X-ray image quality intelligent evaluation fusion model and use method

Abstract

The present application relates to the technical field of image data processing, and more particularly to an x-ray image quality intelligent evaluation fusion model and a use method thereof, comprising a data acquisition module, an unlabeled data processing module, a comprehensive quality evaluation module and a low-quality x-ray image processing module.The present application greatly improves the efficiency and accuracy of x-ray image data processing through the closed-loop processing of data acquisition, model dynamic optimization, accurate quality evaluation and low-quality image improvement by means of the comprehensive quality evaluation module and the low-quality x-ray image processing module.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of image data processing, in particular to an X-ray image quality intelligent evaluation fusion model and a use method thereof. BACKGROUND

[0002] The quality of images obtained by ray digital imaging is directly related to subsequent automatic detection and intelligent interpretation of defects, accurate and intuitive display of internal structures of power equipment, correct judgment of detection results by detection personnel, and the technical level of detection personnel, and the currently developed X-ray detection robot cannot autonomously judge whether the quality of the photographed image meets the diagnostic requirements, which seriously affects the intelligent level, and the current X-ray image quality determination technology has the following deficiencies: 1. Many businesses and lack of experts: due to the uneven shooting technology of each detection personnel, the traditional X-ray image quality determination technology needs a large number of professional and technical personnel to determine the image quality, which is inefficient and the professional and technical personnel are in short supply. 2. Low intelligent level and high labor cost: the currently developed X-ray intelligent detection robot has not realized remote control of more than 200 meters, and the shooting system does not have a shooting quality recognition function, so manual determination of image quality is required, which is time-consuming and labor-intensive, and the professional and technical personnel at the remote end need to be requested for assistance when problems occur. 3. The current X-ray image quality determination method is time-consuming and labor-intensive, and errors are prone to occur when the business volume is large, and visual fatigue problems occur when the image quality is determined manually, which affects the determination result. 4. The recognition technology of X-ray detection images in the current medical industry develops rapidly, but the structure and composition of the human body are relatively fixed, the structure and material of power equipment are complex, and laboratory detection is not possible, so on-site detection is required, the factors affecting the quality of X-ray detection images are more, and the intelligent recognition of image quality is difficult.

[0003] Chinese Patent Publication No. CN103814395B discloses a device 16 for displaying an x-ray image, the device comprising a display 20 for displaying an x-ray image, a workstation 22 for processing image data, and a user interface 24 for receiving commands from a user of the device. The user interface 24 is adapted to allow the user to select a master image 34a and a slave image 34b from a plurality of images. The workstation 22 is adapted to transform the slave image 34b by generating a color transform 40 for visually adapting the slave image to the master image based on the master image 34a and the slave image 34b and by applying the color transform 40 to the slave image 34b. The display 20 is adapted to display the transformed slave image 42. However, this scheme has serious deficiencies in terms of low accuracy of x-ray image evaluation and low quality of x-ray image repair. SUMMARY

[0004] To this end, the application provides an x-ray image quality intelligent evaluation fusion model and a use method to overcome the problems of low x-ray image evaluation accuracy and low x-ray image repair quality in the prior art.

[0005] To achieve the above-mentioned purpose, in one aspect, the application provides an x-ray image quality intelligent evaluation fusion model, comprising:

[0006] A data acquisition module is configured to acquire x-image data;

[0007] An unlabeled data processing module is configured to obtain unlabeled x-ray images according to the x-image data and the MAE processing model, to adjust the MAE processing model according to the sample size in the x-image data, and to revise the adjustment process of the MAE processing model according to the stringency index in the x-image data;

[0008] A comprehensive quality evaluation module is configured to obtain evaluated x-ray images according to the unlabeled x-ray images and the comprehensive quality evaluation model, to calculate the root mean square error and correct the comprehensive quality evaluation model according to the root mean square error, to optimize the correction process of the comprehensive quality evaluation model according to the data noise level in the x-image data, and to correct the optimization process of the correction of the comprehensive quality evaluation model according to the device acquisition x-ray image time in the x-image data;

[0009] A low-quality x-ray image processing module is configured to obtain distortion type labeled x-ray images according to the evaluated x-ray images by the discrimination model, to obtain high-quality x-ray images according to the GAN generation model and the distortion type labeled x-ray images, to obtain the high-quality x-ray images, and to send the high-quality x-ray images to an x-ray image quality intelligent evaluation terminal.

[0010] Further, the unlabeled data processing module constructs the MAE model by a MAE processing model construction method, inputs the initial x-ray images in the x-image data into the MAE processing model to obtain the unlabeled x-ray images output by the MAE processing model, and the MAE processing model construction method comprises:

[0011] Step A1, the mask autoencoder is used as the initial architecture of the MAE processing model to obtain an initial MAE processing model;

[0012] Step A2, the mask proportion pj0 of the initial MAE processing model is set, pj0 is set to 75%, and a to-be-trained MAE processing model is obtained;

[0013] Step A3, the to-be-trained MAE processing model is trained by processing the data set to obtain a trained MAE processing model, and the trained MAE processing model is output as the MAE processing model.

[0014] Further, the unlabeled data processing module compares the sample quantity Yb in the x image data with a preset sample quantity Yb0, judges the sample quantity condition according to the comparison result, and adjusts the MAE processing model according to the judgment result, wherein:

[0015] When Yb≤Yb0, the unlabeled data processing module determines that the sample quantity condition is that the sample quantity is small, and does not adjust the MAE processing model;

[0016] When Yb>Yb0, the unlabeled data processing module determines that the sample quantity condition is that the sample quantity is large, adjusts the MAE processing model, adjusts the mask rate pj0 through the adjustment coefficient Tj, Tj=(1-Yb0 / Yb), obtains the adjusted mask rate pj0`, sets pj0`=Tj×pj0, replaces the mask rate pj0 with the adjusted mask rate pj0`, and re-trains the MAE model according to the mask rate pj0.

[0017] Further, the unlabeled data processing module compares the sample quantity Yb in the x image data with a preset sample quantity Yb0, judges the sample quantity condition according to the comparison result, and adjusts the MAE processing model according to the judgment result, wherein:

[0018] When Yb≤Yb0, the unlabeled data processing module determines that the sample quantity condition is that the sample quantity is small, and does not adjust the MAE processing model;

[0019] When Yb>Yb0, the unlabeled data processing module determines that the sample quantity condition is that the sample quantity is large, adjusts the MAE processing model, adjusts the mask rate pj0 through the adjustment coefficient Tj, Tj=(1-Yb0 / Yb), obtains the adjusted mask rate pj0`, sets pj0`=Tj×pj0, replaces the mask rate pj0 with the adjusted mask rate pj0`, and re-trains the MAE model according to the mask rate pj0.

[0020] Further, the comprehensive quality evaluation module constructs the comprehensive quality evaluation model through a comprehensive quality evaluation model construction method, obtains the comprehensive quality evaluation model, and inputs the unlabeled x-ray image into the comprehensive quality evaluation model to obtain the evaluated x-ray image output by the comprehensive quality evaluation model, wherein the evaluated x-ray image includes a comprehensive quality evaluation score The comprehensive quality evaluation model construction method comprises:

[0021] The evaluation dataset is divided into a 70% training set, a 20% validation set, and a 10% test set. The training set is input into the ResNet50 model for training. The validation set is input into the trained ResNet50 model for iterative hyperparameter optimization. The test set is input into the optimized ResNet50 model for evaluation testing. The total number of test samples is set as f0, the number of correctly evaluated test samples is f, and the evaluation accuracy is F, where F = f / f0. The evaluation accuracy F is compared with the preset evaluation accuracy F0. F0 ≥ 9. Based on the comparison results, the training performance of the optimized ResNet50 model is judged, and the judgment result is output.

[0022] When F≥F0, the comprehensive quality assessment module determines that the training of the iteratively optimized ResNet50 model has reached the target, and outputs the iteratively optimized ResNet50 model as the comprehensive quality assessment model.

[0023] When F < F0, the comprehensive quality assessment module determines that the training of the iteratively optimized ResNet50 model is substandard, updates the assessment dataset to obtain the updated assessment dataset, and trains, iteratively optimizes hyperparameters, and analyzes and tests the ResNet50 model based on the updated assessment dataset until the ResNet50 model training meets the standards.

[0024] Furthermore, the comprehensive quality assessment module bases its assessment on the comprehensive quality assessment score. Data volume and true X-ray image quality score For root mean square error Perform calculations and set Root mean square error The model is compared with the preset root mean square error RMSE0, where 0 ≤ RMSE0 ≤ 0.5. The model performance is assessed based on the comparison results, and the overall quality assessment model is then revised accordingly.

[0025] when When ≤RMSE0, the comprehensive quality assessment module determines that the model performance is qualified and does not make any corrections to the comprehensive quality assessment model;

[0026] when When RMSE0, the comprehensive quality evaluation module determines that the model performance is unqualified, modifies the comprehensive quality evaluation model, modifies the preset evaluation test accuracy F0 through a modification coefficient Czx, sets Czx = 1.04-0.03xe -（RMSE-RMSE0） , obtains a modified preset evaluation test accuracy F01, sets F01 = F0xCzx, replaces the preset evaluation test accuracy F0 with the modified preset evaluation test accuracy F01, and recompares the evaluation test accuracy F with the preset evaluation test accuracy F0.

[0027] Further, the comprehensive quality evaluation module compares the data noise level τ in the x image data with a preset data noise level τ0, judges the data noise condition according to the comparison result, and optimizes the process of modifying the model according to the judgment result, wherein:

[0028] When τ≤τ0, the comprehensive quality evaluation module determines that the data noise condition is low noise level, and does not optimize the process of modifying the model;

[0029] When τ>τ0, the comprehensive quality evaluation module determines that the data noise condition is high noise level, and optimizes the process of modifying the model, optimizes the preset root mean square error RMSE0 through a noise coefficient qbz, sets qbz = 0.75+0.2xe -0.7×（τ-τ0） , obtains an optimized preset root mean square error RMSE0`, sets RMSE0` = RMSE0Xqbz, replaces the preset root mean square error RMSE0 with the optimized preset root mean square error RMSE0`, and recompares the root mean square error RMSE with the preset root mean square error RMSE0.

[0030] Further, the comprehensive quality evaluation module compares the device acquisition x-ray image time Ts in the x image data with a preset device acquisition x-ray image time Ts0, judges the acquisition time condition according to the comparison result, and corrects the optimization process of modifying the comprehensive quality evaluation model according to the judgment result, wherein:

[0031] When Ts≤Ts0, the comprehensive quality evaluation module determines that the acquisition time condition is time shortage, and does not correct the optimization process of modifying the model;

[0032] When Ts>Ts0, the comprehensive quality evaluation module determines that the acquisition time condition is time excess, corrects the optimization process of model correction, corrects the preset data noise level τ0 through a correction coefficient jz, sets jz=Ts0 / Ts, obtains the corrected preset data noise level τ0`, sets τ0`=τ×jz, replaces the preset data noise level τ0 with the corrected preset data noise level τ0`, and compares the data noise level τ with the preset data noise level τ0 again.

[0033] Further, the comprehensive quality evaluation module compares the comprehensive quality evaluation score yz of the x-ray image with a preset comprehensive quality evaluation score yz0, judges the comprehensive quality evaluation score condition of the x-ray image according to the comparison result, and outputs the image property of the evaluated x-ray image according to the judgment result, wherein:

[0034] When yz <yz0, the comprehensive quality evaluation module determines that the comprehensive quality evaluation score condition of the x-ray image is low score, outputs the low-quality x-ray image as the image property of the evaluated x-ray image, and sends the evaluated x-ray image to the low-quality x-ray image processing module.

[0035] When yz >yz0, the comprehensive quality evaluation module determines that the comprehensive quality evaluation score condition of the x-ray image is high score, outputs the high-quality x-ray image as the property of the evaluated x-ray image, and sends the evaluated x-ray image as the learning data set to the unlabeled data processing module, and re-trains the MAE model according to the learning data.

[0036] On the other hand, the application also provides a use method of the x-ray image quality intelligent evaluation fusion model, comprising:

[0037] Step S1, acquiring x image data through a data acquisition module;

[0038] Step S2, acquiring an unlabeled x-ray image through an unlabeled data processing module according to the x image data and the MAE processing model, adjusting the MAE processing model according to the sample amount in the x image data, and revising the adjustment process of the MAE processing model according to the stringency index in the x image data;

[0039] ​In step S3, the comprehensive quality evaluation module is used to obtain the evaluated x-ray image according to the unmarked x-ray image and the comprehensive quality evaluation model, the root mean square error is calculated, the comprehensive quality evaluation model is corrected according to the root mean square error, the correction process of the comprehensive quality evaluation model is optimized according to the data noise level in the x image data, and the optimization process of the correction of the comprehensive quality evaluation model is corrected according to the device acquisition x-ray image time in the x image data.

[0040] In step S4, the low-quality x-ray image processing module is used to obtain the distortion type labeled x-ray image according to the evaluated x-ray image through the discrimination model, and obtain the high-quality x-ray image according to the GAN generation model and the distortion type labeled x-ray image, obtain the high-quality x-ray image, and send the high-quality x-ray image to the x-ray image quality intelligent evaluation terminal.

[0041] Compared with the prior art, the model has the advantages that the closed-loop processing from data acquisition, model dynamic optimization, accurate quality evaluation to low-quality image improvement greatly improves the efficiency and accuracy of x-ray image data processing, the model obtains x image data through the data acquisition module to provide a basis for subsequent processing, the model also obtains unmarked x-ray images through the unmarked data processing module using the MAE processing model according to x image data, adjusts and revises the model, continuously optimizes the performance of the MAE processing model, and further improves the accuracy of the MAE processing model, so as to output more accurate unmarked x-ray images, the model also obtains evaluated x-ray images through the comprehensive quality evaluation module with the aid of the comprehensive quality evaluation model, and combines the root mean square error correction model to optimize the correction process according to the x image data, finally obtains accurate evaluated x-ray images, improves the evaluation accuracy of the x-ray image, and the model also generates high-quality images according to the evaluated x-ray image, the discrimination model and the GAN generation model through the low-quality x-ray image processing module, so as to improve the image quality of the x-ray image. BRIEF DESCRIPTION OF DRAWINGS

[0042] Figure 1 It is a structure schematic diagram of the x-ray image quality intelligent evaluation fusion model of the embodiment;

[0043] Figure 2 It is a flowchart of the use method of the x-ray image quality intelligent evaluation fusion model of the embodiment. DETAILED DESCRIPTION

[0044] In order to make the purpose and advantages of the present application clearer, the present application will be further described below in conjunction with the embodiments; it should be understood that the specific embodiments described herein are only used to explain the present application, and are not used to limit the present application.

[0045] The preferred embodiments of the present application will be described below with reference to the accompanying drawings. Those skilled in the art will understand that the embodiments are only used to explain the technical principles of the present application and are not intended to limit the protection scope of the present application.

[0046] It should be noted that, in the description of the present application, the terms indicating the direction or positional relationship of "upper", "lower", "left", "right", "inner", "outer" and the like are based on the direction or positional relationship shown in the drawings, which is only for the convenience of description, and does not indicate or imply that the device or element must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation on the present application.

[0047] In addition, it should be noted that, in the description of the present application, unless otherwise explicitly specified and limited, the terms "mounting", "connecting", "connection" should be understood broadly, for example, it can be fixed connection, or detachable connection, or integral connection; it can be mechanical connection, or electrical connection; it can be direct connection, or indirect connection through intermediate medium, or internal communication of two elements. Those skilled in the art can understand the specific meaning of the above terms in the present application according to the specific circumstances.

[0048] Please refer to Figure 1 As shown in the figure, it is a structural schematic diagram of the x-ray image quality intelligent evaluation fusion model of the embodiment, the model comprises:

[0049] The data acquisition module is used to acquire x image data;

[0050] The unlabeled data processing module is used to acquire unlabeled x-ray images according to the x image data and the MAE processing model, obtain the unlabeled x-ray images, and is also used to adjust the MAE processing model according to the sample size in the x image data, and is also used to revise the adjustment process of the MAE processing model according to the stringency index in the x image data, the unlabeled data processing module is connected with the data acquisition module;

[0051] The comprehensive quality evaluation module is used to acquire the evaluated x-ray images according to the unlabeled x-ray images and the comprehensive quality evaluation model, and is also used to calculate the root mean square error and correct the comprehensive quality evaluation model according to the root mean square error, and is also used to optimize the correction process of the comprehensive quality evaluation model according to the data noise level in the x image data, and is also used to correct the optimization process of the correction of the comprehensive quality evaluation model according to the device acquisition x-ray image time in the x image data, the comprehensive quality evaluation module is connected with the unlabeled data processing module;

[0052] The low-quality x-ray image processing module is configured to acquire the x-ray image with distortion type label according to the evaluation model, and acquire the high-quality x-ray image according to the GAN generation model and the x-ray image with distortion type label, so as to obtain the high-quality x-ray image, and send the high-quality x-ray image to the x-ray image quality intelligent evaluation terminal. The low-quality x-ray image processing module is connected with the comprehensive quality evaluation module.

[0053] Specifically, the x-ray image quality intelligent evaluation fusion model is applied to the x-ray image quality intelligent evaluation terminal. The model is constructed by a deep learning model and a machine learning model, such as an MAE processing model, a comprehensive quality evaluation model and a GAN generation model, to form an x-ray image quality intelligent evaluation fusion model. The model realizes a closed-loop processing from data collection, model dynamic optimization, accurate quality evaluation to low-quality image improvement, greatly improves the efficiency and accuracy of x-ray image data processing, and comprehensively improves the x-ray image data processing capability through the collaborative operation of multiple modules. The model acquires x-ray image data through the data collection module to provide a basis for subsequent processing. The model also acquires un-labeled x-ray images according to x-ray image data by using the MAE processing model through the un-labeled data processing module, adjusts and revises the model, continuously optimizes the performance of the MAE processing model, and further improves the accuracy of the MAE processing model to output more accurate un-labeled x-ray images. The model also acquires evaluated x-ray images by using the comprehensive quality evaluation model through the comprehensive quality evaluation module, combines the root mean square error correction model, optimizes the correction process according to x-ray image data, and finally obtains accurate evaluated x-ray images to improve the evaluation accuracy of x-ray images. The model also generates high-quality images according to the evaluated x-ray images, a discrimination model and a GAN generation model through the low-quality x-ray image processing module to improve the image quality of x-ray images.

[0054] Specifically, the data acquisition module acquires x image data, the x image data including an initial x-ray image, a sample size, a stringency index, a data noise level, and a device acquisition x-ray image time, the initial x-ray image being an initial acquired x-ray image, the data acquisition module acquiring the initial x-ray image through an x-ray generator, the sample size being a number of initial x-ray images input into the MAE model, the data acquisition module acquiring the sample size through a computer counter, the stringency index being a numerical mapping of a judgment of the stringency of the initial x-ray image according to expert experience, the mapping being between 0 and 1, the closer to 1, the higher the stringency, the data acquisition module acquiring the stringency index through expert experience, the data noise level being a numerical mapping of the intensity of non-ideal, interfering pixel changes in the image, the embodiment not limiting the specific acquisition method of the data acquisition module for acquiring the data noise level, and a person skilled in the art can set it according to the actual situation, such as acquiring the data noise level through a smoothing region method, the smoothing region method being a noise level estimation technique without reference to an image, and the device acquisition x-ray image time being a time consumed by the x-ray generator to acquire the initial x-ray image, the data acquisition module acquiring the device acquisition x-ray image time through a timer.

[0055] Specifically, the unlabeled data processing module constructs the MAE model through an MAE processing model construction method, inputs the initial x-ray image in the x image data into the MAE processing model, and obtains an unlabeled x-ray image output by the MAE processing model, the MAE processing model construction method including:

[0056] Step A1, taking a mask autoencoder as an initial architecture of the MAE processing model to obtain an initial MAE processing model;

[0057] Step A2, setting a mask ratio pj0 of the initial MAE processing model, setting pj0 = 75%, and obtaining a to-be-trained MAE processing model;

[0058] Step A3, training the to-be-trained MAE processing model through a processing data set to obtain a trained MAE processing model, and outputting the trained MAE processing model as the MAE processing model.

[0059] Specifically, the English full name of the MAE is Masked Autoencoder, and the Chinese name is Masked Autoencoder. The Masked Autoencoder refers to a deep learning architecture based on the concept of self-supervised learning. The mask ratio refers to the percentage of the part randomly covered in the input initial x-ray image. The processing data set refers to the training data set for training the MAE processing model. The processing data set includes the initial x-ray image obtained historically and the unlabeled x-ray image corresponding to the initial x-ray image obtained historically. The unlabeled x-ray image refers to the x-ray image repaired by the MAE processing model. The embodiment does not limit the specific implementation of training the MAE processing model by the processing data set. The person skilled in the art can set it according to the actual situation, such as dividing the processing data set into 70% processing training set, 15% processing verification set and 15% processing test set. The trained MAE processing model is obtained by training the MAE processing model through the processing training set, and the performance of the MAE processing model is verified through the processing verification set. The trained MAE processing model is obtained by testing the accuracy of the MAE processing model through the processing test set.

[0060] Specifically, the unlabeled data processing module inputs the initial x-ray image in the x image data into the MAE processing model to obtain the unlabeled x-ray image output by the MAE processing model, supplements the missing part of the initial x-ray image, avoids inaccurate analysis of the x-ray image in the future, and improves the processing accuracy of the x-ray image.

[0061] Specifically, the unlabeled data processing module compares the sample quantity Yb in the x image data with the preset sample quantity Yb0, judges the sample quantity according to the comparison result, and adjusts the MAE processing model according to the judgment result, wherein:

[0062] When Yb≤Yb0, the unlabeled data processing module determines that the sample quantity is small, and does not adjust the MAE processing model;

[0063] When Yb>Yb0, the unlabeled data processing module determines that the sample quantity is large, and adjusts the MAE processing model. The mask rate pj0 is adjusted by adjusting coefficient Tj, Tj=(1-Yb0 / Yb), to obtain the adjusted mask rate pj0`. Set pj0`=Tj×pj0, replace the mask rate pj0 with the adjusted mask rate pj0`, and retrain the MAE model according to the mask rate pj0.

[0064] Specifically, the preset sample quantity refers to a preset value for judging the sample quantity condition, and the embodiment does not limit the specific value of the preset sample quantity. A person skilled in the art can set it according to the actual situation, for example, the preset sample quantity can be set according to the processing efficiency of the MAE processing model, and 80≤Yb0≤120 can be set. The sample quantity condition refers to the amount of sample quantity according to the sample quantity and the preset sample quantity. The sample quantity condition includes a large sample quantity and a small sample quantity.

[0065] Specifically, the unlabeled data processing module judges the sample quantity condition and adjusts the MAE processing model according to the judgment result. When the sample quantity condition is a large sample quantity, the mask rate is adjusted by setting an adjustment coefficient that decreases with the increase of the sample quantity, so as to speed up the processing speed of the MAE processing model and achieve the purpose of efficient image supplement.

[0066] Specifically, the unlabeled data processing module compares the stringency index Yy in the x image data with the preset stringency index Yy0, judges the stringency index condition according to the comparison result, and revises the adjustment process of the MAE processing model according to the judgment result, wherein:

[0067] When Yy≤Yy0, the unlabeled data processing module determines that the stringency index condition is low, and does not revise the adjustment process of the update of the MAE processing model;

[0068] When Yy>Yy0, the unlabeled data processing module determines that the stringency index condition is high, and revises the adjustment process of the update of the MAE processing model. The preset sample quantity Yb0 is revised by a revision coefficient cjb, which is set as cjb=Yy0 / Yy, to obtain the revised preset sample quantity Yb0x1, which is set as Yb0x1=Yb0×cjb. The preset sample quantity Yb0 is replaced by the revised preset sample quantity Yb0x1, and the sample quantity Yb is compared with the preset sample quantity Yb0 again.

[0069] Specifically, the preset stringency index refers to a preset value for judging the stringency index condition, and the embodiment does not limit the specific value of the preset stringency index. For example, the specific value of the preset stringency index can be set according to the quality requirement of the x image. The stringency index condition refers to the high and low condition of the stringency index according to the stringency index and the preset stringency index. The stringency index condition includes low stringency index and high stringency index.

[0070] Specifically, the unlabeled data processing module adjusts the value of the preset sample size by judging the case of the stringency index, so as to increase the sample quantity in the case of high stringency, thereby improving the accuracy of the data. When the stringency index case is high stringency index, by setting a revision coefficient that increases with the increase of the stringency index, the revision coefficient is multiplied by the preset sample size, thereby increasing the value of the preset sample size, and improving the accuracy of the sample data.

[0071] Specifically, the comprehensive quality evaluation module constructs the comprehensive quality evaluation model by a comprehensive quality evaluation model construction method, obtains the comprehensive quality evaluation model, and inputs the unlabeled x-ray image into the comprehensive quality evaluation model to obtain the evaluated x-ray image output by the comprehensive quality evaluation model. The evaluated x-ray image includes a comprehensive quality evaluation score The comprehensive quality evaluation model construction method includes:

[0072] The evaluation data set is divided into 70% evaluation training set, 20% evaluation validation set and 10% evaluation test set. The evaluation training set is input into the ResNet50 model to train the ResNet50 model. The evaluation validation set is input into the trained ResNet50 model to perform hyperparameter iterative optimization on the trained ResNet50 model. The evaluation test set is input into the iterative optimized ResNet50 model to perform evaluation test on the iterative optimized ResNet50 model, and the evaluation test result is obtained. The total sample quantity of the evaluation test set is set as f0, the correct evaluation test sample quantity is f, the evaluation test accuracy is F, F=f / f0, the evaluation test accuracy F is compared with the preset evaluation test accuracy F0, F0≥9, the training of the iterative optimized ResNet50 model is judged according to the comparison result, and the output is performed according to the judgment result, wherein:

[0073] When F is greater than or equal to F0, the comprehensive quality evaluation module determines that the iterative optimized ResNet50 model training is up to standard, and outputs the iterative optimized ResNet50 model as the comprehensive quality evaluation model;

[0074] When F is less than F0, the comprehensive quality evaluation module determines that the iterative optimized ResNet50 model training is not up to standard, updates the evaluation data set to obtain an updated evaluation data set, and trains, iteratively optimizes and analyzes the test of the ResNet50 model according to the updated evaluation data set until the ResNet50 model training is up to standard.

[0075] Specifically, the comprehensive quality assessment score refers to the comprehensive quality assessment score of the comprehensive quality assessment module for the unlabeled X-ray image. The comprehensive quality assessment model refers to a ResNet50 model that takes the unlabeled X-ray image as input and the evaluated X-ray image as output. The ResNet50 model is a type of deep residual network. The full English name of ResNet50 is Residual Network 50-layer, and its full Chinese name is Deep Residual Network (50 layers). This embodiment does not limit the construction method of the comprehensive quality assessment model. Those skilled in the art can set it according to the actual situation. For example, the ResNet50 model can be trained using an evaluation dataset to obtain the comprehensive quality assessment model. The evaluation dataset refers to the training dataset used to construct the comprehensive quality assessment model. The evaluation dataset includes historically acquired unlabeled X-ray images and the evaluated X-ray images corresponding to the historically acquired unlabeled X-ray images. The preset evaluation test accuracy rate refers to a preset value used to judge the training achievement of the iteratively optimized decision tree model. The iteratively optimized decision tree... The training achievement status of the model refers to the accuracy achievement status of the trained decision tree model. The training achievement status of the iteratively optimized decision tree model includes both the training achievement status and the training failure status. The evaluation training set refers to the training data set used to train the ResNet50 model. The evaluation validation set refers to the validation data set used to verify the performance of the ResNet50 model. The evaluation test set refers to the test data set used to test the accuracy of the ResNet50 model. This embodiment does not limit the specific implementation method of updating the evaluation dataset. Those skilled in the art can set it according to the actual situation.

[0076] Specifically, the comprehensive quality assessment module constructs a comprehensive quality assessment model and inputs an unlabeled X-ray image into the model to obtain the assessed X-ray image output by the model, so that the assessed X-ray image can be processed subsequently.

[0077] Specifically, the comprehensive quality assessment module is based on the comprehensive quality assessment score. Data volume and true X-ray image quality score For root mean square error Perform calculations and set Root mean square error The model is compared with the preset root mean square error RMSE0, where 0 ≤ RMSE0 ≤ 0.5. The model performance is assessed based on the comparison results, and the overall quality assessment model is then revised accordingly.

[0078] When When RMSE0, the comprehensive quality evaluation module determines that the model performance is qualified, and does not correct the comprehensive quality evaluation model;

[0079] When When RMSE0, the comprehensive quality evaluation module determines that the model performance is unqualified, corrects the comprehensive quality evaluation model, corrects the preset evaluation test accuracy F0 through a correction coefficient Czx, sets Czx=1.04-0.03×e -（RMSE-RMSE0） , obtains a corrected preset evaluation test accuracy F01, sets F01=F0×Czx, replaces the preset evaluation test accuracy F0 with the corrected preset evaluation test accuracy F01, and recompares the evaluation test accuracy F with the preset evaluation test accuracy F0.

[0080] Specifically, the data quantity refers to the number of unlabeled x-ray images input into the comprehensive quality evaluation model, and the specific acquisition method of the data quantity is not limited in the embodiment, for example, the data quantity can be acquired by a computer counting method. The preset root mean square error refers to a preset value for judging the model performance. The model performance refers to the performance of the comprehensive quality evaluation model determined according to the root mean square error and the preset root mean square error. The model performance includes model performance qualified and model performance unqualified. The real x-ray image quality score refers to the actual comprehensive quality evaluation score of the evaluated x-ray image, and the acquisition method of the real x-ray image quality score is not limited in the embodiment, for example, the real x-ray image quality score can be obtained by evaluating the actual comprehensive quality of the evaluated x-ray image by an expert.

[0081] Specifically, the comprehensive quality evaluation module judges the model performance, and when the model performance is unqualified, sets the correction coefficient to increase from 1.01 to 1.04 with the increase of the root mean square error, adjusts the value of the preset evaluation test accuracy, and increases the value of the preset evaluation test accuracy, so as to increase the accuracy of the comprehensive quality evaluation model, thereby improving the accuracy of the comprehensive quality evaluation.

[0082] Specifically, the comprehensive quality evaluation module compares the data noise level τ in the x image data with the preset data noise level τ0, judges the data noise according to the comparison result, and optimizes the process of correcting the model according to the judgment result, wherein:

[0083] When τ≤τ0, the comprehensive quality evaluation module determines that the data noise level is low, and does not optimize the process of correcting the model;

[0084] When τ>τ0, the comprehensive quality evaluation module determines that the data noise condition is high noise level, optimizes the process of correcting the model, optimizes the preset root mean square error RMSE0 through the noise coefficient qbz, sets qbz=0.75+0.2×e -0.7×（τ-τ0） , obtains the optimized preset root mean square error RMSE0`, sets RMSE0`=RMSE0×qbz, replaces the preset root mean square error RMSE0 with the optimized preset root mean square error RMSE0`, and compares the root mean square error RMSE with the preset root mean square error RMSE0 again.

[0085] Specifically, the preset data noise level refers to a preset value for judging the data noise condition, and the embodiment does not limit the specific value of the preset data noise level. For example, the specific value of the preset data noise level can be limited according to the image quality requirement, and 0≤τ0≤1.23 is set. The data noise condition refers to the noise condition of the evaluated x-ray graph line determined according to the data noise level and the preset data noise level. The data noise condition includes low noise level and high noise level.

[0086] Specifically, the comprehensive quality evaluation module judges the data noise condition. When the data noise condition is high noise level, the output of the comprehensive quality evaluation score of the comprehensive quality evaluation module is disturbed by noise, which leads to inaccurate evaluated x-ray graph line. The noise coefficient is set to decrease from 0.95 to 0.75 approaching to 0.75 infinitely with the increase of noise data level, to optimize the preset root mean square error, reduce the value of the preset root mean square error, and make the model performance condition more likely to be determined as model performance unqualified. Thus, the comprehensive quality evaluation model is corrected, the accuracy of the comprehensive quality evaluation model is improved, and the influence of noise on the accuracy of the output comprehensive quality evaluation score of the comprehensive quality evaluation model is reduced.

[0087] Specifically, the comprehensive quality evaluation module compares the device collected x-ray image time Ts in the x image data with the preset device collected x-ray image time Ts0, judges the collection time condition according to the comparison result, and corrects the optimization process of the correction of the comprehensive quality evaluation model according to the judgment result, wherein:

[0088] When Ts≤Ts0, the comprehensive quality evaluation module determines that the collection time condition is time shortage, and does not correct the optimization process of the model correction;

[0089] When Ts>Ts0, the comprehensive quality evaluation module determines that the acquisition time condition is time excess, corrects the optimization process of model correction, corrects the preset data noise level τ0 through the correction coefficient jz, sets jz=Ts0 / Ts, obtains the corrected preset data noise level τ0`, sets τ0`=τ×jz, replaces the preset data noise level τ0 with the corrected preset data noise level τ0`, and compares the data noise level τ with the preset data noise level τ0 again.

[0090] Specifically, the preset device acquisition x-ray image time refers to a preset value for judging the acquisition time condition, and the embodiment does not limit the specific value of the preset device acquisition x-ray image time. For example, the specific value of the preset device acquisition x-ray image time can be limited according to the quality requirement of the picture, and 50ms≤Ts0≤100ms is set. The acquisition time condition refers to the condition of how much acquisition time according to the device acquisition x-ray image time and the preset device acquisition x-ray image time. The acquisition time condition includes time shortage and time excess.

[0091] Specifically, the comprehensive quality evaluation module corrects the value of the preset data noise level by setting a correction coefficient that decreases with the increase of the device acquisition x-ray image time when the acquisition time condition is time excess, avoids the interference caused by the optimization process of the model correction process in which more data noise is collected due to the excessively long acquisition time, and further improves the accuracy of the comprehensive quality evaluation model.

[0092] Specifically, the comprehensive quality evaluation module compares the comprehensive quality evaluation score yz of the x-ray image with the preset comprehensive quality evaluation score yz0, judges the x-ray image comprehensive quality evaluation score condition according to the comparison result, and outputs the image attribute of the evaluated x-ray image according to the judgment result, wherein:

[0093] When yz>yz0, the comprehensive quality evaluation module determines that the x-ray image comprehensive quality evaluation score condition is high score, outputs the high-quality x-ray image as the image attribute of the evaluated x-ray image, and sends the evaluated x-ray image to the unmarked data processing module as the learning data set, and re-trains the MAE model according to the learning data.

[0094] When yz>yz0, the comprehensive quality evaluation module determines that the x-ray image comprehensive quality evaluation score condition is high score, outputs the high-quality x-ray image as the image attribute of the evaluated x-ray image, and sends the evaluated x-ray image to the unmarked data processing module as the learning data set, and re-trains the MAE model according to the learning data.

[0095] ​​​Specifically, the preset comprehensive quality evaluation score refers to a preset value for judging the comprehensive quality evaluation score of the x-ray image. The specific value of the preset comprehensive quality evaluation score is not limited in the embodiment, and can be set by the person skilled in the art according to the actual situation, such as setting the specific value of the preset comprehensive quality evaluation score according to the image quality requirement, setting 2≤yz0≤4, the comprehensive quality evaluation score of the x-ray image refers to the comprehensive quality evaluation score of the unlabeled x-ray image judged by the comprehensive quality evaluation score and the preset comprehensive quality evaluation score, the comprehensive quality evaluation score of the x-ray image includes low score and high score, the x-ray image attribute refers to the image attribute of the unlabeled x-ray image, the x-ray image attribute includes low quality x-ray image and high quality x-ray image, the specific implementation of sending the evaluated x-ray image to the low quality x-ray image processing module is not limited in the embodiment, such as sending the evaluated x-ray image to the low quality x-ray image processing module in the form of wireless transmission, the specific implementation of sending the evaluated x-ray image as a learning data set to the unlabeled data processing module is not limited in the embodiment, such as sending the evaluated x-ray image as a learning data set to the unlabeled data processing module in the form of wireless transmission.

[0096] Specifically, the comprehensive quality evaluation module judges the comprehensive quality evaluation score of the x-ray image, and outputs the image attribute of the evaluated x-ray image according to the judgment result, so as to process the evaluated x-ray image.

[0097] Specifically, the low quality x-ray image processing module inputs the evaluated x-ray image into the discrimination model to obtain the distortion type label x-ray image output by the discrimination model.

[0098] Specifically, the discrimination model refers to a convolutional neural network model taking the evaluated x-ray image as input and taking the distortion type label x-ray image as output, the distortion type label x-ray image refers to an unlabeled x-ray image whose x-ray image attribute containing distortion type label on the image is a low quality x-ray image, the distortion type label refers to a label of an unexpected change occurring in the process of acquisition, processing or transmission of the evaluated x-ray image, the specific construction method of the discrimination model is not limited in the embodiment, and can be set by the person skilled in the art according to the actual situation, such as training the convolutional neural network model through the distortion data set to obtain the discrimination model, the distortion data set refers to a learning data set for constructing the discrimination model, and the distortion data set includes the evaluated x-ray image acquired historically and the distortion type label x-ray image corresponding to the evaluated x-ray image acquired historically.

[0099] Specifically, the low-quality x-ray image processing module inputs the evaluated x-ray image into a discrimination model to obtain a distortion type labeled x-ray image output by the discrimination model, so as to subsequently repair the distortion type labeled x-ray image to obtain a high-quality x-ray image.

[0100] Specifically, the low-quality x-ray image processing module inputs the distortion type labeled x-ray image into a GAN generation model to obtain a generated image output by the GAN generation model, and outputs the generated image as a high-quality x-ray image, thereby obtaining the high-quality x-ray image and sending the high-quality x-ray image to an x-ray image quality intelligent evaluation terminal.

[0101] Specifically, the GAN generation model refers to a generative adversarial network model taking the distortion type labeled x-ray image as input and taking the generated image as output. The full name of the GAN is Generative Adversarial Network, and the Chinese name is generative adversarial network model. The specific construction method of the GAN generation model is not limited in the embodiment, and a person skilled in the art can set it according to the actual situation, such as training the generative adversarial network model through a generated data set to obtain the GAN generation model. The generated data set refers to a training data set used to construct the GAN generation model. The generated data set includes the distortion type labeled x-ray image obtained historically and the generated image corresponding to the distortion type labeled x-ray image obtained historically. The specific method of sending the high-quality x-ray image to the x-ray image quality intelligent evaluation terminal is not limited in the embodiment, and a person skilled in the art can set it according to the actual situation, such as sending the high-quality x-ray image to the x-ray image quality intelligent evaluation terminal through wireless transmission.

[0102] Specifically, the low-quality x-ray image processing module inputs the distortion type labeled x-ray image into a GAN generation model to obtain a generated image output by the GAN generation model, and outputs the generated image as a high-quality x-ray image, thereby obtaining the high-quality x-ray image and sending the high-quality x-ray image to an x-ray image quality intelligent evaluation terminal.

[0103] Please refer to Figure 2 The method includes the following steps:

[0104] In step S1, the x-ray image data is collected through a data collection module.

[0105] In step S2, the unlabeled x-ray image is obtained through an unlabeled data processing module according to the x-ray image data and the MAE processing model. The MAE processing model is adjusted according to the sample size in the x-ray image data, and the adjustment process of the MAE processing model is revised according to the stringency index in the x-ray image data.

[0106] In step S3, the comprehensive quality evaluation module is used to obtain the evaluated x-ray image according to the un-labeled x-ray image and the comprehensive quality evaluation model, calculate the root mean square error, correct the comprehensive quality evaluation model according to the root mean square error, optimize the correction process of the comprehensive quality evaluation model according to the data noise level in the x image data, and correct the optimization process of the correction of the comprehensive quality evaluation model according to the x-ray image acquisition time of the equipment in the x image data.

[0107] In step S4, the low-quality x-ray image processing module is used to obtain the distortion type labeled x-ray image according to the evaluated x-ray image through the discrimination model, obtain the high-quality x-ray image according to the GAN generation model and the distortion type labeled x-ray image, obtain the high-quality x-ray image, and send the high-quality x-ray image to the x-ray image quality intelligent evaluation terminal.

[0108] So far, the technical solutions of the present application have been described in combination with the preferred embodiments shown in the drawings, but those skilled in the art can easily understand that the protection scope of the present application is obviously not limited to these specific embodiments. Those skilled in the art can make equivalent changes or replacements to the related technical features without departing from the principles of the present application, and the technical solutions after these changes or replacements will fall within the protection scope of the present application.

Claims

1. A smart evaluation and fusion system for X-ray image quality, characterized in that, include: The data acquisition module is used to acquire x-image data; The unlabeled data processing module is used to acquire unlabeled X-ray images based on X-ray image data and MAE processing model, and to adjust the MAE processing model based on the sample size in the X-ray image data. It is also used to revise the adjustment process of the MAE processing model based on the rigor index in the X-ray image data. The comprehensive quality assessment module is used to acquire the assessed X-ray images based on the unlabeled X-ray images and the comprehensive quality assessment model. It is also used to calculate the root mean square error and correct the comprehensive quality assessment model based on the root mean square error. Furthermore, it is used to optimize the correction process of the comprehensive quality assessment model based on the data noise level in the X-ray image data, and to correct the optimization process of the comprehensive quality assessment model based on the equipment acquisition time of the X-ray images in the X-ray image data. The low-quality X-ray image processing module is used to obtain the distortion type labeled X-ray image based on the evaluated X-ray image through the discrimination model, and to obtain the high-quality X-ray image based on the GAN generation model and the distortion type labeled X-ray image, thus obtaining the high-quality X-ray image, and sending the high-quality X-ray image to the X-ray image quality intelligent evaluation terminal. in: The MAE processing model is adjusted based on the sample size in the x-image data as follows: When the sample size is greater than the preset sample size, the masking rate of the MAE processing model is adjusted to reduce the value of the masking rate. The process of revising the MAE processing model based on the rigor index in the x-image data is as follows: When the rigor index is greater than the preset rigor index, the preset sample size is revised and the value of the preset sample size is reduced. The root mean square error (RMSE) is calculated, and the comprehensive quality assessment model is then corrected based on the RMSE. Specifically: When the root mean square error is greater than the preset root mean square error, the preset evaluation test accuracy of the comprehensive quality assessment model is corrected and the value of the preset evaluation test accuracy is increased. The optimization process for the comprehensive quality assessment model based on the noise level in the x-image data is as follows: When the data noise level is greater than the preset data noise level, the preset root mean square error is optimized to reduce the value of the preset root mean square error. The optimization process for correcting the comprehensive quality assessment model based on the device acquisition time of X-ray images in the X-ray image data is as follows: When the time for the device to acquire X-ray images exceeds the preset time for acquiring X-ray images, the preset data noise level is corrected to reduce the value of the preset data noise level.

2. The intelligent evaluation and fusion system for X-ray image quality according to claim 1, characterized in that, The unlabeled data processing module constructs the MAE model using the MAE processing model construction method. It inputs the initial x-ray image from the x-image data into the MAE processing model to obtain the unlabeled x-ray image output by the MAE processing model. The MAE processing model construction method includes: Step A1: Use the mask autoencoder as the initial architecture of the MAE processing model to obtain the initial MAE processing model. Step A2: Set the mask ratio pj0 of the initial MAE processing model, set pj0=75%, and obtain the MAE processing model to be trained. Step A3: Train the MAE processing model to be trained by processing the dataset to obtain the trained MAE processing model, and output the trained MAE processing model as the MAE processing model.

3. The intelligent evaluation and fusion system for X-ray image quality according to claim 2, characterized in that, The unlabeled data processing module compares the sample size Yb in the x-image data with the preset sample size Yb0, judges the sample size based on the comparison result, and adjusts the MAE processing model based on the judgment result, wherein: When Yb≤Yb0, the unlabeled data processing module determines that the sample size is too small and does not adjust the MAE processing model. When Yb > Yb0, the unlabeled data processing module determines that the sample size is too large, adjusts the MAE processing model, adjusts the mask ratio pj0 by adjusting the adjustment coefficient Tj, Tj = (1 - Yb0 / Yb), and obtains the adjusted mask ratio pj0'. Set pj0' = Tj × pj0, replace the mask ratio pj0 with the adjusted mask ratio pj0', and retrain the MAE model based on the mask ratio pj0.

4. The intelligent evaluation and fusion system for X-ray image quality according to claim 3, characterized in that, The unlabeled data processing module compares the rigor index Yy in the x-image data with the preset rigor index Yy0, judges the rigor index status based on the comparison result, and revises the adjustment process of the MAE processing model based on the judgment result, wherein: When Yy≤Yy0, the unlabeled data processing module determines the rigor index as low and does not revise the adjustment process for updating the MAE processing model. When Yy > Yy0, the unlabeled data processing module determines that the rigor index is high and revises the adjustment process of updating the MAE processing model. The preset sample size Yb0 is revised by revising the revision coefficient cjb. cjb = Yy0 / Yy is set to obtain the revised preset sample size Yb0x1. Yb0x1 = Yb0 × cjb is set to replace the preset sample size Yb0x1 with the revised preset sample size Yb0x1, and the sample size Yb is re-compared with the preset sample size Yb0.

5. The intelligent evaluation and fusion system for X-ray image quality according to claim 2, characterized in that, The comprehensive quality assessment module constructs a comprehensive quality assessment model using a comprehensive quality assessment model construction method. An unlabeled X-ray image is then input into the comprehensive quality assessment model to obtain the assessed X-ray image output by the model. This assessed X-ray image includes the comprehensive quality assessment score. The method for constructing the comprehensive quality assessment model includes: The evaluation dataset is divided into a 70% training set, a 20% validation set, and a 10% test set. The training set is input into the ResNet50 model for training. The validation set is input into the trained ResNet50 model for iterative hyperparameter optimization. The test set is input into the optimized ResNet50 model for evaluation testing. The total number of test samples is set as f0, the number of correctly evaluated test samples is f, and the evaluation accuracy is F, where F = f / f0. The evaluation accuracy F is compared with the preset evaluation accuracy F0. F0 ≥ 9. Based on the comparison results, the training performance of the optimized ResNet50 model is judged, and the judgment result is output. When F≥F0, the comprehensive quality assessment module determines that the training of the iteratively optimized ResNet50 model has reached the target, and outputs the iteratively optimized ResNet50 model as the comprehensive quality assessment model. When F < F0, the comprehensive quality assessment module determines that the training of the iteratively optimized ResNet50 model is substandard, updates the assessment dataset to obtain the updated assessment dataset, and trains, iteratively optimizes hyperparameters, and analyzes and tests the ResNet50 model based on the updated assessment dataset until the ResNet50 model training meets the standards.

6. The intelligent evaluation and fusion system for X-ray image quality according to claim 5, characterized in that, The comprehensive quality assessment module is based on the comprehensive quality assessment score. Data volume and true X-ray image quality score For root mean square error Perform calculations and set Root mean square error The model is compared with the preset root mean square error RMSE0, where 0 ≤ RMSE0 ≤ 0.

5. The model performance is assessed based on the comparison results, and the overall quality assessment model is then revised accordingly. when When ≤RMSE0, the comprehensive quality assessment module determines that the model performance is qualified and does not make any corrections to the comprehensive quality assessment model; when When the RMSE > 0, the comprehensive quality assessment module determines that the model performance is unqualified, and corrects the comprehensive quality assessment model by adjusting the preset assessment test accuracy F0 using a correction coefficient Czx, where Czx = 1.04 - 0.03 × e -（RMSE-RMSE0） The corrected preset evaluation test accuracy F01 is obtained. F01 is set to F0×Czx. The preset evaluation test accuracy F0 is replaced with the corrected preset evaluation test accuracy F01. The evaluation test accuracy F is then compared with the preset evaluation test accuracy F0 again.

7. The intelligent evaluation and fusion system for X-ray image quality according to claim 6, characterized in that, The comprehensive quality assessment module compares the data noise level τ in the x-image data with the preset data noise level τ0, judges the data noise situation based on the comparison result, and optimizes the model correction process based on the judgment result, wherein: When τ≤τ0, the comprehensive quality assessment module determines that the data noise level is low and does not optimize the process of correcting the model. When τ>τ0, the comprehensive quality assessment module determines that the data noise level is high, and optimizes the model correction process by using the noise figure qbz to optimize the preset root mean square error RMSE0, setting qbz=0.75+0.2×e -0.7×（τ-τ0） The optimized preset root mean square error RMSE0` is obtained. RMSE0` is set to RMSE0 × qbz. The preset root mean square error RMSE0 is replaced with the optimized preset root mean square error RMSE0`. The root mean square error RMSE is then compared with the preset root mean square error RMSE0 again.

8. The intelligent evaluation and fusion system for X-ray image quality according to claim 7, characterized in that, The comprehensive quality assessment module compares the device acquisition time Ts of the X-ray image data with the preset device acquisition time Ts0, judges the acquisition time based on the comparison result, and corrects the optimization process of the comprehensive quality assessment model based on the judgment result, wherein: When Ts≤Ts0, the comprehensive quality assessment module determines that the data collection time is too short and does not correct the optimization process of model correction. When Ts > Ts0, the comprehensive quality assessment module determines that the data acquisition time is too long and corrects the optimization process of the model correction. The preset data noise level τ0 is corrected by the correction coefficient jz. jz = Ts0 / Ts is set to obtain the corrected preset data noise level τ0`. τ0` = τ × jz is set to replace the preset data noise level τ0 with the corrected preset data noise level τ0`, and the data noise level τ is compared with the preset data noise level τ0 again.

9. The intelligent evaluation and fusion system for X-ray image quality according to claim 8, characterized in that, The comprehensive quality assessment module will assign a comprehensive quality assessment score. The image is compared with a preset comprehensive quality assessment score yz0. Based on the comparison results, the comprehensive quality assessment score of the X-ray image is judged, and the image attributes of the evaluated X-ray image are output according to the judgment results, including: when When ≤yz0, the comprehensive quality assessment module determines that the comprehensive quality assessment score of the X-ray image is low, outputs the low-quality X-ray image as the image attribute of the assessed X-ray image, and sends the assessed X-ray image to the low-quality X-ray image processing module. when When yz > 0, the comprehensive quality assessment module determines that the comprehensive quality assessment score of the X-ray image is high, outputs the high-quality X-ray image as an attribute of the assessed X-ray image, and sends the assessed X-ray image as a learning dataset to the unlabeled data processing module, and retrains the MAE model based on the learning data.

10. A method of using the intelligent evaluation and fusion system for X-ray image quality according to any one of claims 1 to 9, characterized in that, include: Step S1: Acquire x image data through the data acquisition module; Step S2: The unlabeled x-ray image is obtained by the unlabeled data processing module based on the x-image data and the MAE processing model. The MAE processing model is adjusted according to the sample size in the x-image data and the adjustment process of the MAE processing model is revised according to the rigor index in the x-image data. Step S3: The comprehensive quality assessment module acquires the assessed X-ray image based on the unlabeled X-ray image and the comprehensive quality assessment model. It also calculates the root mean square error and corrects the comprehensive quality assessment model based on the root mean square error. Furthermore, it optimizes the correction process of the comprehensive quality assessment model based on the data noise level in the X-ray image data and corrects the optimization process of the comprehensive quality assessment model based on the equipment acquisition time of the X-ray image data. Step S4: Using the low-quality X-ray image processing module, the distortion type labeled X-ray image is obtained based on the evaluated X-ray image through the discrimination model, and the high-quality X-ray image is obtained based on the GAN generation model and the distortion type labeled X-ray image, thus obtaining the high-quality X-ray image, and then sending the high-quality X-ray image to the X-ray image quality intelligent evaluation terminal.

Images