A lobster-eye optical device calibration method based on a neural network

Abstract

The application discloses a lobster eye optical device calibration method based on a neural network, and comprises the following steps: simulating the simulation X-ray images of the lobster eye optical device under different structural characteristic parameters by using an optical model as a sample database; constructing a neural network about the X-ray images and the structural characteristic parameters of the lobster eye optical device; training the neural network by using the sample database and updating the neural network parameters; calculating the X-ray images of the real lobster eye optical device by using the trained neural network to obtain the structural characteristic parameters of the real lobster eye optical device; and correcting the optical model by using the structural characteristic parameters of the real lobster eye optical device to realize the calibration of the real lobster eye optical device. The application establishes the relationship between the structural characteristics and the imaging by using the neural network, combines the actual imaging results, obtains the real structural characteristic parameters, establishes the accurate optical model, obtains the accurate calibration result, improves the calibration accuracy compared with the traditional method, and shortens the calibration time.

Description

Technical Field

[0001] This invention relates to the field of X-ray lobster eye optical device technology, and more specifically to a lobster eye optical device calibration method based on neural networks. Background Technology

[0002] Lobster-eye telescopes with large field of view have significant applications in X-ray time-domain astronomy. Their optical system is composed of multiple lobster-eye optics (MPO) lenses. Each MPO lens consists of millions of square micro-apertures evenly distributed on a spherical surface, with the center of each aperture pointing towards the center of the sphere. When X-rays are incident on the MPO lens, the inner walls of the micro-apertures perform total internal reflection, focusing the grazingly incident X-rays. Therefore, the geometry of the micro-aperture inner walls has a significant impact on the imaging quality of the MPO lens. The fabrication process of MPO lenses is highly complex, and the intricate micro-aperture structure inevitably introduces structural defects. These defects cause deviations from the theoretical positions and angles of the micro-aperture inner walls, resulting in a degree of blurring in the focused image and reducing the optical performance of the MPO lens. The aperture diameter of MPO lens micro-apertures is typically tens of micrometers, with millions of micro-apertures on a single MPO lens. Simultaneously, the thickness of an MPO lens is on the order of millimeters, resulting in extremely small aperture ratios, making it extremely difficult to measure the internal structure of the micro-apertures using microscopic methods. Research on defects in MPO lenses generally focuses on imaging quality. It uses experimental methods to measure the focused image of X-rays by the MPO lens and analyzes information such as the point spread function of the image to measure the lens quality.

[0003] Calibration is a crucial part of telescope development. The calibration of a lobster-eye telescope mainly consists of optical response calibration and detector response calibration, which are coupled to form the overall telescope calibration. The optical system of a lobster-eye telescope is primarily composed of multiple lobster-eye optical elements pieced together; therefore, optical response calibration mainly involves calibrating the assembled lobster-eye optical elements. X-ray telescope calibration is typically performed in a beam apparatus, using the characteristic radiation of a metal target to test the optical response at different energies, and scanning at different angles within the field of view by adjusting the telescope's attitude. The accuracy of this method depends on the number of energy points and scanning angle points selected; the greater the number, the higher the accuracy. For lobster-eye telescopes with large fields of view, the required number of scanning angle points far exceeds that of ordinary X-ray telescopes. Due to practical engineering limitations, traditional methods are insufficient to obtain accurate performance calibration results for lobster-eye telescopes. Summary of the Invention

[0004] Objective of the Invention: This invention addresses the shortcomings of existing calibration techniques by providing a neural network-based calibration method for lobster-eye optical devices. Starting with the micropore structure characteristics of lobster-eye optical devices, this invention utilizes a neural network to obtain the relationship between structural feature parameters and focused images. This relationship is then combined with actual X-ray imaging results to obtain the micropore structure feature parameters of the MPO lens, thereby establishing a precise optical model of the MPO lens. Based on this model, accurate calibration results are obtained.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] A method for calibrating lobster eye optical devices based on neural networks, comprising:

[0007] Simulated X-ray images of lobster eye optical devices under different structural feature parameters were used as a sample database, simulating optical models.

[0008] Construct a neural network for X-ray images and structural feature parameters of lobster eye optics;

[0009] The neural network is trained using a sample database, and its parameters are updated.

[0010] The structural feature parameters of a real lobster eye optical device are obtained by using a trained neural network to calculate X-ray images of the device.

[0011] The optical model is corrected by using the structural characteristic parameters of real lobster eye optical devices to achieve the calibration of real lobster eye optical devices.

[0012] A further technical solution of the present invention is as follows: the simulated X-ray images of the lobster eye optical device under different structural feature parameters, using an optical model as a sample database, specifically include:

[0013] The optical model simulates the focused image of X-rays incident on a lobster-eye optical device using ray tracing. The structural feature parameters of the lobster-eye optical device are the element values ​​of eight M×N dimensional matrices. The lobster-eye optical device has M×N square micro-holes, and the eight matrices are the slope and intercept matrices of the four inner walls of the M×N micro-holes, respectively. Different structural feature parameters are used to characterize different processing errors of the lobster-eye optical device. The structural feature parameter values ​​are sample labels, and the energy, direction, and position information of the X-ray source in the simulated X-ray image and the optical model are the sample data to construct a sample database.

[0014] A further technical solution of the present invention is as follows: the construction of the neural network for X-ray images and structural feature parameters of lobster eye optical devices specifically includes:

[0015] The input layer of the neural network includes the pixel values ​​of m×n pixels of the X-ray image of the lobster eye optics and the energy, direction, and position information of the X-ray source. The X-ray image is the focused image formed by the lobster eye optics on the X-ray point source or parallel source. The output layer of the neural network consists of the structural feature parameters of the lobster eye optics, including the element values ​​of 8 M×N dimensional matrices.

[0016] A further technical solution of the present invention is as follows: the step of training the neural network using a sample database and updating the neural network parameters specifically includes:

[0017] The neural network uses simulated X-ray images of the optical model and the energy, direction, and position information of the X-ray source in the optical model as sample data, and the corresponding structural feature parameters of the lobster eye optical device as sample labels. The output parameters of the neural network are the estimated values ​​of the structural feature parameters of the lobster eye optical device. The loss function is defined by the difference between the estimated values ​​and the sample labels, and the neural network parameters are updated.

[0018] A further technical solution of the present invention is as follows: the method of using a trained neural network to calculate the X-ray image of a real lobster eye optical device to obtain the structural feature parameters of the real lobster eye optical device specifically includes:

[0019] Using real lobster eye optical devices to capture X-ray images of known X-ray sources, and combining the energy, direction, and position information of the X-ray source, the data is input into an updated neural network to obtain the structural feature parameter values ​​of the real lobster eye optical devices.

[0020] A further technical solution of the present invention is as follows: the calibration of the real lobster eye optical device by correcting the optical model using the structural feature parameters of the real lobster eye optical device specifically includes:

[0021] By utilizing the structural characteristic parameters of a real lobster eye optical device, modifying the theoretical parameters in the optical model, and setting different X-ray source energies, directions, and positions in the model, the performance of the real lobster eye optical device under different working conditions is obtained, thus achieving the calibration of the real lobster eye optical device.

[0022] The beneficial effects of this invention are:

[0023] A neural network-based calibration method for lobster-eye optics is proposed. This method utilizes a neural network to establish a mapping relationship between the structural feature parameters of the lobster-eye optics and X-ray images. Simulated X-ray images of the optical model and the energy, direction, and position information of the X-ray source in the optical model are used as sample data. The corresponding structural feature parameter values ​​of the lobster-eye optics are then used as sample labels to update the neural network parameters. The updated neural network is then used to calculate the structural feature parameters of the actual lobster-eye optics from the X-ray images of the real lobster-eye optics. This corrects the optical model and achieves the calibration of the actual lobster-eye optics. Therefore, the proposed lobster-eye optics calibration method can obtain accurate calibration results under different operating conditions. Attached Figure Description

[0024] Figure 1 This is a flowchart of a lobster eye optical device calibration method based on neural networks according to the present invention;

[0025] Figure 2 This is a schematic diagram of the inner wall structure of a single microchannel in the lobster eye optical device of the present invention;

[0026] Figure 3 Figure (a) shows the structural feature parameters of the inner wall of a single microchannel in the lobster eye optical device of the present invention, and Figure (b) shows the structural feature parameters of the upper and lower inner walls.

[0027] Explanation of reference numerals in the attached figures:

[0028] 201 - Lobster eye optical device; 202 - Single microchannel; 301 - Equation of the plane containing the upper inner wall of the microchannel; 302 - Equation of the plane containing the lower inner wall of the microchannel; 303 - Equation of the plane containing the right inner wall of the microchannel; 304 - Equation of the plane containing the left inner wall of the microchannel. Detailed Implementation

[0029] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0030] It should also be noted that, in order to avoid obscuring the present invention with unnecessary details, only the structures and / or processing steps closely related to the present invention are shown in the accompanying drawings, while other details that are not closely related to the present invention are omitted.

[0031] Additionally, it should be noted that the terms “comprising,” “including,” or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0032] Please see Figure 1 As shown, the present invention proposes a calibrating method for lobster eye optical devices based on neural networks, comprising the following steps:

[0033] S1. Establish an optical model;

[0034] An optical model is established based on the ray tracing method. The light source information in the model includes the position of the light source, the energy of the photons, and the vector of the emitted light rays. Combined with the principle of grazing incidence total internal reflection of X-rays, the starting position and direction of the emitted light rays after passing through the lobster eye optical device are calculated, thereby obtaining the simulated imaging results at the image plane position.

[0035] S2. Establish the structural feature parameter matrix of the lobster eye optical device;

[0036] The geometry of lobster eye optical devices is as follows Figure 2 As shown, a diagram contains M×N square micropores, each with four inner walls (top, bottom, left, and right). Projecting the top and bottom inner walls onto the YOZ plane and the left and right inner walls onto the XOZ plane yields the equations of the planes containing the four inner walls. These planes can be represented by the slope k and intercept b of a straight line, as shown below. Figure 3 As shown, the structural features of the lobster-eye optical device can be characterized by eight matrices, namely:

[0037]

[0038] and

[0039]

[0040] These eight matrices constitute the structural feature parameters of the lobster-eye optical device. The deviation between these parameters and the theoretical values ​​represents the manufacturing error of the lobster-eye optical device. In the optical model, by combining the equation of the incident ray and the equation of the plane containing the inner wall, the equation of the outgoing ray can be calculated. By modifying the structural feature parameters and the light source information, the imaging results of different lobster-eye optical devices under various working conditions can be simulated. The structural feature parameters of the lobster-eye optical device, the light source information, and the corresponding simulated imaging are recorded as a set of samples. By modifying the parameters, a large number of simulations are performed to form a sample database.

[0041] S3. Construct a neural network;

[0042] A neural network is constructed, with the input layer including the pixel values ​​of m×n pixels of the X-ray image, the energy value of the X-ray source, the direction value of the X-ray source, and the position value of the X-ray source, and the output layer containing the structural feature parameters of the lobster eye optical device.

[0043] S4. Train the neural network;

[0044] The neural network outputs simulated X-ray images and energy, direction and position information of X-ray sources from the sample database as sample data. The output parameters are estimated values ​​of structural feature parameters of the lobster eye optical device. The corresponding structural feature parameters of the lobster eye optical device in the sample database are sample labels. The loss function is defined by the difference between the estimated values ​​and the sample labels. The backpropagation algorithm is used to update the neural network parameters.

[0045] S5. Conduct imaging tests on real lobster eye optical devices;

[0046] The lobster-eye optical device is illuminated by light emitted from a known X-ray source. Information such as the source energy and flux of the X-ray source is obtained based on information such as the target material, voltage, and current of the X-ray source. Information such as the direction and position of the X-ray source is obtained based on the relative positional relationship between the X-ray source and the lobster-eye optical device. The imaging result of the lobster-eye optical device is received by an X-ray detector at the image plane position, and an X-ray focused image of m×n pixels is output.

[0047] S6. Obtain the structural feature parameters of real lobster eye optical devices using neural networks;

[0048] The X-ray image of a real lobster eye optical device and the light source information are input into a trained neural network, and the structural feature parameters of the real lobster eye optical device are output.

[0049] S7. Calibrate real lobster eye optical devices using structural feature parameters;

[0050] By utilizing the structural characteristic parameters of a real lobster-eye optical device, the structural characteristic parameters of the lobster-eye optical device in the optical model are modified, making the lobster-eye optical device in the optical model a digital counterpart of the real lobster-eye optical device. By setting different X-ray source energies, directions, and positions in the model, the performance of the lobster-eye optical device under different working conditions is obtained, thus achieving accurate calibration of the real lobster-eye optical device.

[0051] The following detailed embodiments further illustrate the calibration method for lobster eye optical devices based on neural networks according to the present invention:

[0052] Example

[0053] This embodiment proposes a calibration method for lobster eye optical devices based on neural networks, including the following steps:

[0054] S1. Establish the optical model:

[0055] In this embodiment, an optical model is established. The center coordinates of the lobster-eye optical device are (0, 0, 0), in mm, with dimensions of 42.5×42.5×2.5mm, a radius of curvature of 750mm, and a focal length of 375mm. The X-ray source is parallel light, with center coordinates of (10000, 0, 0), in mm, a photon energy of 1keV, a principal axis vector of (-1, 0, 0), a divergence angle of 0, and a spot diameter of 50mm. A detector is set at the focal plane of the lobster-eye optical device to receive the focused photons. The detector center coordinates are (-375, 0, 0), in mm, with a pixel size of 1024×1024 and a single pixel size of 15×15μm.

[0056] S2. Establish the structural feature parameter matrix of the lobster eye optical device:

[0057] The lobster-eye optical device has a micropore size of 40×40μm and a wall thickness of 8μm. First, the theoretical values ​​of the slope and intercept of each inner wall were calculated. Then, a certain degree of error was added to the values ​​to obtain the structural feature parameter matrix. The matrix has dimensions of 885×885, and the values ​​of the eight matrices are as follows:

[0058]

[0059]

[0060] and

[0061] The simulated imaging of the lobster eye optical device under the structural feature parameters is calculated using an optical model. The resulting image, X-ray source information, and structural feature parameters are used as a set of samples. By modifying the structural feature parameters or X-ray source information, more corresponding simulated images are obtained, and a sample database is constructed. In this embodiment, a total of 1,000 sets of samples are generated as a training set and 100 sets of samples are generated as a validation set.

[0062] S3. Constructing a neural network:

[0063] The input layer of this embodiment includes the pixel values ​​of the 1024×1024 pixels of the X-ray image, the energy value of the X-ray source, the direction value of the X-ray source, and the position value of the X-ray source. The output layer is the structural feature parameters of the lobster eye optical device. In this embodiment, the neural network is a multi-input neural network. The convolutional neural network (CNN) is used to process the X-ray image data, and the fully connected layer neural network is used to process the X-ray source information. The image features and the source information features are fused through the feature fusion layer, and after further processing, the micropore structure feature parameters are output.

[0064] S4. Training the neural network:

[0065] The simulated X-ray images and the energy, direction, and position information of the X-ray source in the sample database are used as sample data output. The output parameters are the estimated values ​​of the structural feature parameters of the lobster eye optical device. The corresponding structural feature parameters of the lobster eye optical device in the sample database are the sample labels. The mean square error (MSE) between the estimated value and the sample label is defined as the loss function. The backpropagation algorithm is used to update the neural network parameters.

[0066] S5. Imaging test of real lobster eye optical components:

[0067] Imaging tests were conducted on a real lobster-eye optical device in an X-ray beam apparatus. The real lobster-eye device has dimensions of 42.5×42.5×2.5mm, a radius of curvature of 750mm, a focal length of 375mm, and a nominal micro-aperture size of 40×40μm with a wall thickness of 8μm. The lobster-eye optical device was fixed in a special fixture, and its position and orientation were adjusted so that the center of the lobster-eye optical device was aligned with the center of the X-ray source and the center of the detector, and the focal point of the lobster-eye optical device was located at the center of the detector. The X-ray source used a characteristic K-line of Mg metal target material with an energy of 1.25keV, and was located 10m away from the lobster-eye optical device with a divergence angle of 2°. The detector was located 361.4mm away from the lobster-eye optical device at its focal plane. The image of the X-rays emitted by the source after being focused by the lobster-eye optical device was recorded and output.

[0068] S6. Obtaining structural feature parameters of real lobster eye optical devices using neural networks:

[0069] The X-ray image of the real lobster eye optical device output by the detector and the light source information are input into the trained neural network, and the output result is the structural feature parameters of the real lobster eye optical device.

[0070] S7. Calibrate real lobster eye optical components using structural feature parameters:

[0071] By utilizing the structural characteristic parameters of a real lobster-eye optics device, the structural characteristic parameters of the lobster-eye optics device in the optical model are modified, making the lobster-eye optics device in the optical model a digital counterpart of the real lobster-eye optics device. The model simulates the imaging effect under parallel light incidence, and by comparing with experimental results, it can more accurately reflect the performance of the lobster-eye telescope in receiving X-rays emitted by distant celestial bodies in space. Furthermore, it can simulate the response of the lobster-eye telescope to X-rays of different energies, achieving precise calibration of the real lobster-eye optics device.

[0072] In summary, this invention proposes a neural network-based calibration method for lobster-eye optical devices. This method constructs a relevant dataset using simulated data from an optical model, then builds and optimizes a neural network. The optimized neural network is then used to obtain the structural feature parameters of a real lobster-eye optical device. These structural feature parameters reflect the manufacturing errors of the actual lobster-eye optical device. Combined with the optical model, the performance of the lobster-eye optical device under different operating conditions can be obtained. Compared to traditional methods, this neural network-based calibration method for lobster-eye optical devices improves calibration accuracy and reduces the calibration time, providing significant reference value in the development of lobster-eye telescopes.

[0073] The above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A calibration method for lobster-eye optical devices based on neural networks, characterized in that, The method includes: S1. Simulated X-ray images of lobster eye optical devices under different structural characteristic parameters are used as a sample database, simulating optical models. S2. Construct a neural network for X-ray images and structural feature parameters of lobster eye optics; Specifically: The input layer of the neural network includes the pixel values ​​of m×n pixels of the X-ray image of the lobster eye optical device and the energy, direction, and position information of the X-ray source. The X-ray image is the focused image formed by the lobster eye optical device on the X-ray point source or parallel source. The output layer of the neural network contains the structural feature parameters of the lobster eye optical device, including the element values ​​of 8 M×N dimensional matrices. The lobster eye optical device has M×N square micropores, and the 8 matrices are the slope values ​​and intercept values ​​of the four inner walls of the M×N micropores, respectively. S3. Train the neural network using the sample database and update the neural network parameters; S4. Calculate the structural feature parameters of the real lobster eye optical device using the trained neural network to calculate the X-ray image of the real lobster eye optical device. S5. The optical model is corrected by using the structural characteristic parameters of real lobster eye optical devices to achieve the calibration of real lobster eye optical devices.

2. The calibration method for lobster-eye optical devices based on neural networks as described in claim 1, characterized in that, The simulated X-ray images of lobster eye optical devices under different structural feature parameters, using an optical model, serve as a sample database, specifically including: The optical model simulates the focused image of X-rays incident on a lobster-eye optical device using ray tracing. The structural feature parameters of the lobster-eye optical device are the element values ​​of eight M×N dimensional matrices. The lobster-eye optical device has M×N square micro-holes, and the eight matrices are the slope and intercept matrices of the four inner walls of the M×N micro-holes, respectively. Different structural feature parameters are used to characterize different processing errors of the lobster-eye optical device. The structural feature parameter values ​​are sample labels, and the energy, direction, and position information of the X-ray source in the simulated X-ray image and the optical model are the sample data to construct a sample database.

3. The calibration method for lobster-eye optical devices based on neural networks as described in claim 1, characterized in that, The process of training the neural network using a sample database and updating the neural network parameters specifically includes: The neural network uses simulated X-ray images of the optical model and the energy, direction, and position information of the X-ray source in the optical model as sample data, and the corresponding structural feature parameters of the lobster eye optical device as sample labels. The output parameters of the neural network are the estimated values ​​of the structural feature parameters of the lobster eye optical device. The loss function is defined by the difference between the estimated values ​​and the sample labels, and the neural network parameters are updated.

4. The calibration method for lobster eye optical devices based on neural networks as described in claim 1, characterized in that, The method of using a trained neural network to calculate the structural feature parameters of a real lobster eye optical device from an X-ray image specifically includes: Using real lobster eye optical devices to capture X-ray images of known X-ray sources, and combining the energy, direction, and position information of the X-ray source, the data is input into an updated neural network to obtain the structural feature parameter values ​​of the real lobster eye optical devices.

5. The calibration method for lobster-eye optical devices based on neural networks as described in claim 1, characterized in that, The calibration of a real lobster-eye optical device by correcting the optical model using the structural feature parameters of the real lobster-eye optical device specifically includes: By utilizing the structural characteristic parameters of a real lobster eye optical device, modifying the theoretical parameters in the optical model, and setting different X-ray source energies, directions, and positions in the model, the performance of the real lobster eye optical device under different working conditions is obtained, thus achieving the calibration of the real lobster eye optical device.

Images