A deep learning-based automatic adjustment method for an X-ray focusing system

Abstract

The application provides an X-ray focusing system automatic adjustment method based on deep learning, comprising: providing an X-ray optical imaging system, which comprises an X-ray focusing device, a detector system and a parameter adjustment system; providing a focal spot quality determination device and method for obtaining focal spot quality data and / or an optical system simulation module; obtaining optimal focal spot quality, recording the optimal driving parameters of the corresponding parameter adjustment system; selecting a plurality of driving parameter values in the parameter adjustment interval of the parameter adjustment system, and recording the error parameters and the optical images of the detector as sample data; training a deep learning calculation system, processing the optical images currently collected by the detector system based on the training results, obtaining the estimation results of the error parameters and adjusting them to improve the focal spot quality. The method of the application can realize high-precision fast automatic calibration of the X-ray focusing system and improve the focal spot quality.

Description

Technical Field

[0001] This invention pertains to an automatic adjustment method for X-ray optical focusing systems, specifically a focusing system parameter optimization method based on deep learning technology. Background Technology

[0002] Deep learning is a data-driven, model-based machine learning method. It is based on a neural network model with a large number of parameters. It learns target information by optimizing the gradient of the objective function, calculating the gradient of all model parameters using backpropagation, and then using gradient descent to optimize the parameters. Deep learning is now widely used in engineering and scientific research, enabling applications such as facial recognition, image classification, and speech recognition. It is also widely adopted in research fields such as biology, materials science, physics, and chemistry.

[0003] X-ray focusing systems, primarily including Kirkpatrick-Baez (KB) mirror systems, Fresnel zone plate focusing systems, compound refractive lens (CRL) focusing systems, Laue lenses, and waveguides, are widely used for beam focusing tasks in X-ray instruments, synchrotron radiation, and free-electron lasers. To achieve good focusing, the X-ray focusing system needs to be in a suitable position and orientation. The calibration error of the X-ray focusing system relative to its ideal orientation directly affects the focal spot quality of the optical system, and consequently, the spatial resolution of X-ray focusing and imaging experiments. In practical applications, complex environmental variables and optical mechanisms exert influence, and the X-ray focused spot is typically extremely small, making it difficult to directly observe subtle changes in focal spot quality. Therefore, it is difficult to finely adjust the focusing system using fixed methods or direct observation and adjustment. Furthermore, methods such as knife-edge scanning to determine and adjust the focal spot state are time-consuming.

[0004] Therefore, there is an urgent need for a method to adjust the X-ray focusing system in order to achieve rapid and automated adjustment of the X-ray focusing system with high precision. Summary of the Invention

[0005] The present invention aims to provide an automatic adjustment method for X-ray focusing systems based on deep learning, so as to achieve rapid and automated calibration of X-ray focusing systems with high precision and improve the quality of the focal spot.

[0006] To achieve the above objectives, the present invention provides an automatic adjustment method for an X-ray focusing system based on deep learning, comprising:

[0007] S1 provides an X-ray optical imaging system and a deep learning computing system; the X-ray optical imaging system includes an X-ray focusing device and a detector system arranged along the incident direction of the X-ray beam, and a parameter adjustment system connected to the X-ray focusing device;

[0008] S2 provides a focal spot quality measurement device for acquiring focal spot quality data and a corresponding focal spot quality measurement method and / or optical system simulation module;

[0009] S3, obtain the focal spot quality data corresponding to the driving parameters of different parameter adjustment systems, select the best focal spot quality from them, and record the best driving parameters of the corresponding parameter adjustment system;

[0010] S4, with the optimal driving parameters of the parameter adjustment system as the center, set the parameter adjustment range of the parameter adjustment system;

[0011] S5. Select the parameter values ​​of multiple driving parameters within the parameter adjustment range of the parameter adjustment system, and record the difference between the driving parameters and the optimal driving parameters, and the optical image acquired by the detector system as sample data, wherein the difference between the driving parameters and the optimal driving parameters is used as the error parameter.

[0012] S6. The sample data is preprocessed. The optical image of the preprocessed sample data is used as the input parameter and the error parameter is used as the label to train the deep learning computing system to obtain the parameter adjustment model of the X-ray focusing system.

[0013] S7. When using an X-ray optical imaging system, the optical image currently acquired by the detector system is processed based on the X-ray focusing system parameter adjustment model to obtain the estimated result of the error parameter. The parameter adjustment system is then used to adjust the parameters according to the estimated result of the error parameter in order to improve the quality of the focal spot.

[0014] The X-ray focusing device is configured to receive an incident X-ray beam and generate a focused X-ray beam; the parameter adjustment system is configured to adjust the adjustable parameters of the X-ray focusing device by changing its driving parameters; the detector system is configured to receive the optical image of the focused X-ray beam.

[0015] If step S2 provides a focal spot quality measuring device, then in step S3, the focal spot quality measuring device is installed in the X-ray optical imaging system. When the driving parameters of the parameter adjustment system are different, the focal spot quality measuring device and the focal spot quality measuring method of step S2 are used to obtain focal spot quality data respectively. If step S2 provides an optical system simulation module, then in step S3, the optical system simulation module is used to obtain focal spot quality data corresponding to the driving parameters of different parameter adjustment systems. The optical system simulation module is configured to output a simulated detector optical image based on the value of the driving parameters of the parameter adjustment system, thereby obtaining the focal spot quality data.

[0016] Step S2 provides a focal spot quality measuring device. In step S3, before installing the focal spot quality measuring device into the X-ray optical imaging system, the method further includes: removing the detector system out of the optical path when the detector system is not needed and the detector system affects the optical path of the focal spot quality measuring device.

[0017] In step S4, based on the accuracy requirements of the optical system and the error adjustment accuracy of the parameter adjustment system, the parameter adjustment range of the parameter adjustment system is set with the optimal driving parameter of the parameter adjustment system as the center.

[0018] The X-ray optical imaging system also includes an optical image enhancement device disposed between the parameter adjustment system and the detector system to increase the features of the optical image. The optical image enhancement device is a scattering, transmission, or reflection device.

[0019] The incident direction of the X-ray beam is the Y direction, the optical image enhancement device is a scattering thin film, and step S5 further includes: adjusting the driving parameter value of each parameter adjustment system with the focal point of the X-ray focusing device as the center to obtain different Y-direction positions of the scattering thin film, and recording the distance between the Y-direction position of the scattering thin film and the focal point of the X-ray focusing device while recording the optical image acquired by the detector system, as sample data, wherein the difference between the driving parameter value of the parameter adjustment system and the optimal driving parameter and the distance between the Y-direction position of the scattering thin film and the focal point of the X-ray focusing device are both used as error parameters.

[0020] Extended exposure is used when the detector system acquires optical images to enhance the features of the optical images.

[0021] After step S5 and before step S6, the method further includes: using the optical system simulation module to obtain optical images corresponding to different error parameters, and using the corresponding optical images and error parameters as simulated sample data; preprocessing the simulated sample data, using the optical images of the preprocessed simulated sample data as input parameters and the error parameters as labels to pre-train the deep learning computing system; wherein the method used for preprocessing the simulated sample data is the same as the method used for preprocessing the sample data, and the parameters used for pre-training are the same as the parameters used for training.

[0022] The deep learning computing system includes a backbone neural network model and a head network containing fully connected layers. The backbone neural network model is configured to extract features from optical images, and the head network is configured to output the estimated results of error parameters.

[0023] The automatic adjustment method for X-ray focusing systems based on deep learning of the present invention utilizes the high adaptive optimization capability of neural networks and adopts a deep learning method based on optical image information feedback to realize the automatic adjustment of X-ray focusing systems. Compared with traditional X-ray focusing system adjustment methods, the automatic adjustment method for X-ray focusing systems based on deep learning of the present invention does not require human intervention and only requires the assistance of other focal spot quality measurement methods during the data acquisition stage, which can achieve rapid and automated adjustment of X-ray focusing systems with high precision. Attached Figure Description

[0024] Figure 1 This is an overall flowchart of an automatic adjustment method for an X-ray focusing system based on deep learning, according to the present invention.

[0025] Figure 2 This is a schematic diagram illustrating the working principle of an automatic adjustment method for an X-ray focusing system based on deep learning, according to the present invention.

[0026] Figure 3 This is an optical image acquired by the detector after step S5 of the automatic adjustment method for an X-ray focusing system based on deep learning according to the present invention.

[0027] Figures 4-6 This is a line graph comparing the estimated error parameters obtained after executing step S7 with the actual error in the automatic adjustment method for X-ray focusing systems based on deep learning according to the present invention. Figures 4-6 The results of three experiments are shown. Detailed Implementation

[0028] The preferred embodiments of the present invention are given below with reference to the accompanying drawings and described in detail.

[0029] like Figure 1 The present invention discloses an automatic adjustment method for an X-ray focusing system based on deep learning, which includes the following steps:

[0030] Step S1, providing an X-ray optical imaging system and a deep learning computing system 50;

[0031] like Figure 2 As shown, the X-ray optical imaging system includes, but is not limited to: an X-ray focusing device 10 and a detector system 30 arranged along the incident direction of the X-ray beam, and a parameter adjustment system 20 connected to the X-ray focusing device 10;

[0032] The various directions involved in the X-ray optical imaging system are defined as follows: the direction along the incident direction of the X-ray beam is the Y direction; the direction perpendicular to the incident direction of the X-ray beam in the horizontal plane is the X direction; and the vertically upward direction perpendicular to the horizontal plane is the Z direction.

[0033] The X-ray focusing device 10 is configured to receive the incident X-ray beam and generate a focused X-ray beam; at least one parameter of the X-ray focusing device 10 is adjustable. The parameters of the X-ray focusing device include, but are not limited to, the X-coordinate, Y-coordinate, Z-coordinate, tilt angle (i.e., in-plane rotation angle), projection angle (i.e., incident tilt angle), roll angle, curvature, etc., of each optical element of the X-ray focusing device. In-plane refers to the transmission or reflection interface of the optical element, and tilt angle is the grazing incident angle of the light rays to the interface of the optical element.

[0034] The parameter adjustment system 20 is configured to adjust the adjustable parameters of the X-ray focusing device 10 by changing its driving parameters.

[0035] The parameter adjustment system 20 includes, but is not limited to, at least one of the following: an X-direction motor, a Y-direction motor, a Z-direction motor, a swing angle adjustment motor, a projection angle adjustment motor, a roll angle adjustment motor, and a curvature adjustment device (such as a bending motor) for each optical element of the X-ray focusing device.

[0036] The parameters vary for different X-ray focusing devices. For example, Fresnel zone plates do not include the bending coefficient parameter. They generally include three translational orientations of the focusing element (i.e., X-coordinate, Y-coordinate, Z-coordinate) and three angular orientations (i.e., swing angle, projection angle, roll angle). This allows us to determine the relative orientations between different optical elements.

[0037] When the optical element of the X-ray focusing device 10 is a mirror, its bending motor is configured to bend the centerline of the mirror relative to both sides along the interface normal direction to provide the required curvature. Thus, for the direction along the light path (i.e., the direction of mirror bending), the magnitude of the mirror deformation gradually decreases from both sides to the middle, and for the direction perpendicular to the light path, the magnitude of the mirror deformation is consistent.

[0038] In this embodiment, the X-ray focusing device 10 is a KB reflector, and the parameter adjustment system 20 is a KB reflector parameter adjustment system.

[0039] The incident X-ray beam is an unfocused hard X-ray emitted by a synchrotron radiation source. It is obtained by generating a secondary source through a secondary slit downstream of the source and then incident on the KB reflector.

[0040] The KB reflector (Kirkpatrick-Baez reflector) contains a pair of mutually perpendicular reflecting and focusing mirrors (including vertical and horizontal mirrors), which are configured to receive the incident X-ray beam generated by the secondary slit source and produce a focused X-ray beam.

[0041] When the X-ray focusing device 10 is a KB reflector, the two mirrors of the X-ray focusing device 10 each have their own tilt angle (i.e., in-plane rotation angle), projection angle (i.e., incident tilt angle), and roll angle. Taking the KB reflector as an example, in addition to the X-coordinate, Y-coordinate, Z-coordinate, tilt angle, and curvature parameters of each optical element, there is also a relative roll angle parameter between the two mirrors. The roll angle is generally 90°, making the two mirrors perpendicular to each other and focusing X-rays in the vertical and horizontal directions respectively. The in-plane rotation angle has a high tolerance for error and generally does not require special adjustment, so it is not considered an adjustable parameter. Therefore, the parameter adjustment system 20 of the KB reflector includes a mirror beam incident tilt angle adjustment system and a mirror curvature bending system. By changing the driving parameters of the KB reflector parameter adjustment system, the incident attitude and bending degree of the reflector can be adjusted, that is, the adjustable parameters of the X-ray focusing device 10 can be adjusted.

[0042] The detector system 30 is configured to receive an optical image of the focused X-ray beam. The detector system 30 includes an image detector, which is positioned behind the X-ray focusing device 10 when the light enters the optical path.

[0043] In this embodiment, the image detector is a CMOS (Complementary Metal Oxide Semiconductor) image detector, positioned in the far field behind the focal point of the X-ray focusing device 10 in the optical path of the X-ray beam. In some embodiments, the image detector can be switched between entering and exiting the optical path.

[0044] The optical image received by the image detector is used as an observation result to infer the calibration error, that is, as the input to the neural network.

[0045] In some embodiments, the X-ray optical imaging system further includes a scattering film 40 disposed between the parameter adjustment system 20 and the detector system 30, thereby employing the scattering film to enhance the characteristics of the optical image, making it easier for the detector system 30 to detect the optical image. In this embodiment, the scattering film is disposed at the focal point of the X-ray focusing device 10, so the scattering film 40 can provide a defocusing error estimate in addition to enhancing the characteristics of the optical image.

[0046] In other embodiments, the scattering film 40 may be replaced by other optical image enhancement devices disposed between the parameter adjustment system and the detector system for increasing the features of the optical image. These optical image enhancement devices may be scattering, transmitting, or reflecting devices. Alternatively, the optical image enhancement device may be omitted, and the features of the optical image may be enhanced by methods such as prolonged exposure when the detector system acquires the optical image. Or, no optical image enhancement device or method may be used.

[0047] The scattering film 40 comprises a film with particles, which is fixed within a sample frame on a support. The particle size depends on the focal spot size and can be nano- or micro-particles (generally, the particle size is set so that multiple particles exist within the focal spot area; nano-particles are typically used for focal spots of one micrometer or smaller, while micro-particles can be used for focal spots of tens of micrometers or larger). The particles have a large refractive index difference compared to air; generally, a larger refractive index difference is better. The film can be processed using various techniques such as deposition to obtain the scattering film 40, as long as the particles are uniformly distributed on the film.

[0048] In this embodiment, the particles are 500-nanometer copper particles, and the film is an optical tape. Accordingly, the preparation method of the scattering film 40 includes the following steps:

[0049] Step S101: Copper particles with a size of 500 nanometers are evenly spread on the optical tape;

[0050] Step S102: Adhere the optical tape containing the copper particles and cut it off;

[0051] Step S103: Install the cut optical tape into a sample frame of a support and bond it with adhesive.

[0052] Step S2: Provide a focal spot quality measurement device and a corresponding focal spot quality measurement method for acquiring focal spot quality data; and / or provide an optical system simulation module for acquiring focal spot quality data, wherein the optical system simulation module is configured to adjust the values ​​of the driving parameters of the system according to the parameters in its simulated object to output a simulated detector optical image, thereby acquiring focal spot quality data;

[0053] In this embodiment, the focal spot quality data generally refers only to the focal spot size. However, in other embodiments, the focal spot quality data may include the light intensity distribution within the focal spot, so it is not explicitly specified. This is also an optimization target of the present invention.

[0054] There are many types of devices for measuring the quality of focal spots, such as knife-edge scanning devices, pencil beams, speckle interferometry, Hartmann sensors, and so on.

[0055] In this embodiment, the simplest and most commonly used knife-edge scanning device is used as the focal spot quality measuring device, and the knife-edge scanning method is used as the corresponding focal spot quality measuring method.

[0056] The blade scanning device includes a blade, an adjustable sample stage, and a fluorescence detector. The blade is positioned at the focal point of the X-ray focusing device 10 of the X-ray optical imaging system, between the X-ray focusing device 10 and the detector system 30, at a 45-degree angle to the incident direction of the X-ray beam (i.e., the Y-direction). The fluorescence detector is positioned relative to the blade along a horizontal direction perpendicular to the incident direction of the X-ray beam (i.e., the X-direction), with the line connecting it to the blade perpendicular to the incident direction of the X-ray beam (i.e., the Y-direction), to receive the fluorescence emitted by the blade.

[0057] The blade is made of a straight, smooth, thin copper wire, which is mounted on the sample frame of the support.

[0058] The adjustable sample stage is installed at the bottom of the blade and includes an X-direction motor, a Y-direction motor and a Z-direction motor arranged sequentially from top to bottom.

[0059] The fluorescence detector can be placed in the optical path simultaneously with the detector system 30, because it is positioned on the side of the optical path relative to the knife edge, and it detects scattered fluorescence. For the knife-edge scanning method, although the detector system 30 is not used, it is not necessary to remove it from the optical path; the position of the detector system 30 does not affect the knife-edge scanning method. However, in step S3 below, when determining the optimal driving parameters, whether to remove the detector system 30 from the optical path depends on whether the focal spot quality measuring device requires the detector system 30 and whether the detector system 30 affects the optical path of the focal spot quality measuring device.

[0060] The steps of the blade scanning method include:

[0061] Step S201: Take one of the directions perpendicular to the incident direction of the X-ray beam (including the X and Z directions) as the direction to be measured, control the blade to move along the direction to be measured, so that the reading of the fluorescence detector rises from near zero, and record the position corresponding to when the reading starts to rise, which is the cutting position.

[0062] Step S202: Move the blade back a fixed distance from the cutting position along the direction to be measured;

[0063] Step S203: Starting from the current position of the blade, scan along the direction to be tested. During the scanning process, record the position of each blade and the corresponding reading of the fluorescence detector. After the reading of the fluorescence detector stops increasing, the blade stops after moving a fixed distance.

[0064] The fixed distance is generally to allow a certain margin for the cutting range (the light transmitted increases when the blade is removed, which is reflected in the increase of the reading until it is completely removed and the reading returns to a stable state; the range of the reading increase is the cutting range), which is generally 1 / 10 of the cutting range.

[0065] Step S204: Extract the sloping portion of the obtained blade movement distance-fluorescence detector reading curve and differentiate it to obtain the corresponding derivative curve;

[0066] Step S205: Perform Gaussian fitting on the obtained derivative curve and calculate the half-width at half-maximum of the fitting result, which is the focal spot size in the measured direction.

[0067] Since the cutting edge is a surface, the entire focal spot surface is gradually exposed as the cutting edge moves, so the final calculation is the average diameter of the light spot.

[0068] Since the blade is placed at a 45-degree angle to the incident direction of the X-ray beam, if the direction to be measured is the X-direction, the focal spot size measured in the X-direction is divided by 2 to obtain the final focal spot size, which is then used as the focal spot quality data. This is merely a special case in this specific embodiment.

[0069] The optical system simulation module can utilize existing mature simulation solutions (such as OASYS) or allow users to write their own programs based on established simulation principles. The simulated object in the optical system simulation module is unlikely to perfectly match the real-world imaging system; there is no foolproof method. The simulated object in the optical system simulation module includes an X-ray optical imaging system. The parameter adjustment system 20 within this simulated object is adjustable. The simulated detector optical image is obtained by setting the driving parameters of the parameter adjustment system 20. This simulated detector optical image is theoretically the detector optical image that the X-ray optical imaging system's detector system should receive. For the optical system simulation module, the driving parameters of the parameter adjustment system 20 must be accurate; therefore, the optical system simulation module can provide accurate labels for the driving parameters of the parameter adjustment system 20.

[0070] It should be noted that the focal spot quality measuring device and optical system simulation module in step S2 are only used for initial calibration by providing focal spot quality data. The calibration results are optimized and finally used to provide the optimal parameters of the parameter adjustment system 20 as labels for the training of the neural network. They are not related to the optical images actually received by the detector system 30.

[0071] Step S3, preliminary calibration, includes: if step S2 provides a focal spot quality measuring device, then the focal spot quality measuring device is installed in the X-ray optical imaging system. When the driving parameters of the parameter adjustment system 20 are different, the focal spot quality measuring device and the focal spot quality measuring method of step S2 are used to obtain focal spot quality data, and the optimal focal spot quality is selected. The corresponding optimal driving parameters of the parameter adjustment system 20 are recorded. If step S2 provides an optical system simulation module, then the optical system simulation module is used to obtain focal spot quality data corresponding to different driving parameters of the parameter adjustment system 20, and the optimal focal spot quality is selected. The corresponding optimal driving parameters of the parameter adjustment system 20 are recorded.

[0072] Before installing the focal spot quality measuring device into the X-ray optical imaging system, the method may further include: removing the detector system 30 from the optical path when the detector system 30 is not needed and the detector system 30 affects the optical path of the focal spot quality measuring device. Whether or not the detector system 30 is removed from the optical path depends on whether the focal spot quality measuring device needs the detector system 30 and whether the detector system 30 affects the optical path of the focal spot quality measuring device.

[0073] The parameter types of the parameter adjustment system 20 optimized to the best driving parameters can include all driving parameter types in the parameter adjustment system 20, or only the driving parameter types of the parameter adjustment system 20 related to the adjustable parameters of the X-ray focusing device 10 of interest to the adjuster. It should be noted that the driving parameters of the mirror parameter adjustment system 20 have already been adjusted to near their optimal parameters before the experiment.

[0074] In this embodiment, since step S2 provides a focal spot quality measuring device and the focal spot quality measuring device is a knife-edge scanning device, step S3 specifically includes:

[0075] Step S301: Place the blade scanning device, adjust the position of the blade along the incident direction of the X-ray beam (i.e., the Y direction) using the adjustable sample stage, and measure the focal spot size at different positions using the blade scanning method to determine the blade position corresponding to the smallest spot size. The blade position at this time is the focal point of the X-ray focusing device 10.

[0076] In step S302, the blade is placed at the focal point of the X-ray focusing device 10, and the driving parameters of the parameter adjustment system 20 are adjusted sequentially. When the driving parameters of the parameter adjustment system 20 are adjusted differently, the focal spot size is obtained by using the blade scanning device and the blade scanning method respectively. The smallest focal spot size is selected from them, and the corresponding optimal driving parameters of the parameter adjustment system 20 are recorded.

[0077] In this embodiment, the X-ray focusing device 10 is a KB reflector. The parameter types of the parameter adjustment system 20 optimized to the best driving parameters include the parameters of the mirror beam incident tilt angle adjustment system for each of the mutually perpendicular reflecting focusing mirrors and the parameters of the mirror curvature bending system, thereby optimizing the projection angle and curvature of each reflecting focusing mirror of the X-ray focusing device 10. Since the vertical and horizontal mirrors are responsible for focusing in mutually perpendicular directions, a grid-like search can be used to perform the pose parameters of the vertical and horizontal mirrors respectively.

[0078] The parameter adjustment system 20 may also include other types of driving parameters, but in this embodiment, the parameter types optimized to the best driving parameters are only the parameters of the mirror beam incident tilt angle adjustment system and the mirror curvature bending system for each reflecting focusing mirror, and other parameters of the reflecting focusing mirror are not optimized. In addition to the mirror parameters, in this embodiment, since a scattering film is also used, the defocusing error is optimized. The defocusing error is the distance between the Y-direction position of the scattering film and the focal point of the X-ray focusing device 10.

[0079] Step S4: Set the parameter adjustment range of the parameter adjustment system 20 with the optimal driving parameter of the parameter adjustment system 20 as the center;

[0080] In step S4, based on the accuracy requirements of the optical system and the error adjustment accuracy of the parameter adjustment system 20, the parameter adjustment range of the parameter adjustment system 20 is set with the optimal driving parameter of the parameter adjustment system 20 as the center.

[0081] The parameter adjustment range of the parameter adjustment system 20 can generally be set to about 20 times the error adjustment accuracy of the parameter adjustment system 20, depending on specific needs. This determines the parameter adjustment range of the parameter adjustment system 20 centered on the optimal driving parameter. The error adjustment accuracy is generally N, but can be slightly changed depending on the situation (e.g., the accuracy that can be accurately adjusted may be slightly greater than N) and requirements.

[0082] Different X-ray focusing systems and different beamlines have different upper and lower limit requirements. For systems where the focal spot size is the primary optimization target, the range is mainly determined by the target focal spot size. For a certain parameter, the upper and lower limits of the error tolerance (i.e., ±N) are generally determined by the maximum parameter error (denoted as N) obtained through theoretical calculation (usually using assumptions of coherent light or geometric optics) when the focal spot size is increased to 1.1 times the target focal spot size (i.e., increased by 10%), assuming that other parameters are accurate. Therefore, according to the accuracy requirements of the optical system, the parameter adjustment range of the parameter adjustment system 20 is usually the parameter adjustment range of the parameter adjustment system 20 corresponding to the focal spot size being at most 1.1 times the target focal spot size, thereby determining the parameter adjustment range of the parameter adjustment system 20.

[0083] Therefore, the parameter adjustment range of the final parameter adjustment system 20 can be the minimum, maximum or average value of the parameter adjustment range of the parameter adjustment system 20 when the error adjustment accuracy of the parameter adjustment system 20 is 20 times and the focal spot size is at most 1.1 times the target focal spot size, and the parameter adjustment interval of the parameter adjustment system 20 is determined accordingly.

[0084] Step S5: Select multiple values ​​of the driving parameters of the parameter adjustment system 20 within the parameter adjustment range of the parameter adjustment system 20, and record the difference between the driving parameters of the parameter adjustment system 20 and the optimal driving parameters, and the optical image acquired by the detector system 30, as sample data. The difference between the driving parameters of the parameter adjustment system 20 and the optimal driving parameters is used as the error parameter.

[0085] One of the optical images acquired in step S5 is as follows: Figure 3 As shown.

[0086] The number of selected parameter values ​​should be large enough, and it is even possible to iterate through the parameter values ​​of the driving parameters of the parameter adjustment system 20 to obtain sufficient sample data.

[0087] When the X-ray optical imaging system includes a scattering film 40, step S5 further includes: adjusting the parameter value of the driving parameter of the parameter adjustment system 20 for each parameter, adjusting the Y-direction position of the scattering film with the focal point of the X-ray focusing device 10 as the center, and recording the distance between the Y-direction position of the scattering film and the focal point of the X-ray focusing device 10 as sample data while recording the optical image acquired by the detector system 30. The difference between the parameter value of the driving parameter of the parameter adjustment system 20 and the optimal driving parameter, as well as the distance between the Y-direction position of the scattering film and the focal point of the X-ray focusing device 10 (i.e., defocusing error) are all used as error parameters.

[0088] In this embodiment, step S5 includes:

[0089] Step S501: Remove the blade and place the scattering film at the focal point;

[0090] Step S502: Within the parameter adjustment range of the X-ray focusing device parameter adjustment system 20, the parameter values ​​of the driving parameters of the parameter adjustment system 20 and the Y-direction position of the scattering film are traversed by sequential adjustment. The difference between the parameter value of the driving parameters of the parameter adjustment system 20 and the optimal driving parameters, the distance between the Y-direction position of the scattering film and the focal point of the X-ray focusing device 10, and the optical image acquired by the detector system 30 are recorded as sample data.

[0091] The scattering film has different defocusing errors at different positions in the Y direction.

[0092] Step S6: The sample data recorded in step S5 is preprocessed. The optical image of the preprocessed sample data is used as the input parameter, and the error parameter is used as the label to train the deep learning computing system 50 to obtain the X-ray focusing system parameter adjustment model. The deep learning computing system 50 includes a backbone neural network model and a head network containing fully connected layers. The backbone neural network model is configured to extract features from the optical image, and the head network is configured to output the estimated result of the error parameter.

[0093] Generally, a high-performing neural network model in image classification is used as the backbone neural network model. Specifically, the best-performing neural network model in image classification within the computational capabilities of the computing system can be selected as the backbone. The input parameters of the backbone neural network model are the optical images acquired by the detector system 30, and the output is the image features. The input of the head network is the image features output by the backbone neural network model, and the output of the head network is the estimated result of the error parameters, indicating how many error parameters need to be estimated, i.e., how many output parameters are set.

[0094] In this embodiment, the deep learning computing system 50 used is as follows: the backbone neural network model is a ResNet network model, and the head network consists of DO layers (Dropout layers) and fully connected layers. The DO layers prevent overfitting of the neural network during training by randomly discarding some features; the fully connected layers convert the features extracted by the backbone network into estimates of the mirror's error parameters.

[0095] The preprocessing that can be performed includes, but is not limited to: normalizing the error parameters of the parameter adjustment system 20 of the X-ray focusing device, cropping the optical image to remove the background, and resampling it to a suitable image size. When the X-ray optical imaging system includes a scattering film 40, the preprocessing also includes: normalizing the error parameters (i.e., defocusing error) corresponding to the scattering film 40.

[0096] The appropriate image size depends on the computing power available to the implementer, the number of detector pixels, and the required accuracy. For neural networks, the initial image input size is generally uniformly 224×224 pixels. However, as time has passed and computing power has increased, high-resolution images have been gradually incorporated into research, and image sizes have been continuously increasing. Generally, higher-resolution images contain more features, resulting in higher training accuracy, but also requiring longer training time and a larger network size. Currently, image pixels are generally below 1024×1024, and even "high-resolution" images are typically around 500×500. Images need to contain light intensity features, and the more prominent the features, the better. In this embodiment, the image size is resampled to 500×500.

[0097] During training, the Adam optimization method (an optimized stochastic gradient descent method) was used, with a batch size of 10 and a learning rate of 10. -4 The learning rate was reduced to 0.98 after each epoch, and a total of 80 epochs were trained. During training, the parameters of the best-performing deep learning computing system 50 on the validation set were recorded.

[0098] After step S5 and before step S6, the method may further include: using the optical system simulation module to obtain optical images corresponding to different error parameters, and using the corresponding optical images and error parameters as simulation sample data.

[0099] The simulated sample data is preprocessed, and the optical image of the preprocessed simulated sample data is used as the input parameter, and the error parameter is used as the label to pre-train the deep learning computing system 50.

[0100] The preprocessing method for the simulated sample data is the same as that for the sample data mentioned above, and the parameters used in the pre-training are the same as those used in the training.

[0101] Since the amount of data collected in experiments is limited, pre-training can use a large amount of simulated data to train the network's feature extraction and parameter fitting capabilities. Combining this with experimental data for training can improve the convergence speed. If the simulation is accurate enough, the estimation accuracy can be improved.

[0102] Step S7: When using the X-ray optical imaging system, the optical image currently acquired by the detector system 30 is processed based on the X-ray focusing system parameter adjustment model to obtain the estimation result of the error parameter. The parameter adjustment system 20 is then used to adjust the parameters according to the estimation result of the error parameter to improve the quality of the focal spot.

[0103] In this embodiment, the error parameters include: the pose calibration error of the reflector (i.e., the difference between the parameter value of the driving parameter of the parameter adjustment system 20 and the optimal driving parameter when the X-ray focusing device 10 is a KB reflector) and the defocusing error of the sample holder where the scattering film is located (the distance between the Y-direction position of the scattering film and the focal point of the X-ray focusing device 10).

[0104] like Figure 4-6 The estimated results obtained by the mirror pose estimation model under three different experiments are shown. Among them, parameter number 1 is the incident angle error of the vertical mirror, 2 is the incident angle error of the horizontal mirror, 3 is the curvature error of the vertical mirror, 4 is the curvature error of the horizontal mirror, 5 is the distance error between the two mirrors, and 6 is the defocusing error of the scattering film position.

[0105] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of the invention. Various variations can be made to the above embodiments of the present invention. All simple and equivalent changes and modifications made based on the claims and description of this application fall within the protection scope of the claims of this patent. All aspects not described in detail in this invention are conventional technical content.

Claims

1. An automatic adjustment method for an X-ray focusing system based on deep learning, characterized in that, include: Step S1: Provide an X-ray optical imaging system and a deep learning computing system; The X-ray optical imaging system includes an X-ray focusing device and a detector system arranged along the incident direction of the X-ray beam, and a parameter adjustment system connected to the X-ray focusing device; Step S2: Provide a focal spot quality measurement device and a corresponding focal spot quality measurement method and / or optical system simulation module for acquiring focal spot quality data; Step S3: Obtain the focal spot quality data corresponding to the driving parameters of different parameter adjustment systems, select the best focal spot quality from them, and record the best driving parameters of the corresponding parameter adjustment system. Step S4: Set the parameter adjustment range of the parameter adjustment system with the optimal driving parameter of the parameter adjustment system as the center; Step S5: Select the parameter values ​​of multiple driving parameters within the parameter adjustment range of the parameter adjustment system, and record the difference between the driving parameters and the optimal driving parameters, as well as the optical image acquired by the detector system, as sample data. The difference between the driving parameters and the optimal driving parameters is used as the error parameter. Step S6: Preprocess the sample data, use the optical image of the preprocessed sample data as input parameter and the error parameter as label to train the deep learning computing system to obtain the X-ray focusing system parameter adjustment model. Step S7: When using the X-ray optical imaging system, the optical image currently acquired by the detector system is processed based on the X-ray focusing system parameter adjustment model to obtain the estimation results of the error parameters. The parameter adjustment system is then used to adjust the parameters according to the estimation results of the error parameters in order to improve the quality of the focal spot.

2. The automatic adjustment method for an X-ray focusing system based on deep learning according to claim 1, characterized in that, The X-ray focusing device is configured to receive the incident X-ray beam and generate a focused X-ray beam. The parameter adjustment system is configured to adjust the adjustable parameters of the X-ray focusing device by changing its driving parameters; The detector system is configured to receive an optical image of the focused X-ray beam.

3. The automatic adjustment method for an X-ray focusing system based on deep learning according to claim 1, characterized in that, If step S2 provides a focal spot quality measuring device, then in step S3, the focal spot quality measuring device is installed in the X-ray optical imaging system. When the driving parameters of the adjustment parameter system are different, the focal spot quality measuring device and the focal spot quality measuring method of step S2 are used to obtain focal spot quality data respectively. If step S2 provides an optical system simulation module, then in step S3, the optical system simulation module is used to obtain the focal spot quality data corresponding to the driving parameters of the different parameter adjustment system. The optical system simulation module is configured to output a simulated detector optical image by adjusting the values ​​of the driving parameters of the system according to the parameters, thereby obtaining focal spot quality data.

4. The automatic adjustment method for an X-ray focusing system based on deep learning according to claim 3, characterized in that, Step S2 provides a focal spot quality measuring device. In step S3, before installing the focal spot quality measuring device into the X-ray optical imaging system, the method further includes: removing the detector system out of the optical path when the detector system is not needed and the detector system affects the optical path of the focal spot quality measuring device.

5. The automatic adjustment method for an X-ray focusing system based on deep learning according to claim 1, characterized in that, In step S4, based on the accuracy requirements of the optical system and the error adjustment accuracy of the parameter adjustment system, the parameter adjustment range of the parameter adjustment system is set with the optimal driving parameter of the parameter adjustment system as the center.

6. The automatic adjustment method for an X-ray focusing system based on deep learning according to claim 1, characterized in that, The X-ray optical imaging system also includes an optical image enhancement device disposed between the parameter adjustment system and the detector system to increase the features of the optical image. The optical image enhancement device is a scattering, transmission, or reflection device.

7. The automatic adjustment method for an X-ray focusing system based on deep learning according to claim 6, characterized in that, The incident direction of the X-ray beam is the Y direction, the optical image enhancement device is a scattering thin film, and step S5 further includes: adjusting the driving parameter value of each parameter adjustment system with the focal point of the X-ray focusing device as the center to obtain different Y-direction positions of the scattering thin film, and recording the distance between the Y-direction position of the scattering thin film and the focal point of the X-ray focusing device while recording the optical image acquired by the detector system, as sample data, wherein the difference between the driving parameter value of the parameter adjustment system and the optimal driving parameter and the distance between the Y-direction position of the scattering thin film and the focal point of the X-ray focusing device are both used as error parameters.

8. The automatic adjustment method for an X-ray focusing system based on deep learning according to claim 1, characterized in that, Extended exposure is used when the detector system acquires optical images to enhance the features of the optical images.

9. The automatic adjustment method for an X-ray focusing system based on deep learning according to claim 1, characterized in that, After step S5 and before step S6, the method further includes: using the optical system simulation module to obtain optical images corresponding to different error parameters, and using the corresponding optical images and error parameters as simulation sample data; The simulated sample data is preprocessed, and the optical image of the preprocessed simulated sample data is used as the input parameter, and the error parameter is used as the label to pre-train the deep learning computing system. The method used for preprocessing the simulated sample data is the same as the method used for preprocessing the sample data, and the parameters used for pre-training are the same as the parameters used for training.

10. The automatic adjustment method for an X-ray focusing system based on deep learning according to claim 1, characterized in that, The deep learning computing system includes a backbone neural network model and a head network containing fully connected layers. The backbone neural network model is configured to extract features from optical images, and the head network is configured to output the estimated results of error parameters.

Images