection image to a second preset value to obtain a second gray value of the pixel at the candidate position, the candidate position being any one of the positions in the air projection image and the initial projection image; calculating a target gray value of the pixel at the candidate position based on the first gray value and the second gray value; combining the target gray values of the pixels at all positions to obtain the candidate projection image according to the coordinates of the positions in the air projection image or the initial projection image.

9. An imaging system characterized by, comprise: a controller, and a ray source and a detector in communication connection with the controller; the controller comprises: a processor and a memory in communication connection with the processor; the memory stores computer program instructions executable by the processor, and the computer program instructions, when executed by the processor, cause the controller to perform the projection image super-resolution reconstruction method according to any one of claims 1-8.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores computer program instructions executable by the processor, and the computer program instructions, when executed by the processor, cause the computer to perform the projection image super-resolution reconstruction method according to any one of claims 1-8.

Images