Spliced mirror co-phase error detection method based on wavelet transform multi-channel input

Abstract

The invention discloses a spliced mirror co-phase error detection method based on wavelet transform multi-channel input. The method comprises the following steps: designing an optical system based on far-field imaging according to an optical system principle; 10000 groups of co-phase errors within one wavelength are introduced, and 20dB Gaussian white noise is introduced; the method comprises the following steps: acquiring 10000 far-field images, recording corresponding co-phase error values, respectively constructing an image data set and a label data set, and selecting a training set from the image data set; configuring a deep learning environment, and building a neural network model; preprocessing image data in the training set to serve as input of a neural network model, and obtaining a predicted co-phase error value through the neural network model; and calculating a loss value between the predicted co-phase error value and a given label, and carrying out optimization and parameter adjustment on the neural network. The method is good in noise immunity and higher in precision, and has practical application significance for co-phase error detection based on the far-field image.

Description

Technical Field

[0001] The present invention relates to the technical field of photoelectric measurement, image processing and deep learning, and in particular to a common phase error detection method of a spliced ​​mirror based on wavelet transform multi-channel input. Background Art

[0002] According to the Rayleigh criterion, for a given wavelength λ, the resolution of the optical system is inversely proportional to the pupil diameter D, that is, the larger the pupil diameter, the higher the system resolution; at the same time, the light-gathering ability of the system is proportional to the square of the pupil diameter. This means that a larger mirror can collect more light to achieve clearer imaging. In order to meet the needs of deeper space and higher resolution observations, scientists have begun to develop giant telescopes with large-aperture primary mirrors. However, the solution of using a single large-aperture reflector faces many technical limitations, including the difficulty in maintaining the accuracy of mirror manufacturing and testing, and the high difficulty of processing and transportation. In addition, the size of the launch vehicle fairing also limits the practical application of large-aperture single mirrors. Spliced ​​mirrors or synthetic aperture technology is an important technical means to develop the next generation of very large telescopes. The high-resolution imaging capability of a single primary mirror with the same aperture is achieved by splicing sub-mirrors in common phase or combining subsystems in common phase. Therefore, the detection of piston error is particularly important in spliced / synthetic aperture telescope systems.

[0003] In recent years, with the development of artificial intelligence and deep learning, common phase error detection based on convolutional neural networks (CNN) has made significant progress. As early as 2018, Guerra-Ramos et al. in Spain used CNN to realize the common phase error piston term detection of 36 spliced ​​mirrors; later, Li Dequan et al. from the Changchun Institute of Optics, Fine Mechanics and Physics of the Chinese Academy of Sciences studied the common phase detection performance of CNN for opposite targets; Ma Xiafei et al. from the Institute of Optoelectronics Technology of the Chinese Academy of Sciences constructed a deep CNN and confirmed the application of this method in the common phase error detection of spliced ​​mirrors through simulation and experiments.

[0004] While the above-mentioned common phase error detection method based on convolutional neural network improves the detection accuracy, it still faces some problems. In practical applications, the far-field image of the telescope system often contains noise, which will cause the image features to be blurred or distorted, thereby affecting the prediction accuracy of the neural network. This greatly reduces the performance of the common phase error detection method based on CNN under noisy conditions. Summary of the invention

[0005] The technical problem to be solved by the present invention is: in view of the poor performance of traditional neural networks in the detection of common phase errors of spliced ​​mirrors in noisy environments, a common phase error detection method for spliced ​​mirrors based on wavelet transform multi-channel input is proposed. The method inputs the multi-channel image after wavelet decomposition into a convolutional neural network (CNN), extracts multi-scale features to suppress the influence of noise, and realizes high-precision prediction of common phase errors, thereby effectively improving the detection accuracy under noisy conditions and meeting the accuracy requirements of the spliced ​​mirror system.

[0006] The technical solution adopted by the present invention to solve the above technical problems is: a method for detecting common phase errors of spliced ​​mirrors based on wavelet transform multi-channel input, comprising the following steps:

[0007] Step 1: According to the principle of optical system, an optical system based on far-field imaging is designed. The optical system includes a main mirror formed by splicing two sub-mirrors, a mask, a focusing lens and a detector. The incident parallel light beam is reflected by the sub-mirror and passes through the mask. After being converged by the focusing lens, it is imaged on the image plane of the detector. There is a common phase error between the sub-mirrors.

[0008] Step 2: Taking one circular hole as the reference, randomly introduce 10,000 groups of common phase errors within one wavelength to the other circular hole, and introduce 20 dB Gaussian white noise to simulate the actual camera readout noise;

[0009] Step 3: Collect 10,000 far-field images and record the corresponding common phase error values, construct image datasets and label datasets respectively, and select a training set from the image dataset;

[0010] Step 4: Configure the deep learning environment and build a neural network model;

[0011] Step 5: Preprocess the image data in the training set, decompose the single-channel image data into 4-channel components through a Haar wavelet, use it as the input of the neural network model, and obtain the predicted common phase error value through the neural network model;

[0012] Step 6: Calculate the loss value between the predicted common error value and the given label, and optimize the neural network and adjust the parameters.

[0013] Compared with the prior art, the present invention has the following beneficial effects:

[0014] (1) The present invention decomposes the far-field image into multi-scale features through wavelet transform and inputs them into CNN in a multi-channel manner, so that the model can more accurately capture the details of the common phase error and improve the prediction accuracy.

[0015] (2) Wavelet multi-channel input has the advantage of noise resistance, which enhances the robustness of the model in noisy environments and enables the system to maintain high-precision detection under noise interference.

[0016] (3) The present invention adopts the ResNet18 structure and improves the feature extraction efficiency through the residual module. The network is lightweight and optimized quickly, which is suitable for the practical application of the splicing mirror system. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] FIG1(a) is a Michelson type spliced ​​telescope system, which is composed of a Michelson type primary mirror 1-1, a Michelson type secondary mirror 1-2 and a Michelson type image plane 1-3; FIG1(b) is a Perestroika type synthetic aperture telescope system, which is composed of an afocal subsystem 2-1, a phase delay device 2-2, a beam synthesizer 2-3 and a Perestroika type image plane 2-4.

[0018] Figure 2 This is a schematic diagram of the optical imaging system corresponding to the present invention. Taking two hexagonal sub-mirrors as an example, the system consists of a sub-mirror 3-1, a mask 3-2, a focusing lens 3-3 and a detector 3-4.

[0019] Figure 3 The figure shows the principle of multi-channel decomposition of wavelet transform in the present invention. The original noisy image (224×224 pixels) is decomposed into: (a) LL (low-frequency approximation component, 112×112 pixels), (b) LH (horizontal high-frequency component, 112×112 pixels), (c) HL (vertical high-frequency component, 112×112 pixels), and (d) HH (diagonal high-frequency component, 112×112 pixels) after one Haar wavelet decomposition.

[0020] Figure 4 This is a schematic diagram of the common phase detection method of splicing mirrors based on wavelet transform multi-channel input proposed by the present invention.

[0021] Figure 5 The specific model used in the convolutional neural network of the present invention is ResNet18. The network includes 17 convolutional layers, 2 pooling layers and 1 fully connected layer, and the input is directly passed to the output through the residual connection in ResNet18.

[0022] Figure 6 This is the distribution diagram of the prediction residual error of the method proposed in the present invention compared with traditional transfer learning. DETAILED DESCRIPTION

[0023] In order to make the process, technical solutions and advantages of the present invention clearer, the invention is further described below with reference to the accompanying drawings, taking a spliced ​​double hexagonal sub-mirror system as an embodiment of the present invention.

[0024] Figure 1(a) shows a Michelson type spliced ​​telescope system. The incident parallel light beam is reflected by the spliced ​​Michelson type primary mirror 1-1 and then reflected by the Michelson type secondary mirror 1-2, and imaged on the Michelson type image plane 1-3. Figure 1(b) shows a Pesso type synthetic aperture telescope system. The parallel light beam is reflected by the afocal subsystem 2-1 and then added with different phases by the phase retarder 2-2. The light beam synthesized by the beam synthesizer 2-3 is imaged on the Pesso type image plane 2-4.

[0025] Figure 2 1 is a schematic diagram of an optical imaging system corresponding to the present invention. The main mirror can be a Michelson type spliced ​​telescope system as shown in FIG1(a) or a Pesso type synthetic aperture telescope system as shown in FIG1(b), which is spliced ​​by two sub-mirrors 3-1. There is a common phase error between the sub-mirrors 3-1. The incident parallel light beam is reflected by the spliced ​​sub-mirrors 3-1, passes through the mask 3-2, and is converged by the focusing lens 3-3 to form an image on the imaging surface of the detector 3-4. Figure 3 The principle diagram of the multi-channel decomposition of wavelet transform in the present invention. After the original noisy image (224×224 pixels) is decomposed by Haar wavelet once, four sub-images are generated: (a) LL component (low-frequency approximation component, 112×112 pixels), which contains the main low-frequency information of the image; (b) LH component (horizontal high-frequency component, 112×112 pixels), which reflects the high-frequency details in the horizontal direction; (c) HL component (vertical high-frequency component, 112×112 pixels), which shows the high-frequency details in the vertical direction; and (d) HH component (diagonal high-frequency component, 112×112 pixels), which contains the high-frequency information in the diagonal direction.

[0026] Figure 4 This is the process of the splicing mirror common phase error detection method based on wavelet transform multi-channel input proposed by the present invention. The 4-channel image obtained after the original single-channel image is subjected to wavelet transform is used as the input of the convolutional neural network (CNN). The CNN model (ResNet18 structure is used in this application) is used to extract features and perform regression mapping on the input image, and a prediction value is output. .

[0027] The common phase error detection method of the splicing mirror based on wavelet transform multi-channel input includes the following steps:

[0028] Step 1: According to the principle of optical system, an optical system based on far-field imaging is designed. The optical system includes a main mirror formed by splicing two sub-mirrors 3-1, a mask 3-2, a focusing lens 3-3 and a detector 3-4. The incident parallel light beam is reflected by the spliced ​​sub-mirror 3-1, passes through the mask 3-2, and is converged by the focusing lens 3-3 to be imaged on the image plane of the detector 3-4. There is a common phase error between the sub-mirrors 3-1. Different phase differences are applied to the spliced ​​sub-mirrors 3-1 to simulate the actual common phase error. The design process includes the design of: the wavelength of the incident light, the wave number, the focal length of the focusing lens 3-3, the circular hole, etc. In one example, the wavelength of the incident light , wave number , lens focal length , round hole The circular hole is the white circular area in the mask 3-2, and the design of the circular hole includes the design of the circular hole diameter and the distance between the two holes.

[0029] Step 2: Using one circular hole as a reference, randomly introduce 10,000 groups of one wavelength into the other circular hole. The common phase error is within , and 20dB Gaussian white noise is introduced to simulate the actual camera readout noise.

[0030] Step 3: Collect 10,000 far-field images and record the corresponding common phase error values, construct image data sets and label data sets respectively, and select training sets from the image data sets. The collected diffraction images contain the common phase error (piston error) between the spliced ​​sub-mirrors. The 10,000 images are obtained by applying different phase differences to the spliced ​​sub-mirror 3-1.

[0031] Figure 2 In the spliced ​​mirror optical system shown, the sub-mirror 3-1 has a common phase error. In the simulation, the common phase error can be simulated by applying a phase difference (label) in the circular hole of the mask 3-2. After being converged by the focusing lens 3-3, the image is imaged on the detector 3-4. When the applied phase difference (label) changes, the final image changes accordingly, that is, the label and the image are one-to-one corresponding. Constructing an image data set and a label data set may include: constructing a folder to store all named images (image data set); constructing another folder to store all label data containing image indexes (label data set). The two are one-to-one corresponding, and one image corresponds to one label.

[0032] In one embodiment, 80% of the image data set is selected as a training set for the neural network model to learn the nonlinear mapping relationship between the image after Haar wavelet decomposition and the common phase error value, and 20% of the image data set is selected as a verification set to measure the accuracy and real-time performance of the neural network model.

[0033] Step 4: Configure the deep learning environment and build a neural network model;

[0034] Step 5: Preprocess the image data in the training set, decompose the single-channel image data into 4-channel components through a Haar wavelet, use it as the input of the neural network model, and obtain the predicted common phase error value through the neural network model.

[0035] Step 6: Calculate the loss value between the predicted common error value and the given label, optimize the neural network and adjust the parameters. The calculation of the loss value includes: calculating the difference between the predicted common error value and the given label through the mean square error loss function (MSE), thereby obtaining the loss value. The given label is in the label data set. The given label is determined by path indexing. For example, if an image in the image data set is named "0001.png", the network searches for the corresponding label value in the label according to this path. For example, the path of the corresponding label at this time is: "0001.png_0.025", and the algorithm extracts the index value of the label under the image path, that is, "0.025" (given label). In this way, the corresponding extraction relationship between the image and the given label is completed.

[0036] After the training network converges, the network only needs to be fed with a single frame of far-field diffraction image, and the network can output the predicted common phase error value. Compared with the traditional common phase error detection method based on transfer learning, the method proposed in the present invention has stronger robustness to noisy images and greatly improves the prediction accuracy. Simulation shows that the time required for the present invention to complete a common phase error prediction is about 10 milliseconds.

[0037] Among them, the mathematical expression of the image plane intensity in step 3 is:

[0038] ,

[0039] in, represents the ideal distribution of point sources on the image plane, represents the convolution operation, is the camera noise, are the horizontal and vertical coordinates of the image plane, respectively. The intensity point spread function PSF can be expressed as:

[0040] ,

[0041] in, is the wavelength of incident light, is the focal length of the lens, represents the generalized pupil function of the optical system, is the Fourier transform.

[0042] Figure 5The specific model used in the convolutional neural network in this invention is ResNet18. ResNet18 contains 17 convolutional layers, 2 pooling layers (1 maximum pooling layer and 1 global average pooling layer) and 1 fully connected layer. In the network, the input is directly transferred to the output through the residual connection to alleviate the gradient vanishing problem. Figure 5 The lower right corner of the figure shows the structure of the residual module, each of which contains two layers of convolution. The input is directly passed to the output through the residual connection (i.e., skip connection), which alleviates the gradient vanishing problem in deep network training and improves the stability and training efficiency of the model.

[0043] In summary, a common phase detection method for spliced ​​mirrors based on wavelet transform multi-channel input is proposed. First, the far-field diffraction noisy image of the spliced ​​mirror containing piston error is decomposed by wavelet to obtain multi-scale feature components, which are provided as multi-channel input to the convolutional neural network (CNN). The network adopts the ResNet18 model structure and contains multiple residual modules to improve the robustness and noise resistance of feature extraction. The network output is a single regression value, which is used to predict the common phase error (piston error) of the spliced ​​mirror. The mean square error loss function (MSE) is used in the training process to minimize the prediction error by optimizing the network parameters.

[0044] Figure 6 This is the distribution diagram of the prediction residual error of the method proposed in the present invention compared with traditional transfer learning.

[0045] The contents not described in detail in the specification of the present invention belong to the prior art known to the professional and technical personnel in this field.

Claims

1. A method for detecting common phase errors of splicing mirrors based on wavelet transform multi-channel input, characterized in that: The following steps are involved: Step 1: According to the principle of optical system, an optical system based on far-field imaging is designed, wherein the optical system comprises a main mirror formed by splicing two sub-mirrors (3-1), a mask (3-2), a focusing lens (3-3) and a detector (3-4), wherein an incident parallel light beam is reflected by the sub-mirror (3-1), passes through the mask (3-2), converges through the focusing lens (3-3), and is imaged on the image plane of the detector (3-4), and there is a common phase error between the sub-mirrors (3-1); Step 2: Taking one circular hole as the reference, randomly introduce 10,000 groups of common phase errors within one wavelength to the other circular hole, and introduce 20 dB Gaussian white noise to simulate the actual camera readout noise; Step 3: Collect 10,000 far-field images and record the corresponding common phase error values, construct image datasets and label datasets respectively, and select a training set from the image dataset; Step 4: Configure the deep learning environment and build a neural network model; Step 5: Preprocess the image data in the training set, decompose the single-channel image data into 4-channel components through a Haar wavelet, use it as the input of the neural network model, and obtain the predicted common phase error value through the neural network model; Step 6: Calculate the loss value between the predicted common error value and the given label, and optimize the neural network and adjust the parameters.

2. The method for detecting common phase errors of splicing mirrors based on wavelet transform multi-channel input according to claim 1 is characterized in that: The far-field image collected in step three includes the common phase error between the sub-mirrors.

3. The method for detecting common phase errors of splicing mirrors based on wavelet transform multi-channel input according to claim 1 is characterized in that: In the step three, 80% of the image data set is selected as a training set for the neural network model to learn the nonlinear mapping relationship between the image after Haar wavelet decomposition and the common phase error value, and 20% of the image data set is used as a validation set to measure the accuracy and real-time performance of the neural network model.

4. The method for detecting common phase errors of splicing mirrors based on wavelet transform multi-channel input according to claim 1 is characterized in that: The neural network model of step 4 adopts the convolutional neural network ResNet18, which includes 17 convolutional layers, 2 pooling layers and 1 fully connected layer, and the input is directly passed to the output through the residual connection in ResNet18.

5. The method for detecting common phase errors of splicing mirrors based on wavelet transform multi-channel input according to claim 1 is characterized in that: In step six, the difference between the predicted common error value and the given label is calculated by the mean square error loss function (MSE) to obtain the loss value.

6. The method for detecting common phase errors of splicing mirrors based on wavelet transform multi-channel input according to claim 1 is characterized in that: According to the principle of optical system, the optical system based on far-field imaging is designed, including: designing: incident light wavelength, wave number, focal length of focusing lens (3-3), and circular aperture.

7. The method for detecting common phase errors of splicing mirrors based on wavelet transform multi-channel input according to claim 6 is characterized in that: Incident light wavelength , wave number , the focal length of the focusing lens (3-3) , round hole .

8. The method for detecting common phase errors of splicing mirrors based on wavelet transform multi-channel input according to claim 1, characterized in that: Step three includes: The common phase error is simulated by applying a phase difference in the circular hole of the mask (3-2). The phase difference is a label. After being converged by the focusing lens (3-3), it is imaged on the detector (3-4). When the applied phase difference changes, the final far-field image changes accordingly, thereby obtaining a one-to-one corresponding label and far-field image.

9. The method for detecting common phase errors of splicing mirrors based on wavelet transform multi-channel input according to claim 8, characterized in that: Building the image dataset and the label dataset includes: building a folder to store all named far-field images; building another folder to store all labels containing image indexes.

10. The method for detecting common phase errors of splicing mirrors based on wavelet transform multi-channel input according to claim 1, characterized in that: In step 3, the mathematical expression of the image plane intensity is: ， in, represents the ideal distribution of point sources on the image plane, represents the convolution operation, is the camera noise, are the horizontal and vertical coordinates of the image plane, respectively. The intensity point spread function PSF is expressed as: ， in, is the wavelength of incident light, is the focal length of the lens, represents the generalized pupil function of the optical system, is the Fourier transform.

Images