Adjustment method of off-axis parabolic mirror misalignment

Abstract

A method for adjusting the misalignment of an off-axis parabolic mirror, the method records the focal spot of the off-axis parabolic mirror once through an image sensor, analyzes the deformed focal spot using an image processing unit containing a deep learning model, obtains the misalignment error of the current off-axis parabolic mirror, and then achieves precise adjustment of the off-axis parabolic mirror through the multi-axis adjustment device. The method has the advantages of strong robustness, high precision, full automation, and high speed, and is conducive to the efficient adjustment of the off-axis parabolic mirror.

Description

Technical Field

[0001] The invention relates to an off-axis parabolic mirror, in particular to a method for adjusting the misalignment of an off-axis parabolic mirror. Background Art

[0002] Since the off-axis parabolic mirror is an off-axis reflective element without monochromatic aberration and nonlinear effect, it is very suitable as an optical element for imaging measurement and strong laser focusing, and is widely used in astronomy, aviation, high-power laser and other fields. However, compared with traditional non-off-axis optical elements, the tolerance of the off-axis parabolic mirror is very small, and the installation and adjustment accuracy is extremely high. A small misalignment error will cause serious degradation of the focal spot. Therefore, in actual experiments, high-quality adjustment of the off-axis parabolic mirror is required. The existing off-axis parabolic mirror calibration method is mainly based on the analysis of the focal spot intensity distribution. It is impossible to directly measure the misalignment according to the focal spot morphology. Therefore, the actual process requires repeated adjustments to gradually approach the optimal focal spot. The adjustment process is cumbersome and affects the experimental efficiency. Summary of the invention

[0003] In order to overcome the shortcomings of the existing off-axis parabolic mirror optical path misalignment measurement method, the present invention proposes an off-axis parabolic mirror misalignment adjustment method. The method has the advantages of strong robustness, high precision, full automation, and high speed, and is conducive to the efficient adjustment of the off-axis parabolic mirror.

[0004] The present invention is implemented by the following technical solutions:

[0005] A method for adjusting misalignment of an off-axis parabolic mirror, characterized in that the method comprises the following steps:

[0006] 1) An off-axis parabolic mirror is fixed to a control end of a multi-axis adjustment device, an image sensor is located at a focal position of the off-axis parabolic mirror, an output end of the image sensor is connected to an input end of an image processing unit containing a deep learning model, and an output end of the image processing unit containing a deep learning model is connected to an input end of the multi-axis adjustment device, and the output light of the input light source is focused through the off-axis parabolic mirror and incident on the image sensor to form a focal spot image;

[0007] 2) The image sensor inputs the focus spot image into the image processing unit containing the deep learning model;

[0008] 3) After processing the focal spot image, the image processing unit containing the deep learning model displays the error vector of N elements between the optical axis of the off-axis parabolic mirror corresponding to the current focal spot image and the optical axis of the input light source, and outputs it to the multi-axis adjustment device, wherein N is the adjustable dimension of the multi-axis adjustment device, and the element values ​​of the error vector of the corresponding dimensions of the error vectors X1, X2, ..., XN of the N elements are E1, E2, ..., EN;

[0009] 4) The multi-axis adjustment device adjusts the corresponding dimensions of the error vector of the N elements according to the element values ​​E1, E2, ..., EN, so that the error between the optical axis of the off-axis parabolic mirror and the optical axis of the input light source is minimized, thereby completing the adjustment of the misalignment of the off-axis parabolic mirror.

[0010] The method for processing the focal spot image by the image processing unit containing the pre-trained deep learning model comprises the following steps:

[0011] 1) Perform image segmentation preprocessing on the input focus spot image to segment the focus spot area from the large background area;

[0012] 2) Perform image enhancement processing on the segmented focus spot image to improve the resolution and adapt the algorithm model;

[0013] 3) Inputting the enhanced focus spot image into the pre-trained CNN regression network, outputting and displaying the error vector, each element of the error vector corresponds to an adjustable dimension of the multi-axis adjustment device, and the element value of the adjustable dimension is the error of the corresponding adjustment dimension of the multi-axis adjustment device.

[0014] The work of the image processing unit containing the pre-trained deep learning model before the CNN regression network pre-training includes the following steps:

[0015] 1) According to the parameters of the current experimental device and based on the light transmission model of the off-axis parabolic mirror, a data set of a focal spot intensity map corresponding to the adjustment error of the off-axis parabolic mirror is obtained;

[0016] 2) by adjusting the input beam parameters of the light transmission model of the off-axis parabolic mirror, such as the beam waist radius, the clear aperture, etc., a data set of the focal spot intensity map corresponding to the adjustment error of the off-axis parabolic mirror under different input light parameters is obtained;

[0017] 3) Before pre-training, the data set needs to be augmented according to scene requirements to obtain a data set of focal spot intensity maps corresponding to multiple adjustment errors of the off-axis parabolic mirror, including: the difference between the pixel size of the actual image sensor and the simulation, and scaling the data set; when adjusting the off-axis parabolic mirror, the focus of the image sensor is translated, and the data set image is translated.

[0018] The method for preparing the CNN regression network data set can establish a model of an off-axis parabolic mirror in optical software, use the physical optics propagation module of the software to output the focal spot image of the off-axis parabolic mirror with the same misalignment error, and save it as a data set; or the physical model of light transmission of the off-axis parabolic mirror can be established in the CNN regression network, and the focal spot images with different misalignment errors can be generated during training.

[0019] Compared with the prior art, the innovation of the present invention is:

[0020] 1. The method of the present invention performs misalignment measurement on the off-axis parabolic mirror based on a convolutional neural network. The precise adjustment of the misalignment error can be achieved through a single misalignment measurement and adjustment, thus avoiding the tedious problem of multiple repeated adjustments required by the traditional method and improving the experimental adjustment efficiency.

[0021] 2. The neural network model proposed by the method of the present invention can take into account the training of multiple input parameters, adapt to various experimental scenarios, and has strong robustness.

[0022] 3. The method of the present invention only requires a small number of optical elements to achieve single misalignment measurement of off-axis parabolic mirrors and is easy to integrate.

[0023] In summary, the method of the present invention has the advantages of strong robustness, high precision, full automation and high speed, which is conducive to the efficient adjustment of off-axis parabolic mirrors. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 Schematic diagram of the optical path of the off-axis parabolic mirror misalignment adjustment method of the present invention

[0025] Figure 2 This is a schematic diagram of a CNN regression network with a 4-layer convolutional unit and a 3-layer fully connected unit. DETAILED DESCRIPTION

[0026] In order to make the implementation purpose, technical scheme and technical effect of the present invention clearer, the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments. In the description of this application, it should be understood that the terms are only for the convenience of describing this application and simplifying the description, and therefore cannot be understood as limiting this application.

[0027] Please refer to Figure 1As can be seen from the figure, the off-axis parabolic mirror misalignment adjustment method of the present invention is characterized in that the method comprises the following steps:

[0028] 1) The off-axis parabolic mirror 2 is fixed to the control end of the multi-axis adjustment device 5, the image sensor 3 is located at the focal position of the off-axis parabolic mirror 2, the output end of the image sensor 3 is connected to the input end of the image processing unit 4 containing the deep learning model, and the output end of the image processing unit 4 containing the deep learning model is connected to the input end of the multi-axis adjustment device 5, and the output light of the input light source 1 is focused through the off-axis parabolic mirror 2 and incident on the image sensor 3 to form a focal spot image;

[0029] 2) The image sensor 3 inputs the focus spot image into the image processing unit 4 containing the deep learning model;

[0030] 3) After processing the focal spot image, the image processing unit 4 containing the deep learning model displays the error vector of N elements between the optical axis of the off-axis parabolic mirror 2 and the optical axis of the input light source 1 corresponding to the current focal spot image, and outputs it to the multi-axis adjustment device 5, wherein N is the adjustable dimension of the multi-axis adjustment device 5, and the element values ​​of the error vector of the corresponding dimensions of the error vectors X1, X2, ..., XN of the N elements are E1, E2, ..., EN;

[0031] 4) The multi-axis adjustment device 5 adjusts the corresponding dimensions of the error vector of the N elements according to the element values ​​E1, E2, ..., EN, so that the error between the optical axis of the off-axis parabolic mirror 2 and the optical axis of the input light source 1 is minimized, thereby completing the adjustment of the misalignment of the off-axis parabolic mirror 2.

[0032] The method for processing the focal spot image by the image processing unit 4 containing the pre-trained deep learning model comprises the following steps:

[0033] 1) Perform image segmentation preprocessing on the input focus spot image to segment the focus spot area from the large background area;

[0034] 2) Perform image enhancement processing on the segmented focus spot image to improve the resolution and adapt the algorithm model;

[0035] 3) The enhanced focus spot image is input into the pre-trained CNN regression network, and an error vector is output and displayed. Each element of the error vector corresponds to an adjustable dimension of the multi-axis adjustment device 5, and the element value of the adjustable dimension is the error of the corresponding adjustment dimension of the multi-axis adjustment device 5.

[0036] The work of the image processing unit 4 containing the pre-trained deep learning model before the CNN regression network pre-training includes the following steps:

[0037] 1) According to the parameters of the current experimental device and based on the light transmission model of the off-axis parabolic mirror 2, a data set of a focal spot intensity map corresponding to the adjustment error of the off-axis parabolic mirror 2 is obtained;

[0038] 2) by adjusting the input beam parameters of the light transmission model of the off-axis parabolic mirror, such as the beam waist radius, the clear aperture, etc., a data set of a focal spot intensity map corresponding to the adjustment error of the off-axis parabolic mirror 2 under different input light parameters is obtained;

[0039] 3) Before pre-training, it is also necessary to augment the data set according to the scene requirements, and expand the data set of the focal spot intensity map corresponding to multiple adjustment errors of the off-axis parabolic mirror 2, including: the difference between the pixel size of the actual image sensor and the simulation, and scaling the data set; when adjusting the off-axis parabolic mirror 2, the focus of the image sensor 3 is translated, and the data set image is translated.

[0040] The method for producing the CNN regression network data set can establish a model of the off-axis parabolic mirror 2 in the optical software, use the physical optics propagation module of the software to output the focal spot images of the off-axis parabolic mirror 2 with different misalignment errors, and save them as a data set; or the optical transmission physical model of the off-axis parabolic mirror 2 can be established in the CNN regression network, and the focal spot images with different misalignment errors can be generated during training.

[0041] The multi-axis adjustment device 5 used in this embodiment has two adjustment dimensions, namely, x-axis rotation and y-axis rotation. Therefore, the image processing unit 4 containing the pre-trained deep learning model outputs an error vector of two elements, and each element X1 and X2 of the error vector are the errors E1 and E2 of the x-axis rotation and y-axis rotation of the current off-axis parabolic mirror 2, respectively, thereby realizing a single measurement of the misalignment of the off-axis parabolic mirror;

[0042] 4. According to the measurement results, the x-axis rotation controller X1 and the y-axis rotation controller X2 of the multi-axis adjustment device (5) are adjusted to make the off-axis parabolic mirror (2) reach the ideal position.

[0043] Specifically, the processing steps of the image processing unit (4) containing the pre-trained deep learning model during measurement are as follows:

[0044] 1) Perform image segmentation preprocessing on the input focus spot image to segment the focus spot area from the large background area;

[0045] 2) Perform image enhancement processing on the segmented focus image, improve the resolution through Lanczos interpolation, and adapt the algorithm model;

[0046] 3) The image is input into the pre-trained CNN regression network, and the output is an error vector, the first element of the error vector is the x-axis rotation error of the off-axis parabolic mirror 2, and the second element is the y-axis rotation error of the off-axis parabolic mirror 2;

[0047] See also Figure 2 , Figure 2 This is a schematic diagram of an embodiment of a CNN regression network with a 4-layer convolutional unit and a 3-layer fully connected unit.

[0048] The CNN regression network of this embodiment includes a convolution unit and a fully connected unit. The CNN regression network of this embodiment has 4 layers of convolution and 3 layers of full connection. The output channels of the 1st to 4th convolution layers are 16, 32, 64, and 96, respectively. Finally, the output tensor of the 4th convolution layer is flattened and then input into the fully connected unit. The adjustable dimension of the multi-axis adjustment device 5 described in this embodiment is 2, so the output node of the fully connected unit network is 2.

[0049] In this embodiment, each convolution layer adopts the same network structure. There is a corresponding batch normalization function and activation function after each convolution layer. In this embodiment, the ReLU activation function is uniformly used.

[0050] The fully connected unit output layer of this embodiment adopts a linear activation function.

[0051] There are many methods for producing the above-mentioned CNN regression network data set. A model of the off-axis parabolic mirror 2 can be established in optical software, and the physical optics propagation module of the software can be used to output the focal spot images of the off-axis parabolic mirror 2 at different misalignment errors, and saved as a data set; the physical model of light transmission of the off-axis parabolic mirror 2 can also be established in the CNN regression network to generate focal spot images during training.

[0052] It is more convenient to use the physical optics propagation module of the optical software, so this embodiment uses the optical software to generate the focus image dataset. According to the parameters of the current experimental device, the dataset of the focus intensity map corresponding to the X-axis and Y-axis rotation of the off-axis parabolic mirror 2 is generated with a maximum misalignment of 3° and a step size of 0.01°;

[0053] In order to enhance the generalization of the optical transmission model and enable the optical transmission model to be applicable to a variety of input beam parameters, the input beam parameters of the optical transmission model of the off-axis parabolic mirror, such as the beam waist radius, the clear aperture, etc., are adjusted to obtain a data set of a focal spot intensity map corresponding to the adjustment error of the multi-dimensional off-axis parabolic mirror 2 under different input light parameters;

[0054] Before pre-training, the data set needs to be augmented according to the scene requirements to obtain a data set of focus intensity maps corresponding to multiple adjustment errors of the off-axis parabolic mirror 2, including: the difference between the pixel size of the actual image sensor and the simulation, and a 20% scaling transformation of the data set; when adjusting the off-axis parabolic mirror 2, the focus of the image sensor 3 is translated, and the data set image is translated in a 20% random direction;

[0055] Experiments show that this off-axis parabolic mirror single misalignment measurement method based on deep learning has the advantages of single, strong robustness, high precision, full automation and high speed. It can be used in astronomy, aviation, high energy density physics and inertial confinement fusion research or laser plasma interaction to improve experimental debugging efficiency and enhance beam quality.

[0056] The present invention mainly solves the problem of assembly misalignment of off-axis parabolic mirrors. According to the technology described in the present invention, a single, fast misalignment measurement of off-axis parabolic mirrors can be achieved, providing a new measurement scheme for optical paths containing off-axis parabolic mirrors. Although certain embodiments have been described, these embodiments are only presented as examples and are not intended to limit the scope of protection. Although the invention has been described in detail with reference to the aforementioned examples, it is still possible for those skilled in the art to modify the schemes recorded in the aforementioned examples, or to make equivalent substitutions for some of the technical features therein. All modifications, equivalent substitutions, etc. made within the spirit and principles of the invention should be included in the scope of protection of the invention.

Claims

1. A method for adjusting misalignment of an off-axis parabolic mirror, characterized in that: The method comprises the following steps: 1) An off-axis parabolic mirror (2) is fixed to a control end of a multi-axis adjustment device (5); an image sensor (3) is located at a focal position of the off-axis parabolic mirror (2); an output end of the image sensor (3) is connected to an input end of an image processing unit (4) containing a deep learning model; and an output end of the image processing unit (4) containing a deep learning model is connected to an input end of the multi-axis adjustment device (5); output light of an input light source (1) is focused through the off-axis parabolic mirror (2) and incident on the image sensor (3), thereby forming a focal spot image; 2) The image sensor (3) inputs the focal spot image into the image processing unit (4) containing the deep learning model; 3) After processing the focal spot image, the image processing unit (4) containing the deep learning model displays the error vector of N elements between the optical axis of the off-axis parabolic mirror (2) and the optical axis of the input light source (1) corresponding to the current focal spot image, and outputs it to the multi-axis adjustment device (5), wherein N is the adjustable dimension of the multi-axis adjustment device (5), and the element values ​​of the error vector of the corresponding dimension of the error vector of the N elements X1, X2, ..., XN are E1, E2, ..., EN; 4) The multi-axis adjustment device (5) adjusts the corresponding dimensions of the error vector of the N elements, based on the element values ​​E1, E2, ..., EN, so that the error between the optical axis of the off-axis parabolic mirror (2) and the optical axis of the input light source (1) is minimized, that is, the adjustment of the misalignment of the off-axis parabolic mirror (2) is completed; The method for processing the focal spot image by the image processing unit (4) containing the deep learning model comprises the following steps: 1) Perform image segmentation preprocessing on the input focus spot image to segment the focus spot area from the large background area; 2) Perform image enhancement processing on the segmented focal spot image to improve the resolution and adapt the algorithm model; 3) Inputting the enhanced focus spot image into a pre-trained CNN regression network, outputting and displaying an error vector, wherein each element of the error vector corresponds to an adjustable dimension of the multi-axis adjustment device (5), and the element value of the adjustable dimension is the error of the multi-axis adjustment device (5) corresponding to the adjustment dimension.

2. The method for adjusting the misalignment of an off-axis parabolic mirror according to claim 1, characterized in that The work of the image processing unit (4) containing the deep learning model before the CNN regression network pre-training includes the following steps: 1) according to the parameters of the current experimental device and based on the light transmission model of the off-axis parabolic mirror (2), a data set of a focal spot intensity map corresponding to the adjustment error of the off-axis parabolic mirror (2) is obtained; 2) by adjusting the input beam parameters of the light transmission model of the off-axis parabolic mirror, i.e., the beam waist radius and the clear aperture, a data set of a focal spot intensity map corresponding to the adjustment error of the off-axis parabolic mirror (2) under different input light parameters is obtained; 3) Before pre-training, the data set needs to be augmented according to scene requirements to obtain a data set of focus intensity maps corresponding to multiple adjustment errors of the off-axis parabolic mirror (2), including: the difference between the pixel size of the actual image sensor and the simulation, scaling the data set; when adjusting the off-axis parabolic mirror (2), the focus of the image sensor (3) is shifted, and the data set image is shifted.

3. The method for adjusting the misalignment of an off-axis parabolic mirror according to claim 2, characterized in that The method for preparing the CNN regression network data set is to establish a model of the off-axis parabolic mirror (2) in optical software, use the physical optics propagation module of the software to output the focal spot images of the off-axis parabolic mirror (2) with different misalignment errors, and save them as a data set; or establish the optical transmission physical model of the off-axis parabolic mirror (2) in the CNN regression network, and generate the focal spot images with different misalignment errors during training.

Images