X-ray image quality evaluation method and device, computer equipment and storage medium

Abstract

The invention relates to an X-ray image quality evaluation method and device, computer equipment and a storage medium. Degradation factor simulation operation and degradation factor combination operation are carried out on original X-ray images, degraded X-ray images corresponding to each degradation factor combination mode and category labels corresponding to the degraded X-ray images are obtained to form a training image set, and category label information of the degraded X-ray images can be reflected. According to the method, features of multiple convolution stages are obtained through processing in a convolutional neural network model, feature saliency learning is carried out, multi-scale saliency features are obtained, local, channel and global features can be better captured, category label information of each distorted image can be identified through the trained convolutional neural network model, and the classification label information of each distorted image can be identified through the trained convolutional neural network model. Therefore, the method is suitable for monitoring the image transmission quality of the security check system, intelligently evaluating the quality and querying the image degradation reason, reduces the participation of technicians to a certain extent, and improves the error correction capability of distorted images.

Description

Technical Field

[0001] The present application relates to the technical field of X-ray image comparison, and in particular to an X-ray image quality assessment method, apparatus, computer equipment, and storage medium. Background Art

[0002] Today, X-ray imaging-based baggage inspection has become the standard for ensuring the safety of transportation and other public places. X-ray imaging systems enable security personnel to identify prohibited items in luggage.

[0003] In intelligent security inspection systems, the system needs to transmit captured data to the backend for processing using detection algorithms to identify prohibited items in luggage, and then return the detection results to the front-end interface. In intelligent security inspection systems, during the formation, transmission, and recording of X-ray images, imperfections in the imaging system, transmission media, and equipment can degrade the quality of the X-ray images, thereby affecting the detection and display of test results. There are many reasons for the degradation of X-ray images, including defocusing of the imaging system, diffraction of the optical system, atmospheric disturbances, motion blur, distortion, sensor noise, and random noise. These factors can degrade the quality of X-ray images during image formation, transmission, or display.

[0004] In order to ensure the quality of image transmission and display in intelligent detection systems, it is urgent to evaluate the quality of X-ray images so as to grasp the image quality and the reasons for image quality degradation in real time, so as to quickly find the root cause of the image quality degradation problem. Summary of the Invention

[0005] Based on this, in order to address the problems of X-ray image degradation being difficult to evaluate and degradation types being difficult to find in intelligent security inspection systems, an X-ray image quality evaluation method, apparatus, computer equipment, and storage medium are provided, which can quickly evaluate the quality of X-ray images and the causes of image degradation, thereby helping intelligent security inspection systems to find the root causes of image quality degradation problems that exist during the transmission and actual processes of the imaging search system.

[0006] In one aspect, a method for evaluating X-ray image quality is provided, the method comprising:

[0007] Acquire an original X-ray image, perform a degradation factor simulation operation and a degradation factor combination operation on the original X-ray image, obtain a degraded X-ray image corresponding to each degradation factor combination and a corresponding category label of the degraded X-ray image, and form a training image set;

[0008] Inputting the training image set into a convolutional neural network model to obtain features of multiple convolution stages, performing feature saliency learning on the features of the multiple convolution stages to obtain multi-scale saliency features, fusing the multi-scale saliency features, performing fully connected feature mapping, and obtaining an output result of the convolutional neural network model;

[0009] Calculate the model loss value using a cross entropy loss function according to the category label and the output result of the convolutional neural network model, and use the convolutional neural network model with the minimum model loss value as the trained convolutional neural network model;

[0010] The X-ray image to be evaluated is input into the trained convolutional neural network model for processing to obtain the degradation category prediction result and image quality assessment score.

[0011] In one embodiment, the factor combination operation includes:

[0012] Set the degradation type and degradation level of each degradation type for degradation treatment;

[0013] A degradation mode combination library is formed according to the combined degradation modes of all degradation types;

[0014] The degradation factor simulation operation includes:

[0015] Performing a degradation factor simulation operation on the original X-ray image according to each degradation mode in the degradation mode combination library to form a degraded X-ray picture;

[0016] Constructing a category label calculation formula according to the degradation type and the degradation level of each degradation type, and calculating and obtaining a category label corresponding to the degraded X-ray image according to the category label calculation formula;

[0017] All degraded X-ray images and their corresponding category labels are aggregated to form a training image set.

[0018] In one embodiment, constructing a category label calculation formula based on the degradation type and the degradation level of each degradation type includes:

[0019] Set the degradation level of the degradation type to n0, n1, ..., n i-1 ; where n i-1 represents the degradation level of the i-th degradation type;

[0020] If the degradation level of each degradation type is M, then n0, n1, ..., n i-1 ∈[0,M-1];

[0021] The constructed category label calculation formula is: class label =(M)0 n0+(M) 1 n1+...+(M) i-1 n i-1 .

[0022] In one embodiment, constructing a category label calculation formula based on the degradation type and the degradation level of each degradation type includes:

[0023] If the number of the degradation types is three, and the degradation types include sensor noise, distortion, and image compression, setting the degradation levels of sensor noise, distortion, and image compression to be represented as a, b, and c, respectively;

[0024] If there are ten degradation levels for each degradation type, then a, b, c∈[0,9];

[0025] The constructed category label calculation formula is class label =1a+10b+100c.

[0026] In one embodiment, inputting the training image set into a convolutional neural network model, obtaining features of multiple convolution stages, performing feature saliency learning on the features of the multiple convolution stages to obtain multi-scale saliency features, fusing the multi-scale saliency features, performing fully connected feature mapping, and obtaining the output result of the convolutional neural network model includes:

[0027] Inputting each degraded X-ray image in the training image set and its corresponding category label into a convolutional neural network model to perform multi-level feature learning to obtain features of multiple convolution stages;

[0028] Inputting the features of the multiple convolution stages into a multi-scale channel segmentation attention module to perform feature saliency learning to obtain salient features;

[0029] The multi-scale salient features are integrated to extract key information at different depths and scales through self-learning;

[0030] After fusing the multi-scale salient features, a fully connected feature map is performed, and the final result is output using a softmax function to obtain the output result of the convolutional neural network model.

[0031] In one embodiment, the cross entropy loss function is used to calculate the model loss value based on the category label and the output result of the convolutional neural network model, and the convolutional neural network model with the minimum model loss value is used as the trained convolutional neural network model;

[0032] Set the softmax function to where xi is the i-th element of the input vector, x i The value of is associated with the category label, and K is the number of degradation types corresponding to the i-th element;

[0033] The output of the softmax function is used as the output result of the convolutional neural network model

[0034] Assume there are k categories of output results The category label is denoted as Y gt , then the cross entropy loss function is used to calculate the model loss value Where k represents the total number of output result categories, Y gt Indicates the true probability distribution of sample pixels, Represents the probability distribution of sample predictions by the convolutional neural network model;

[0035] The weights of the convolutional neural network model are reversely updated using a stochastic gradient descent optimization algorithm, and the weight with the smallest model loss value is taken as the optimal weight of the convolutional neural network model to obtain a trained convolutional neural network model.

[0036] In one embodiment, inputting the X-ray image to be evaluated into the trained convolutional neural network model for processing to obtain the degradation category prediction result and the image quality assessment score includes:

[0037] If the number of degradation types is N, the image quality assessment score corresponding to each category label of the X-ray image to be evaluated is:

[0038] In another aspect, an X-ray image quality assessment device is provided, comprising:

[0039] a training image set forming module, configured to obtain original X-ray images, perform degradation factor simulation operations and degradation factor combination operations on the original X-ray images, obtain degraded X-ray images corresponding to each degradation factor combination and corresponding category labels for the degraded X-ray images, and form a training image set;

[0040] A model training module is used to input the training image set into a convolutional neural network model, obtain features of multiple convolution stages, perform feature saliency learning on the features of the multiple convolution stages, obtain multi-scale saliency features, fuse the multi-scale saliency features, perform fully connected feature mapping, and obtain the output result of the convolutional neural network model;

[0041] A model screening module is used to calculate the model loss value using a cross entropy loss function based on the category label and the output result of the convolutional neural network model, and take the convolutional neural network model with the minimum model loss value as the trained convolutional neural network model;

[0042] The image evaluation module is used to input the X-ray image to be evaluated into the trained convolutional neural network model for processing to obtain the degradation category prediction result and image quality evaluation score.

[0043] In another aspect, a computer device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the following steps are implemented:

[0044] Acquire an original X-ray image, perform a degradation factor simulation operation and a degradation factor combination operation on the original X-ray image, obtain a degraded X-ray image corresponding to each degradation factor combination and a corresponding category label of the degraded X-ray image, and form a training image set;

[0045] Inputting the training image set into a convolutional neural network model to obtain features of multiple convolution stages, performing feature saliency learning on the features of the multiple convolution stages to obtain multi-scale saliency features, fusing the multi-scale saliency features, performing fully connected feature mapping, and obtaining an output result of the convolutional neural network model;

[0046] Calculate the model loss value using a cross entropy loss function according to the category label and the output result of the convolutional neural network model, and use the convolutional neural network model with the minimum model loss value as the trained convolutional neural network model;

[0047] The X-ray image to be evaluated is input into the trained convolutional neural network model for processing to obtain the degradation category prediction result and image quality assessment score.

[0048] In another aspect, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

[0049] Acquire an original X-ray image, perform a degradation factor simulation operation and a degradation factor combination operation on the original X-ray image, obtain a degraded X-ray image corresponding to each degradation factor combination and a corresponding category label of the degraded X-ray image, and form a training image set;

[0050] Inputting the training image set into a convolutional neural network model to obtain features of multiple convolution stages, performing feature saliency learning on the features of the multiple convolution stages to obtain multi-scale saliency features, fusing the multi-scale saliency features, performing fully connected feature mapping, and obtaining an output result of the convolutional neural network model;

[0051] Calculate the model loss value using a cross entropy loss function according to the category label and the output result of the convolutional neural network model, and use the convolutional neural network model with the minimum model loss value as the trained convolutional neural network model;

[0052] The X-ray image to be evaluated is input into the trained convolutional neural network model for processing to obtain the degradation category prediction result and image quality assessment score.

[0053] The above-mentioned X-ray image quality assessment method, device, computer equipment and storage medium perform degradation factor simulation operations and degradation factor combination operations on the original X-ray image to obtain degraded X-ray images corresponding to each degradation factor combination and the corresponding category labels of the degraded X-ray images to form a training image set, which can reflect the category label information of the degraded X-ray images, process the features of multiple convolution stages in the convolutional neural network model, perform feature significance learning, and obtain multi-scale significant features, which can better capture local, channel and global features. The trained convolutional neural network model can identify the category label information of each distorted image, further improving the classification performance, and is suitable for monitoring the image transmission quality of security inspection systems, intelligently assessing quality and querying the causes of image degradation, reducing the participation of technical personnel to a certain extent, and improving the error correction capability of distorted images. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0055] Figure 1 This is a flow chart of an X-ray image quality assessment method according to an embodiment of the present application;

[0056] Figure 2 This is a structural diagram of a convolutional neural network model in one embodiment of the present application;

[0057] Figure 3 This is a structural diagram of a multi-scale channel segmentation attention module in one embodiment of the present application;

[0058] Figure 4 A flowchart illustrating the steps of calculating the model loss value using the cross entropy loss function based on the category label and the output result of the convolutional neural network model in one embodiment of the present application, and using the convolutional neural network model with the minimum model loss value as the trained convolutional neural network model;

[0059] Figure 5 A flowchart illustrating the steps of inputting an X-ray image to be evaluated into a trained convolutional neural network model for processing to obtain a degradation category prediction result and an image quality assessment score in one embodiment of the present application;

[0060] Figure 6 This is a structural block diagram of an X-ray image quality assessment device in one embodiment of the present application;

[0061] Figure 7 This is a diagram of the internal structure of a computer device in one embodiment of the present application. DETAILED DESCRIPTION

[0062] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0063] As described in the background technology, most of the existing methods can only evaluate the degradation of image quality from the perspective of numerical statistics, and cannot find the cause of the degradation of image quality very well. In addition, the evaluation of image quality will have certain differences due to different evaluation functions. In actual application scenarios, it may be necessary to try various evaluation functions, and it is not possible to intelligently provide an image quality evaluation method that meets the needs of actual scenarios.

[0064] To address these issues, the present invention innovatively proposes an X-ray image quality assessment method that can rapidly assess image quality and the causes of image degradation, thereby helping security inspection systems quickly identify problems with system transmission and real-world processes. Overall, this algorithm transforms the numerical statistics problem inherent in image processing into a self-supervised feature classification problem. This intelligently assesses image quality and identifies the causes of image degradation, minimizing the need for technical personnel and improving the intelligent inspection system's ability to correct distorted images.

[0065] In one embodiment, Figure 1 As shown, a method for evaluating X-ray image quality is provided, comprising the following steps:

[0066] Step S1: obtaining an original X-ray image, performing degradation factor simulation operations and degradation factor combination operations on the original X-ray image, obtaining a degraded X-ray image corresponding to each degradation factor combination and a corresponding category label of the degraded X-ray image, and forming a training image set;

[0067] Step S2: input the training image set into the convolutional neural network model to obtain features of multiple convolution stages, perform feature saliency learning on the features of multiple convolution stages, obtain multi-scale saliency features, fuse the multi-scale saliency features, perform fully connected feature mapping, and obtain the output result of the convolutional neural network model;

[0068] Step S3, according to the category label and the output result of the convolutional neural network model, the model loss value is calculated using the cross entropy loss function, and the convolutional neural network model with the minimum model loss value is used as the trained convolutional neural network model;

[0069] In step S4, the X-ray image to be evaluated is input into the trained convolutional neural network model for processing to obtain the degradation category prediction result and image quality assessment score.

[0070] The X-ray image obtained by an X-ray machine or other imaging device is transmitted to the back-end, or the X-ray image from the back-end is transmitted to the front-end for display. The image before transmission is called the original X-ray image, and the image after transmission is called the degraded X-ray image.

[0071] Specifically, by performing degradation factor simulation operations and degradation factor combination operations on the original X-ray image, the degraded X-ray image corresponding to each degradation factor combination and the corresponding category labels of the degraded X-ray image are obtained to form a training image set, which can reflect the category label information of the degraded X-ray image. The features of multiple convolution stages are obtained in the convolutional neural network model, and feature significance learning is performed to obtain multi-scale significant features, which can better capture local, channel and global features. The trained convolutional neural network model can identify the category label information of each distorted image, further improving the classification performance, making it suitable for monitoring the image transmission quality of the security inspection system, intelligently evaluating the quality and querying the causes of image degradation, reducing the participation of technical personnel to a certain extent, and improving the error correction ability of distorted images.

[0072] In this embodiment, the factor combination operation includes:

[0073] Set the degradation type and degradation level of each degradation type for degradation treatment;

[0074] A degradation mode combination library is formed according to the combined degradation modes of all degradation types;

[0075] Degradation factor simulation operations include:

[0076] Performing a degradation factor simulation operation on the original X-ray image according to each degradation mode in the degradation mode combination library to form a degraded X-ray image;

[0077] A category label calculation formula is constructed based on the degradation type and the degradation level of each degradation type, and the category label corresponding to the degraded X-ray image is calculated according to the category label calculation formula;

[0078] All degraded X-ray images and their corresponding category labels are aggregated to form a training image set.

[0079] Specifically, such as Figure 4 As shown, the original X-ray image between transmissions is read in, denoted as X t . t The degradation factor simulation and factor combination module is sent to the degradation factor simulation and factor combination module. Different types and different levels of degradation degree of each type are set for X t Make a selection. Setting up the degradation method combination library can ensure that all degradation types of combined degradation methods can be covered, so that all degraded X-ray images can be added to the training image set. t Randomly perform a set of degradation simulation operations to obtain a simulated degradation X-ray image X s The training image set is dynamically generated during the training process based on the permutations and combinations of degradation factors to generate degraded X-ray images and their corresponding class labels.

[0080] In this embodiment, the category label calculation formula is constructed based on the degradation type and the degradation level of each degradation type, including:

[0081] Set the degradation level of the degradation type to n0, n1, ..., n i-1 ; where n i-1 represents the degradation level of the i-th degradation type;

[0082] If the degradation level of each degradation type is M, then n0, n1, ..., n i-1 ∈[0,M-1];

[0083] The constructed category label calculation formula is: class label =(M) 0 n0+(M) 1 n1+...+(M) i-1 n i-1 .

[0084] In this embodiment, the category label calculation formula is constructed based on the degradation type and the degradation level of each degradation type, including:

[0085] If the number of degradation types is three, the degradation types include sensor noise, distortion, and image compression, and the degradation levels of sensor noise, distortion, and image compression are set to be a, b, and c, respectively;

[0086] If there are ten degradation levels for each degradation type, then a, b, c∈[0,9];

[0087] The constructed category label calculation formula is class label =1a+10b+100c.

[0088] like Figure 4 As shown, when performing degradation factor simulation and combination operations in the degradation factor simulation and factor combination modules, a corresponding category label and quality evaluation score label are set for each simulated degradation factor combination distortion mode. Assume that the degradation levels of sensor noise, distortion, and image compression are represented by a, b, and c. The values of a, b, and c are 0, 1, 2, 3, 4, 5, 6, 7, 8, and 9. For example, when the category label value is 8, the degradation type can be directly determined to be sensor noise, and the degradation level of sensor noise is 8, because only sensor noise will limit the category label value to less than 10. When the category label value is 32, the degradation type is sensor noise and distortion, the degradation level of sensor noise is 2, and the degradation level of distortion is 3. When the category label value is 549, the degradation type is sensor noise, distortion, and image compression, the degradation level of sensor noise is 9, the degradation level of distortion is 4, and the degradation level of image compression is 5. Each time training is performed, a possible permutation and combination is extracted to perform image changes and annotations to complete the training.

[0089] This application sets different types and different levels of degradation for each type to choose from. In the security inspection system scenario, only three types of distortion are used here, namely sensor noise, distortion and image compression, and each type has 10 degradation levels. At the same time, these three distortion types can be freely combined, so the number of distortions that can be simulated here is 1000. If there are N types of degradation and M levels of degradation, then there are a total of N M After each degradation simulation, a set of labels will be obtained, which are class labels. label .

[0090] In this embodiment, the training image set is input into the convolutional neural network model, features of multiple convolution stages are obtained, feature saliency learning is performed on the features of multiple convolution stages, multi-scale saliency features are obtained, the multi-scale saliency features are fused, and fully connected feature mapping is performed. The output result of the convolutional neural network model is obtained, including:

[0091] Each degraded X-ray image in the training image set and its corresponding category label are input into the convolutional neural network model for multi-level feature learning to obtain features from multiple convolution stages;

[0092] The features of multiple convolutional stages are input into the multi-scale channel segmentation attention module to learn the saliency of features and obtain salient features;

[0093] The multi-scale salient features are integrated to extract key information at different depths and scales through self-learning;

[0094] After fusing the multi-scale salient features, a fully connected feature map is performed, and the final result is output using the softmax function to obtain the output result of the convolutional neural network model.

[0095] The simulated degraded X-ray image X s And the corresponding label information is sent to the deep convolutional neural network for feature learning. The specific convolutional neural network model structure is as follows Figure 2 shown.

[0096] like Figure 2 As shown, first simulate the degraded X-ray image X s The feature expressions of different depth semantic information of the four convolution stages are obtained, namely stage1∈R 64*56*56 ,stage2∈R 128*28*28 ,stage3∈R 256*14*14 ,stage4∈R 512*7*7 Here, shallow features learn detailed features such as image texture and color, while deep features learn overall information such as contours. C = 3 represents RGB, and an image with H*W*C = HW3 is input into the convolutional neural network model, generating features at multiple different convolution stages.

[0097] The four features obtained at different scales and depths (stage 1, stage 2, stage 3, and stage 4) are fed into the multi-scale channel segmentation attention module for feature saliency learning. This extracts salient features by focusing on multi-scale learning across different channels within the same feature through channel segmentation. Simultaneously, channel and spatial attention are used to self-learn saliency features from the multi-scale channel segmentation extracted features, facilitating the subdivision of different types and levels of degradation within subtle differences. The convolution stage transforms spatial features HW into channel features C.

[0098] like Figure 3 As shown, Figure 3 This is the architecture of the multi-scale channel segmentation attention module. The obtained salient features are fused through a multi-scale fusion network to self-learn and focus on extracting key information from HWs at different depths and scales. This approach primarily employs the FPN fusion concept. For multi-scale fusion, a simple and lightweight fusion is sufficient, rather than using a complex FPN fusion network module.

[0099] like Figure 4 As shown, in this embodiment, according to the category label and the output result of the convolutional neural network model, the cross entropy loss function is used to calculate the model loss value, and the convolutional neural network model with the minimum model loss value is used as the trained convolutional neural network model, including:

[0100] Set the softmax function to where x i is the i-th element of the input vector, x i The value of is associated with the category label, and K is the number of degradation types corresponding to the i-th element;

[0101] The output of the softmax function is used as the output of the convolutional neural network model

[0102] Assume there are k categories of output results The class label is represented as Y gt , then the cross entropy loss function is used to calculate the model loss value Where k represents the total number of output result categories, Y gt Indicates the true probability distribution of sample pixels, Represents the probability distribution of sample predictions by the convolutional neural network model;

[0103] The stochastic gradient descent optimization algorithm (SGD) is used to reversely update the weights of the convolutional neural network model. The weight with the smallest model loss value is taken as the optimal weight of the convolutional neural network model to obtain the trained convolutional neural network model.

[0104] In this embodiment, the X-ray image to be evaluated is input into a trained convolutional neural network model for processing, and the degradation category prediction result and image quality assessment score are obtained, including:

[0105] If the number of degradation types is N, the image quality assessment score corresponding to each category label of the X-ray image to be evaluated is:

[0106] It can be understood that if there are three degradation types, including sensor noise, distortion, and image compression, and the degradation levels of sensor noise, distortion, and image compression are set as a, b, and c respectively, the image quality assessment score corresponding to each category label is: This calculation method can help understand the degree of image quality degradation.

[0107] like Figure 5 As shown, the X-ray image to be evaluated Xv is input into the trained convolutional neural network model for processing, and the degradation category prediction result Y is output after being processed by the softmax function. pre , and display the image quality assessment score class sorce .

[0108] In the above-mentioned X-ray image quality assessment method, by performing degradation factor simulation operations and degradation factor combination operations on the original X-ray image, the degraded X-ray image corresponding to each degradation factor combination and the corresponding category label of the degraded X-ray image are obtained to form a training image set, which can reflect the category label information of the degraded X-ray image. The features of multiple convolution stages are processed in the convolutional neural network model, and feature significance learning is performed to obtain multi-scale significant features, which can better capture local, channel and global features. The trained convolutional neural network model can identify the category label information of each distorted image, further improving the classification performance, and thus being suitable for monitoring the image transmission quality of the security inspection system, intelligently assessing the quality and querying the cause of image degradation, reducing the participation of technical personnel to a certain extent, and improving the error correction capability of distorted images.

[0109] In one embodiment, Figure 6 As shown, an X-ray image quality assessment device 10 is provided, comprising: a training image set forming module 1, a model training module 2, a model screening module 3, and an image assessment module 4.

[0110] The training image set forming module 1 is used to obtain the original X-ray image, perform degradation factor simulation operation and degradation factor combination operation on the original X-ray image, obtain the degraded X-ray image corresponding to each degradation factor combination and the corresponding category label of the degraded X-ray image, and form a training image set.

[0111] The model training module 2 is used to input the training image set into the convolutional neural network model, obtain the features of multiple convolution stages, perform feature saliency learning on the features of multiple convolution stages, obtain multi-scale saliency features, fuse the multi-scale saliency features, perform fully connected feature mapping, and obtain the output results of the convolutional neural network model.

[0112] The model screening module 3 is used to calculate the model loss value using the cross entropy loss function according to the category label and the output results of the convolutional neural network model, and take the convolutional neural network model with the smallest model loss value as the trained convolutional neural network model.

[0113] The image evaluation module 4 is used to input the X-ray image to be evaluated into the trained convolutional neural network model for processing to obtain the degradation category prediction result and image quality evaluation score.

[0114] In this embodiment, the factor combination operation includes:

[0115] Set the degradation type and degradation level of each degradation type for degradation treatment;

[0116] A degradation mode combination library is formed according to the combined degradation modes of all degradation types;

[0117] Degradation factor simulation operations include:

[0118] Performing a degradation factor simulation operation on the original X-ray image according to each degradation mode in the degradation mode combination library to form a degraded X-ray image;

[0119] A category label calculation formula is constructed based on the degradation type and the degradation level of each degradation type, and the category label corresponding to the degraded X-ray image is calculated according to the category label calculation formula;

[0120] All degraded X-ray images and their corresponding category labels are aggregated to form a training image set.

[0121] In this embodiment, the category label calculation formula is constructed based on the degradation type and the degradation level of each degradation type, including:

[0122] Set the degradation level of the degradation type to n0, n1, ..., n i-1 ; where n i-1 represents the degradation level of the i-th degradation type;

[0123] If the degradation level of each degradation type is M, then n0, n1, ..., n i-1 ∈[0,M-1];

[0124] The constructed category label calculation formula is: class label =(M) 0n0+(M) 1 n1+...+(M) i-1 n i-1 .

[0125] In this embodiment, the category label calculation formula is constructed based on the degradation type and the degradation level of each degradation type, including:

[0126] If the number of degradation types is three, the degradation types include sensor noise, distortion, and image compression, and the degradation levels of sensor noise, distortion, and image compression are set to be a, b, and c, respectively;

[0127] If there are ten degradation levels for each degradation type, then a, b, c∈[0,9];

[0128] The constructed category label calculation formula is class label =1a+10b+100c.

[0129] In this embodiment, the training image set is input into the convolutional neural network model, features of multiple convolution stages are obtained, feature saliency learning is performed on the features of multiple convolution stages, multi-scale saliency features are obtained, the multi-scale saliency features are fused, and fully connected feature mapping is performed. The output result of the convolutional neural network model is obtained, including:

[0130] Each degraded X-ray image in the training image set and its corresponding category label are input into the convolutional neural network model for multi-level feature learning to obtain features from multiple convolution stages;

[0131] The features of multiple convolutional stages are input into the multi-scale channel segmentation attention module to learn the saliency of features and obtain salient features;

[0132] The multi-scale salient features are integrated to extract key information at different depths and scales through self-learning;

[0133] After fusing the multi-scale salient features, a fully connected feature map is performed, and the final result is output using the softmax function to obtain the output result of the convolutional neural network model.

[0134] In this embodiment, the cross entropy loss function is used to calculate the model loss value according to the category label and the output result of the convolutional neural network model, and the convolutional neural network model with the minimum model loss value is used as the trained convolutional neural network model;

[0135] Set the softmax function to where x i is the i-th element of the input vector, x iThe value of is associated with the category label, and K is the number of degradation types corresponding to the i-th element;

[0136] The output of the softmax function is used as the output of the convolutional neural network model

[0137] Assume there are k categories of output results The class label is represented as Y gt , then the cross entropy loss function is used to calculate the model loss value Where k represents the total number of output result categories, Y gt Indicates the true probability distribution of sample pixels, Represents the probability distribution of sample predictions by the convolutional neural network model;

[0138] The stochastic gradient descent optimization algorithm (SGD) is used to reversely update the weights of the convolutional neural network model. The weight with the smallest model loss value is taken as the optimal weight of the convolutional neural network model to obtain the trained convolutional neural network model.

[0139] In this embodiment, the X-ray image to be evaluated is input into a trained convolutional neural network model for processing, and the degradation category prediction result and image quality assessment score are obtained, including:

[0140] If the number of degradation types is N, the image quality assessment score corresponding to each category label of the X-ray image to be evaluated is:

[0141] In the above-mentioned X-ray image quality assessment device, by performing degradation factor simulation operations and degradation factor combination operations on the original X-ray image, degraded X-ray images corresponding to each degradation factor combination method and the corresponding category labels of the degraded X-ray images are obtained to form a training image set, which can reflect the category label information of the degraded X-ray images. The features of multiple convolution stages are processed in the convolutional neural network model, and feature significance learning is performed to obtain multi-scale significant features, which can better capture local, channel and global features. The trained convolutional neural network model can identify the category label information of each distorted image, further improving the classification performance, and thus being suitable for monitoring the image transmission quality of the security inspection system, intelligently assessing the quality and querying the causes of image degradation, reducing the participation of technical personnel to a certain extent, and improving the error correction capability of distorted images.

[0142] The specific definitions of the X-ray image quality assessment device can be found in the definitions of the X-ray image quality assessment method above and will not be repeated here. Each module in the aforementioned X-ray image quality assessment device may be implemented in whole or in part via software, hardware, or a combination thereof. Each of these modules may be embedded in or independent of a processor within a computer device in hardware form, or may be stored in a computer device memory in software form, allowing the processor to call and execute the corresponding operations of each module.

[0143] In one embodiment, a computer program product is provided, comprising a computer program, which, when executed by a processor, implements the following steps:

[0144] Obtaining an original X-ray image, performing a degradation factor simulation operation and a degradation factor combination operation on the original X-ray image, obtaining a degraded X-ray image corresponding to each degradation factor combination and a corresponding category label of the degraded X-ray image, and forming a training image set;

[0145] Input the training image set into the convolutional neural network model to obtain features of multiple convolution stages, perform feature saliency learning on the features of multiple convolution stages, obtain multi-scale salient features, fuse the multi-scale salient features, perform fully connected feature mapping, and obtain the output results of the convolutional neural network model;

[0146] According to the category labels and the output results of the convolutional neural network model, the cross entropy loss function is used to calculate the model loss value, and the convolutional neural network model with the minimum model loss value is regarded as the trained convolutional neural network model;

[0147] The X-ray image to be evaluated is input into the trained convolutional neural network model for processing to obtain the degradation category prediction result and image quality assessment score.

[0148] In one embodiment, the computer program further performs the following steps when executed by a processor:

[0149] Factor combination operations include:

[0150] Set the degradation type and degradation level of each degradation type for degradation treatment;

[0151] A degradation mode combination library is formed according to the combined degradation modes of all degradation types;

[0152] Degradation factor simulation operations include:

[0153] Performing a degradation factor simulation operation on the original X-ray image according to each degradation mode in the degradation mode combination library to form a degraded X-ray image;

[0154] A category label calculation formula is constructed based on the degradation type and the degradation level of each degradation type, and the category label corresponding to the degraded X-ray image is calculated according to the category label calculation formula;

[0155] All degraded X-ray images and their corresponding category labels are aggregated to form a training image set.

[0156] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented:

[0157] The calculation formula for constructing the category label based on the degradation type and the degradation level of each degradation type includes:

[0158] Set the degradation level of the degradation type to n0, n1, ..., n i-1 ; where n i-1 represents the degradation level of the i-th degradation type;

[0159] If the degradation level of each degradation type is M, then n0, n1, ..., n i-1 ∈[0,M-1];

[0160] The constructed category label calculation formula is: class label =(M) 0 n0+(M) 1 n1+...+(M) i-1 n i-1 .

[0161] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented:

[0162] The calculation formula for constructing the category label based on the degradation type and the degradation level of each degradation type includes:

[0163] If the number of degradation types is three, the degradation types include sensor noise, distortion, and image compression, and the degradation levels of sensor noise, distortion, and image compression are set to be a, b, and c, respectively;

[0164] If there are ten degradation levels for each degradation type, then a, b, c∈[0,9];

[0165] The constructed category label calculation formula is class label =1a+10b+100c.

[0166] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented:

[0167] Input the training image set into the convolutional neural network model, obtain the features of multiple convolution stages, perform feature saliency learning on the features of multiple convolution stages, obtain multi-scale saliency features, fuse the multi-scale saliency features, perform fully connected feature mapping, and obtain the output results of the convolutional neural network model, including:

[0168] Each degraded X-ray image in the training image set and its corresponding category label are input into the convolutional neural network model for multi-level feature learning to obtain features from multiple convolution stages;

[0169] The features of multiple convolutional stages are input into the multi-scale channel segmentation attention module to learn the saliency of features and obtain salient features;

[0170] The multi-scale salient features are integrated to extract key information at different depths and scales through self-learning;

[0171] After fusing the multi-scale salient features, a fully connected feature map is performed, and the final result is output using the softmax function to obtain the output result of the convolutional neural network model.

[0172] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented:

[0173] According to the category labels and the output results of the convolutional neural network model, the cross entropy loss function is used to calculate the model loss value, and the convolutional neural network model with the minimum model loss value is regarded as the trained convolutional neural network model;

[0174] Set the softmax function to where x i is the i-th element of the input vector, x i The value of is associated with the category label, and K is the number of degradation types corresponding to the i-th element;

[0175] The output of the softmax function is used as the output of the convolutional neural network model

[0176] Assume there are k categories of output results The class label is represented as Y gt , then the cross entropy loss function is used to calculate the model loss value Where k represents the total number of output result categories, Y gt Indicates the true probability distribution of sample pixels, Represents the probability distribution of sample predictions by the convolutional neural network model;

[0177] The stochastic gradient descent optimization algorithm (SGD) is used to reversely update the weights of the convolutional neural network model. The weight with the smallest model loss value is taken as the optimal weight of the convolutional neural network model to obtain the trained convolutional neural network model.

[0178] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented:

[0179] The X-ray image to be evaluated is input into the trained convolutional neural network model for processing to obtain the degradation category prediction results and image quality assessment scores including:

[0180] If the number of degradation types is N, the image quality assessment score corresponding to each category label of the X-ray image to be evaluated is:

[0181] For the specific limitations on the steps implemented when the computer program is executed by the processor, please refer to the limitations on the method for X-ray image quality assessment above, which will not be repeated here.

[0182] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Figure 7 As shown. The computer device includes a processor, a memory, a network interface and a database connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store X-ray image quality assessment data. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, an X-ray image quality assessment method is implemented.

[0183] Those skilled in the art will understand that Figure 7 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0184] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the following steps are performed:

[0185] Obtaining an original X-ray image, performing a degradation factor simulation operation and a degradation factor combination operation on the original X-ray image, obtaining a degraded X-ray image corresponding to each degradation factor combination and a corresponding category label of the degraded X-ray image, and forming a training image set;

[0186] Input the training image set into the convolutional neural network model to obtain features of multiple convolution stages, perform feature saliency learning on the features of multiple convolution stages, obtain multi-scale salient features, fuse the multi-scale salient features, perform fully connected feature mapping, and obtain the output results of the convolutional neural network model;

[0187] According to the category labels and the output results of the convolutional neural network model, the cross entropy loss function is used to calculate the model loss value, and the convolutional neural network model with the minimum model loss value is regarded as the trained convolutional neural network model;

[0188] The X-ray image to be evaluated is input into the trained convolutional neural network model for processing to obtain the degradation category prediction result and image quality assessment score.

[0189] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0190] Factor combination operations include:

[0191] Set the degradation type and degradation level of each degradation type for degradation treatment;

[0192] A degradation mode combination library is formed according to the combined degradation modes of all degradation types;

[0193] Degradation factor simulation operations include:

[0194] Performing a degradation factor simulation operation on the original X-ray image according to each degradation mode in the degradation mode combination library to form a degraded X-ray image;

[0195] A category label calculation formula is constructed based on the degradation type and the degradation level of each degradation type, and the category label corresponding to the degraded X-ray image is calculated according to the category label calculation formula;

[0196] All degraded X-ray images and their corresponding category labels are aggregated to form a training image set.

[0197] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0198] The calculation formula for constructing the category label based on the degradation type and the degradation level of each degradation type includes:

[0199] Set the degradation level of the degradation type to n0, n1, ..., n i-1 ; where n i-1represents the degradation level of the i-th degradation type;

[0200] If the degradation level of each degradation type is M, then n0, n1, ..., n i-1 ∈[0,M-1];

[0201] The constructed category label calculation formula is: class label =(M) 0 n0+(M) 1 n1+...+(M) i-1 n i-1 .

[0202] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0203] The calculation formula for constructing the category label based on the degradation type and the degradation level of each degradation type includes:

[0204] If the number of degradation types is three, the degradation types include sensor noise, distortion, and image compression, and the degradation levels of sensor noise, distortion, and image compression are set to be a, b, and c, respectively;

[0205] If there are ten degradation levels for each degradation type, then a, b, c∈[0,9];

[0206] The constructed category label calculation formula is class label =1a+10b+100c.

[0207] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0208] Input the training image set into the convolutional neural network model, obtain the features of multiple convolution stages, perform feature saliency learning on the features of multiple convolution stages, obtain multi-scale saliency features, fuse the multi-scale saliency features, perform fully connected feature mapping, and obtain the output results of the convolutional neural network model, including:

[0209] Each degraded X-ray image in the training image set and its corresponding category label are input into the convolutional neural network model for multi-level feature learning to obtain features from multiple convolution stages;

[0210] The features of multiple convolutional stages are input into the multi-scale channel segmentation attention module to learn the saliency of features and obtain salient features;

[0211] The multi-scale salient features are integrated to extract key information at different depths and scales through self-learning;

[0212] After fusing the multi-scale salient features, a fully connected feature map is performed, and the final result is output using the softmax function to obtain the output result of the convolutional neural network model.

[0213] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0214] According to the category labels and the output results of the convolutional neural network model, the cross entropy loss function is used to calculate the model loss value, and the convolutional neural network model with the minimum model loss value is regarded as the trained convolutional neural network model;

[0215] Set the softmax function to where x i is the i-th element of the input vector, x i The value of is associated with the category label, and K is the number of degradation types corresponding to the i-th element;

[0216] The output of the softmax function is used as the output of the convolutional neural network model

[0217] Assume there are k categories of output results The class label is represented as Y gt , then the cross entropy loss function is used to calculate the model loss value Where k represents the total number of output result categories, Y gt Indicates the true probability distribution of sample pixels, Represents the probability distribution of sample predictions by the convolutional neural network model;

[0218] The stochastic gradient descent optimization algorithm (SGD) is used to reversely update the weights of the convolutional neural network model. The weight with the smallest model loss value is taken as the optimal weight of the convolutional neural network model to obtain the trained convolutional neural network model.

[0219] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0220] The X-ray image to be evaluated is input into the trained convolutional neural network model for processing to obtain the degradation category prediction results and image quality assessment scores including:

[0221] If the number of degradation types is N, the image quality assessment score corresponding to each category label of the X-ray image to be evaluated is:

[0222] For the specific limitations on the steps implemented when the processor executes the computer program, please refer to the limitations on the method for X-ray image quality assessment above, which will not be repeated here.

[0223] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the following steps are implemented:

[0224] Obtaining an original X-ray image, performing a degradation factor simulation operation and a degradation factor combination operation on the original X-ray image, obtaining a degraded X-ray image corresponding to each degradation factor combination and a corresponding category label of the degraded X-ray image, and forming a training image set;

[0225] Input the training image set into the convolutional neural network model to obtain features of multiple convolution stages, perform feature saliency learning on the features of multiple convolution stages, obtain multi-scale salient features, fuse the multi-scale salient features, perform fully connected feature mapping, and obtain the output results of the convolutional neural network model;

[0226] According to the category labels and the output results of the convolutional neural network model, the cross entropy loss function is used to calculate the model loss value, and the convolutional neural network model with the minimum model loss value is regarded as the trained convolutional neural network model;

[0227] The X-ray image to be evaluated is input into the trained convolutional neural network model for processing to obtain the degradation category prediction result and image quality assessment score.

[0228] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented:

[0229] Factor combination operations include:

[0230] Set the degradation type and degradation level of each degradation type for degradation treatment;

[0231] A degradation mode combination library is formed according to the combined degradation modes of all degradation types;

[0232] Degradation factor simulation operations include:

[0233] Performing a degradation factor simulation operation on the original X-ray image according to each degradation mode in the degradation mode combination library to form a degraded X-ray image;

[0234] A category label calculation formula is constructed based on the degradation type and the degradation level of each degradation type, and the category label corresponding to the degraded X-ray image is calculated according to the category label calculation formula;

[0235] All degraded X-ray images and their corresponding category labels are aggregated to form a training image set.

[0236] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented:

[0237] The calculation formula for constructing the category label based on the degradation type and the degradation level of each degradation type includes:

[0238] Set the degradation level of the degradation type to n0, n1, ..., n i-1 ; where n i-1 represents the degradation level of the i-th degradation type;

[0239] If the degradation level of each degradation type is M, then n0, n1, ..., n i-1 ∈[0,M-1];

[0240] The constructed category label calculation formula is: class label =(M) 0 n0+(M) 1 n1+...+(M) i-1 n i-1 .

[0241] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented:

[0242] The calculation formula for constructing the category label based on the degradation type and the degradation level of each degradation type includes:

[0243] If the number of degradation types is three, the degradation types include sensor noise, distortion, and image compression, and the degradation levels of sensor noise, distortion, and image compression are set to be a, b, and c, respectively;

[0244] If there are ten degradation levels for each degradation type, then a, b, c∈[0,9];

[0245] The constructed category label calculation formula is class label =1a+10b+100c.

[0246] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented:

[0247] Input the training image set into the convolutional neural network model, obtain the features of multiple convolution stages, perform feature saliency learning on the features of multiple convolution stages, obtain multi-scale saliency features, fuse the multi-scale saliency features, perform fully connected feature mapping, and obtain the output results of the convolutional neural network model, including:

[0248] Each degraded X-ray image in the training image set and its corresponding category label are input into the convolutional neural network model for multi-level feature learning to obtain features from multiple convolution stages;

[0249] The features of multiple convolutional stages are input into the multi-scale channel segmentation attention module to learn the saliency of features and obtain salient features;

[0250] The multi-scale salient features are integrated to extract key information at different depths and scales through self-learning;

[0251] After fusing the multi-scale salient features, a fully connected feature map is performed, and the final result is output using the softmax function to obtain the output result of the convolutional neural network model.

[0252] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented:

[0253] According to the category labels and the output results of the convolutional neural network model, the cross entropy loss function is used to calculate the model loss value, and the convolutional neural network model with the minimum model loss value is regarded as the trained convolutional neural network model;

[0254] Set the softmax function to where x i is the i-th element of the input vector, x i The value of is associated with the category label, and K is the number of degradation types corresponding to the i-th element;

[0255] The output of the softmax function is used as the output of the convolutional neural network model

[0256] Assume there are k categories of output results The class label is represented as Y gt , then the cross entropy loss function is used to calculate the model loss value Where k represents the total number of output result categories, Y gt Indicates the true probability distribution of sample pixels, Represents the probability distribution of sample predictions by the convolutional neural network model;

[0257] The stochastic gradient descent optimization algorithm (SGD) is used to reversely update the weights of the convolutional neural network model. The weight with the smallest model loss value is taken as the optimal weight of the convolutional neural network model to obtain the trained convolutional neural network model.

[0258] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented:

[0259] The X-ray image to be evaluated is input into the trained convolutional neural network model for processing to obtain the degradation category prediction results and image quality assessment scores including:

[0260] If the number of degradation types is N, the image quality assessment score corresponding to each category label of the X-ray image to be evaluated is:

[0261] For the specific limitations on the steps implemented when the computer program is executed by the processor, please refer to the limitations on the method for X-ray image quality assessment above, which will not be repeated here.

[0262] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0263] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0264] The above embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art could make various modifications and improvements without departing from the spirit of the present application, all of which fall within the scope of protection of the present application. Therefore, the scope of protection of the present patent application shall be determined by the appended claims.

Claims

1. A method for evaluating X-ray image quality, characterized in that: include: Acquire an original X-ray image, perform a degradation factor simulation operation and a degradation factor combination operation on the original X-ray image, obtain a degraded X-ray image corresponding to each degradation factor combination and a corresponding category label of the degraded X-ray image, and form a training image set; Inputting the training image set into a convolutional neural network model to obtain features of multiple convolution stages, performing feature saliency learning on the features of the multiple convolution stages to obtain multi-scale saliency features, fusing the multi-scale saliency features, performing fully connected feature mapping, and obtaining an output result of the convolutional neural network model; Calculate the model loss value using a cross entropy loss function according to the category label and the output result of the convolutional neural network model, and use the convolutional neural network model with the minimum model loss value as the trained convolutional neural network model; The X-ray image to be evaluated is input into the trained convolutional neural network model for processing to obtain the degradation category prediction result and image quality assessment score.

2. The X-ray image quality assessment method according to claim 1, characterized in that: The factor combination operation includes: Set the degradation type and degradation level of each degradation type for degradation treatment; A degradation mode combination library is formed according to the combined degradation modes of all degradation types; The degradation factor simulation operation includes: Performing a degradation factor simulation operation on the original X-ray image according to each degradation mode in the degradation mode combination library to form a degraded X-ray picture; Constructing a category label calculation formula according to the degradation type and the degradation level of each degradation type, and calculating and obtaining a category label corresponding to the degraded X-ray image according to the category label calculation formula; All degraded X-ray images and their corresponding category labels are aggregated to form a training image set.

3. The X-ray image quality assessment method according to claim 2, characterized in that: The calculation formula for constructing a category label according to the degradation type and the degradation level of each degradation type includes: Set the degradation level of the degradation type to n0, n1, ..., n i-1 ; where n i-1 represents the degradation level of the i-th degradation type; If the degradation level of each degradation type is M, then n0, n1, ..., n i-1 ∈[0,M-1]; The constructed category label calculation formula is: class label =(M) 0 n0+(M) 1 n1+...+(M) i-1 n i-1 .

4. The X-ray image quality assessment method according to claim 3, wherein: The calculation formula for constructing a category label according to the degradation type and the degradation level of each degradation type includes: If the number of the degradation types is three, and the degradation types include sensor noise, distortion, and image compression, setting the degradation levels of sensor noise, distortion, and image compression to be represented as a, b, and c, respectively; If there are ten degradation levels for each degradation type, then a, b, c∈[0,9]; The constructed category label calculation formula is class label =1a+10b+100c.

5. The X-ray image quality assessment method according to claim 1, wherein: Inputting the training image set into a convolutional neural network model, obtaining features of multiple convolution stages, performing feature saliency learning on the features of the multiple convolution stages to obtain multi-scale saliency features, fusing the multi-scale saliency features, performing fully connected feature mapping, and obtaining an output result of the convolutional neural network model includes: Inputting each degraded X-ray image in the training image set and its corresponding category label into a convolutional neural network model to perform multi-level feature learning to obtain features of multiple convolution stages; Inputting the features of the multiple convolution stages into a multi-scale channel segmentation attention module to perform feature saliency learning to obtain salient features; The multi-scale salient features are integrated to extract key information at different depths and scales through self-learning; After fusing the multi-scale salient features, a fully connected feature map is performed, and the final result is output using a softmax function to obtain the output result of the convolutional neural network model.

6. The X-ray image quality assessment method according to claim 5, characterized in that: The cross entropy loss function is used to calculate the model loss value according to the category label and the output result of the convolutional neural network model, and the convolutional neural network model with the minimum model loss value is used as the trained convolutional neural network model; Set the softmax function to where x i is the i-th element of the input vector, x i The value of is associated with the category label, and K is the number of degradation types corresponding to the i-th element; The output of the softmax function is used as the output result of the convolutional neural network model Assume there are k categories of output results The category label is denoted as Y gt , then the cross entropy loss function is used to calculate the model loss value Where k represents the total number of output result categories, Y gt Indicates the true probability distribution of sample pixels, Represents the probability distribution of sample predictions by the convolutional neural network model; The weights of the convolutional neural network model are reversely updated using a stochastic gradient descent optimization algorithm, and the weight with the smallest model loss value is taken as the optimal weight of the convolutional neural network model to obtain a trained convolutional neural network model.

7. The X-ray image quality assessment method according to claim 3, characterized in that: Inputting the X-ray image to be evaluated into the trained convolutional neural network model for processing to obtain the degradation category prediction result and image quality assessment score includes: If the number of degradation types is N, the image quality assessment score corresponding to each category label of the X-ray image to be evaluated is:

8. An X-ray image quality assessment device, characterized in that: The device comprises: a training image set forming module, configured to obtain original X-ray images, perform degradation factor simulation operations and degradation factor combination operations on the original X-ray images, obtain degraded X-ray images corresponding to each degradation factor combination and corresponding category labels for the degraded X-ray images, and form a training image set; A model training module is used to input the training image set into a convolutional neural network model, obtain features of multiple convolution stages, perform feature saliency learning on the features of the multiple convolution stages, obtain multi-scale saliency features, fuse the multi-scale saliency features, perform fully connected feature mapping, and obtain the output result of the convolutional neural network model; A model screening module is used to calculate the model loss value using a cross entropy loss function based on the category label and the output result of the convolutional neural network model, and take the convolutional neural network model with the minimum model loss value as the trained convolutional neural network model; The image evaluation module is used to input the X-ray image to be evaluated into the trained convolutional neural network model for processing to obtain the degradation category prediction result and image quality evaluation score.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images