Self-adaptive optical system non-common-path aberration correction method based on neural network

Abstract

The invention discloses an adaptive optical system non-common-path aberration correction method based on a neural network. The adaptive optical system non-common-path aberration correction method comprises the following steps: constructing a lightweight neural network CP-Net; simulating and generating a perfect point spread function image; generating a deformable mirror control voltage matrix based on the image; taking the voltage matrix as the input of CP-Net for training, and ending the training when the loss function value is smaller than a preset minimum value; the trained CP-Net is connected to an actual control loop, real-time images acquired by a scientific camera are received, and perfect reference voltage is generated and directly used for controlling and changing the mirror surface type of the deformable mirror. According to the method, the deformable mirror control command can be directly generated, aberration does not need to be measured, real-time automatic adjustment can be achieved, compared with a traditional phase diversity method and a stochastic gradient descent algorithm, the operation difficulty can be greatly lowered, the correction time can be greatly shortened, and an operator does not need to manually adjust related parameters in the correction process.

Description

Technical Field

[0001] The invention relates to an adaptive control method of a deformable mirror, and in particular to a non-common optical path aberration correction method of an adaptive optical system based on a neural network. Background Art

[0002] As the exploration of the universe deepens, direct imaging of targets such as exoplanets is an important goal in the field of astronomy. However, it is very difficult to observe high-value targets such as exoplanets at night, and the light signals emitted by most observation targets are relatively weak. In addition, when the light signal passes through the Earth's atmosphere, it is affected by adverse factors such as air flow disturbances and temperature changes in the Earth's atmosphere, which will cause distortion, resulting in blurred observed targets and the inability to analyze their original physical properties. Therefore, adaptive optical systems came into being to correct the distortion of light signals and allow telescopes to obtain imaging with diffraction-limited capabilities.

[0003] The adaptive optical system is mainly composed of a deformable mirror, a wavefront sensor, a controller, and an optical imaging component. The wavefront sensor detects the wavefront slope to calculate the distortion information of the wavefront, and then sends the distortion information to the controller. The controller calculates the control signal according to the distortion information and sends it to the deformable mirror actuator to compensate for the wavefront distortion caused by factors such as atmospheric turbulence by changing the mirror surface shape of the deformable mirror. In order to improve the correction capability of the entire adaptive system, the detection accuracy of the wavefront sensor is crucial. However, there is a difference between the optical path where the wavefront sensor is located and the scientific imaging optical path of the final imaging camera, which results in some aberrations that cannot be detected by the wavefront sensor, thus forming non-common path aberrations (NCPE).

[0004] There are currently two conventional methods for correcting non-common optical path aberrations: the first is the phase diversity algorithm, which is cumbersome and requires the collection of a large number of focal and defocused images on site, and has a low degree of intelligence; the second is the stochastic parallel gradient descent algorithm (SPGD), the performance of which is extremely sensitive to the setting of the evaluation function. The correction results after using different evaluation functions vary greatly, and the correction process is time-consuming, requiring thousands or even tens of thousands of iterations. In addition, both methods require operators to manually adjust relevant parameters during the correction process, and are difficult to adjust automatically in real time. In addition, the existing wavefront neural network method is mainly used to generate the Zenith polynomial coefficients of the wavefront signal and then calculate the control command rate. It still has not gotten rid of the two-step strategy of first detection and then correction in the traditional method. At the same time, it requires expensive customized hardware such as DSP, making it difficult for the algorithm to be directly connected to the existing hardware system. Therefore, the existing methods for correcting non-common optical path aberrations are difficult to meet the intelligent and high-performance requirements of future large telescope adaptive optical systems. Summary of the invention

[0005] In view of the above problems existing in the prior art, the present invention proposes a method based on a neural network that can directly control a deformable mirror to autonomously correct the non-common optical path aberration in an adaptive optical system. The perfect point spread function (PSF) image generated by simulation software is used as a benchmark for training, and the neural network algorithm directly generates the deformable mirror control voltage to compensate for the non-common optical path aberration in the optical system, thereby improving the correction capability of the adaptive optical system.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] The method for correcting non-common optical path aberration of an adaptive optical system based on a neural network comprises the following steps:

[0008] Step 1: Build a lightweight neural network CP-Net;

[0009] Step 2: Simulate and generate a perfect point spread function image;

[0010] Step 3: Generate a deformable mirror control voltage matrix as a reference voltage based on the perfect point spread function image used as a training set;

[0011] Step 4: The deformable mirror control voltage matrix is ​​used as the input of the neural network CP-Net for training. When the loss function value is less than the preset minimum value, the output voltage is a perfect reference voltage, and the training is terminated at this time.

[0012] Step 5: Connect the trained neural network CP-Net to the actual control loop, receive the real-time image acquired by the scientific camera, generate a perfect reference voltage, and directly use it to control the change of the deformable mirror's mirror surface shape.

[0013] Furthermore, in step 1, the neural network CP-Net includes three input channels, namely, upper, middle and lower, and one output channel; wherein:

[0014] The upper and lower input channels are the same, and both include several repeated comprehensive layers. Each comprehensive layer includes a convolutional layer, a nonlinear activation layer, a normalization layer, and a maximum pooling layer from the input side to the back.

[0015] The intermediate input channel includes a fusion layer, a convolution layer, a nonlinear activation layer, a normalization layer, and a maximum pooling layer from the input side to the back;

[0016] The output channel includes a fusion layer, a normalization layer, two consecutive convolutional layers, a nonlinear activation layer, two consecutive fully connected layers, and a regression layer from the input side to the back.

[0017] Furthermore, in the upper and lower input channels, the convolution kernel of the convolution layer is 4*4, the step size is 1, the nonlinear activation layer uses the ReLu function, the data size processed by the normalization layer is 8, the number of channels is 2, the mean is planned to 0.5, the variance is planned to 1, and the maximum pooling layer uses 2*2 maximum pooling processing; in the middle input channel, the fusion layer unifies the data stream into 4*4*2, the convolution kernel of the convolution layer is 1*1, the step size is 1, the nonlinear activation layer uses the ReLu function, and the data size processed by the normalization layer is The size is 4, the number of channels is 2, the mean is planned to 0.5, the variance is planned to 1, and the maximum pooling layer uses 2*2 maximum pooling processing; in the output channel, the fusion layer unifies the data stream to 4*4*2, the data size processed by the standardization layer is 4, the number of channels is 2, the mean is planned to 0.5, the variance is planned to 1, the convolution kernel of the convolution layer is 2*2, the step size is 1, the nonlinear activation layer uses the ReLu function, the fully connected layer uses 1*1 fully connected, and the regression layer outputs a control signal of 1*1*88.

[0018] Furthermore, in step 2, the formula for generating the perfect point spread function image is:

[0019] ;

[0020] Where A is the light intensity of the input light source, Besselj( ) represents the Bessel function, (x, y) is the horizontal and vertical coordinates of each point on the PSF image, k is the wave number of the input light source, and Na represents the numerical aperture.

[0021] Furthermore, in step 3, the perfect point spread function image used as a training set is partially covered to simulate the signal loss under a dim light source.

[0022] Furthermore, in step 3, the control voltage of the deformable mirror is generated using the stochastic gradient descent method to form a deformable mirror control voltage matrix, which serves as a label for training the CP-Net network to generate voltage.

[0023] The formula for generating the control voltage using the stochastic gradient descent method is as follows:

[0024] ;

[0025] Where k is the number of iterations, u is the control voltage of the deformable mirror actuator, is the gain coefficient, which is positive when the evaluation function is optimized toward the minimum value, and negative otherwise. is the change of the evaluation function, is the applied random disturbance voltage signal and obeys the Bernoulli distribution.

[0026] Furthermore, the stochastic gradient descent method is evaluated using the following evaluation function:

[0027] ;

[0028] Among them, (x, y) is the horizontal and vertical coordinates of each point on the generated perfect point spread function image, (x i , y i ) is the horizontal and vertical coordinates of each point on the real-shot image of the scientific camera in the adaptive optical system; when the value of the evaluation function reaches the preset threshold, the control voltage v output by the CP-Net network ir That is the perfect reference voltage, which is used as the training label of the CP-Net network.

[0029] Furthermore, the loss function used in step 4 is:

[0030] ;

[0031] Among them, v i Represents the actual control voltage of the deformable mirror.

[0032] Furthermore, the calculation stage of the non-common optical path aberration is skipped, and the control command of the deformable mirror is directly generated to automatically correct the non-common optical path aberration in the adaptive optical system, and automatically make adjustments during the correction process.

[0033] Compared with the prior art, the present invention has the following beneficial effects:

[0034] (1) The CP-Net network structure adopted by the neural network-based adaptive optical system non-common optical path aberration correction method of the present invention is simple and can be directly run on a commercial computer without the need for professional equipment such as DSP.

[0035] (2) The method of the present invention can directly generate deformable mirror control commands without measuring aberrations, and can achieve real-time automatic adjustment. Compared with the traditional phase diversity method and stochastic gradient descent algorithm, the method can greatly reduce the operation difficulty and correction time. During the correction process, there is no need for the operator to manually adjust the relevant parameters.

[0036] (3) It can be easily integrated into existing adaptive optics systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 This is a schematic diagram of the CP-Net structure.

[0038] Figure 2 This is an example of a perfect PSF image in the CP-Net dataset.

[0039] Figure 3 This is the overall training and application flow chart of CP-Net.

[0040] Figure 4 The figures are effect diagrams of the adaptive optical system before and after the method of the present invention is adopted; wherein, (a) is the effect diagram of the adaptive optical system before the method of the present invention is adopted; and (b) is the effect diagram of the adaptive optical system after the method of the present invention is adopted. DETAILED DESCRIPTION

[0041] The present invention will be further described in detail below in conjunction with the accompanying drawings.

[0042] The present invention provides a method for correcting non-common optical path aberrations of an adaptive optical system based on a neural network, which can make the adaptive optical system unaffected by non-common optical path aberrations when performing scientific observations, and automatically adjust the surface shape of the deformable mirror in real time. The method comprises the following steps:

[0043] Step 1: Build a lightweight neural network CP-Net;

[0044] Step 2: Simulate and generate a perfect point spread function image;

[0045] Step 3: Generate a deformable mirror control voltage matrix as a reference voltage based on the perfect point spread function image used as a training set;

[0046] Step 4: The deformable mirror control voltage matrix is ​​used as the input of the neural network CP-Net for training. When the loss function value is less than the preset minimum value, the output voltage is a perfect reference voltage, and the training is terminated at this time.

[0047] Step 5: Connect the trained neural network CP-Net to the actual control loop, receive the real-time image acquired by the scientific camera, generate a perfect reference voltage, and directly use it to control the change of the deformable mirror's mirror surface shape.

[0048] Figure 1 : is a schematic diagram of the specific structure of the neural network CP-Net of this embodiment. The neural network CP-Net includes three input channels, upper, middle and lower, and one output channel; wherein:

[0049] The upper and lower input channels are the same, and both include three repeated comprehensive layers. Each comprehensive layer performs the following operations from the input side to the back: first perform a convolution operation with a convolution kernel of 4*4 and a step size of 1, then use the ReLu function for nonlinear activation processing, and then perform a standardization operation with a data size of 8 and a number of channels of 2. The mean is planned to 0.5 and the variance is planned to 1, and then a 2*2 maximum pooling process is performed. Repeat the above operation three times and enter the output channel;

[0050] The intermediate input channel performs the following operations from the input side to the back: first perform a fusion operation to unify the data stream into 4*4*2, then the convolution kernel entering the convolution layer is 1*1, the step size is 1, the nonlinear activation layer uses the ReLu function, the data size processed by the standardization layer is 4, the number of channels is 2, the mean is planned to 0.5, the variance is planned to 1, the maximum pooling layer uses 2*2 maximum pooling processing, and then enters the output channel;

[0051] The output channel includes the following operations from the input side to the back: first, a fusion operation is performed to unify the data stream into 4*4*2, and then a standardization process is performed with a data size of 4 and a number of channels of 2. The mean is planned to 0.5 and the variance is planned to 1. Then, two convolution operations are performed with a convolution kernel of 2*2 and a step size of 1. After a nonlinear activation of the ReLu function, two 1*1 fully connected operations are performed, and finally the regression layer 1*1*88 is entered to output the control signal.

[0052] During the training phase, CP-Net used 5,000 simulated perfect point spread function (PSF) images as the data set, 3,000 as the training set, with the wavefront images in the training set partially covered with white squares to simulate the signal loss under weak light sources, 1,000 as the test set, and 1,000 as the validation set. Figure 2 shown.

[0053] In the process of simulating the PSF image, various parameters of the adaptive optical system that need to be corrected are taken into account, including aperture, wavelength, light intensity, etc. The calculation formula is implemented step by step according to formulas (1), (2), and (3).

[0054] (1)

[0055] Where k is the wave number of the input light source and wl is the wavelength of the input light source.

[0056] (2)

[0057] Among them, Na represents the numerical aperture, Dia represents the aperture of the adaptive optical system, and f represents the focal length of the optical system.

[0058] The final simulated PSF image calculation formula is as follows:

[0059] (3)

[0060] Where A is the intensity of the input light source, Besselj( ) represents the Bessel function, and (x, y) is the horizontal and vertical coordinates of each point on the PSF image.

[0061] Then, the perfect PSF image generated above is used as a benchmark and optimization target, and the control voltage of the deformable mirror is generated using the stochastic gradient descent method as a reference voltage, which is used as a label for training the CP-Net network to generate voltage. The formula for generating the control voltage using the stochastic gradient descent method is as follows:

[0062] ;

[0063] Where k is the number of iterations, u is the control voltage of the deformable mirror actuator, is the gain coefficient, which is positive when the evaluation function is optimized toward the minimum value, and negative otherwise. is the change of the evaluation function, is the applied random disturbance voltage signal and obeys the Bernoulli distribution.

[0064] In order to generate a perfect reference voltage, the metrics function of the stochastic gradient descent method is set to:

[0065]

[0066] Among them, (x, y) is the horizontal and vertical coordinates of each point on the generated perfect PSF image, (x i , y i ) is the horizontal and vertical coordinates of each point on the real-time image taken by the camera in the system. When the value of the evaluation function reaches the order of magnitude of 0.0001, it can be considered that the PSF of the actual system is consistent with the perfect PSF. At this time, the control voltage v ir It is used as a perfect reference voltage and as a training label for CP-Net. The following formula is used as the loss function during CP-Net training:

[0067]

[0068] Among them, v iRepresents the actual control voltage of the deformable mirror. When the loss function value reaches the interval [0.9, 1.1], it is considered that the optimal network training result is achieved.

[0069] After training, CP-Net can be connected to the control loop of the adaptive optical system to automatically correct the non-common optical path aberrations in the system. Compared with existing neural network algorithms, CP-Net only takes a long time to train. Once it arrives at the observation site and starts running, it can directly eliminate the non-common optical path aberrations in the system. There is no need to measure aberrations or adjust parameters on site, and it is more suitable for ground-based large-aperture telescopes that are difficult to adjust.

[0070] The training and application process of the neural network CP-Net is as follows Figure 3 As shown in the figure, using the perfect PSF image produced by simulation, the reference voltage control matrix is ​​first calculated using the stochastic gradient descent algorithm, and provided as training label data to the CP-Net network proposed in the present invention. After training, the CP-Net is deployed in a commercial computer (controller), and the deformable mirror control command is directly generated according to the real-time image, and then sent to the deformable mirror (DM). The operation control results of the system are shown in Figure 4 It can be clearly seen from the figure that after the system is in operation, it can effectively concentrate energy and eliminate the influence of non-common optical path aberrations in the system on the observation results.

[0071] In summary, in order to ensure that the adaptive optical system is not affected by non-common optical path aberrations during scientific observation, the present invention proposes a method for correcting non-common optical path aberrations of an adaptive optical system based on a neural network. Based on the lightweight neural network CP-Net constructed by the present invention, the non-common optical path aberrations in the adaptive optical system are automatically corrected before observation, the calculation stage of the non-common optical path aberrations is skipped, and the deformable mirror control command is directly generated to correct the non-common optical path aberrations, and adjustments are automatically made during the correction process. The defect of the traditional phase diversity method that detection and correction are performed separately is solved, the degree of intelligence is greatly improved, the manual parameter adjustment during the correction process is eliminated, and the correction time is saved. At the same time, the CP-Net network structure is simple, and it can run smoothly even on a commercial PC, and can be incorporated into the existing adaptive optical system at a low cost.

[0072] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modification, equivalent replacement and improvement made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for correcting non-common optical path aberrations in an adaptive optical system based on a neural network, characterized in that: The steps include: Step 1: Build a lightweight neural network CP-Net; Step 2: Simulate and generate a perfect point spread function image; Step 3: Generate a deformable mirror control voltage matrix as a reference voltage based on the perfect point spread function image used as a training set; Step 4: The deformable mirror control voltage matrix is ​​used as the input of the neural network CP-Net for training. When the loss function value is less than the preset minimum value, the output voltage is a perfect reference voltage, and the training is terminated at this time. Step 5: Connect the trained neural network CP-Net to the actual control loop, receive the real-time image acquired by the scientific camera, generate a perfect reference voltage, and directly use it to control the change of the deformable mirror's mirror surface shape.

2. The method for correcting non-common optical path aberrations of an adaptive optical system based on a neural network according to claim 1, characterized in that: In step 1, the neural network CP-Net includes three input channels, upper, middle and lower, and one output channel; wherein: The upper and lower input channels are the same, and both include several repeated comprehensive layers. Each comprehensive layer includes a convolutional layer, a nonlinear activation layer, a normalization layer, and a maximum pooling layer from the input side to the back. The intermediate input channel includes a fusion layer, a convolution layer, a nonlinear activation layer, a normalization layer, and a maximum pooling layer from the input side to the back; The output channel includes a fusion layer, a normalization layer, two consecutive convolutional layers, a nonlinear activation layer, two consecutive fully connected layers, and a regression layer from the input side to the back.

3. The method for correcting non-common optical path aberrations of an adaptive optical system based on a neural network according to claim 2, characterized in that: In the upper and lower input channels, the convolution kernel of the convolution layer is 4*4, the step size is 1, the nonlinear activation layer uses the ReLu function, the data size processed by the standardization layer is 8, the number of channels is 2, the mean is planned to 0.5, the variance is planned to 1, and the maximum pooling layer uses 2*2 maximum pooling processing; In the middle input channel, the fusion layer unifies the data stream into 4*4*2, the convolution kernel of the convolution layer is 1*1, the step size is 1, the nonlinear activation layer uses the ReLu function, the data size processed by the standardization layer is 4, the number of channels is 2, the mean is planned to 0.5, the variance is planned to 1, and the maximum pooling layer uses 2*2 maximum pooling processing; In the output channel, the fusion layer unifies the data stream into 4*4*2, the data size processed by the normalization layer is 4, the number of channels is 2, the mean is planned to 0.5, the variance is planned to 1, the convolution kernel of the convolution layer is 2*2, the step size is 1, the nonlinear activation layer uses the ReLu function, the fully connected layer uses 1*1 fully connected, and the regression layer outputs a control signal of 1*1*88.

4. The method for correcting non-common optical path aberrations of an adaptive optical system based on a neural network according to claim 1, characterized in that: In step 2, the formula for generating the perfect point spread function image is: ； Where A is the light intensity of the input light source, Besselj( ) represents the Bessel function, (x, y) is the horizontal and vertical coordinates of each point on the PSF image, k is the wave number of the input light source, and Na represents the numerical aperture.

5. The method for correcting non-common optical path aberrations of an adaptive optical system based on a neural network according to claim 1, characterized in that: In step 3, the perfect point spread function image used as a training set is partially covered to simulate the signal loss under a dim light source.

6. The method for correcting non-common optical path aberrations of an adaptive optical system based on a neural network according to claim 1, characterized in that: In step 3, the control voltage of the deformable mirror is generated using the stochastic gradient descent method to form a deformable mirror control voltage matrix, which serves as a label for training the CP-Net network to generate voltage.

7. The method for correcting non-common optical path aberrations of an adaptive optical system based on a neural network according to claim 6, characterized in that: The formula for generating the control voltage using the stochastic gradient descent method is as follows: ； Where k is the number of iterations, u is the control voltage of the deformable mirror actuator, is the gain coefficient, which is positive when the evaluation function is optimized toward the minimum value, and negative otherwise. is the change of the evaluation function, is the applied random disturbance voltage signal and obeys the Bernoulli distribution.

8. The method for correcting non-common optical path aberrations of an adaptive optical system based on a neural network according to claim 6, characterized in that: The stochastic gradient descent method is evaluated using the following evaluation function: ； Among them, (x, y) is the horizontal and vertical coordinates of each point on the generated perfect point spread function image, (x i , y i ) is the horizontal and vertical coordinates of each point on the real-shot image of the scientific camera in the adaptive optical system; when the value of the evaluation function reaches the preset threshold, the control voltage v output by the CP-Net network ir That is the perfect reference voltage, which is used as the training label of the CP-Net network.

9. The method for correcting non-common optical path aberrations of an adaptive optical system based on a neural network according to claim 8, characterized in that: The loss function used in step 4 is: ； Among them, v i Represents the actual control voltage of the deformable mirror.

10. The method for correcting non-common optical path aberrations of an adaptive optical system based on a neural network according to claim 1, characterized in that: The calculation stage of non-common optical path aberrations is skipped, and control commands for the deformable mirror are directly generated to automatically correct non-common optical path aberrations in the adaptive optical system and make automatic adjustments during the correction process.

Images