Polarization optical detection system polarization measurement error calibration method based on deep learning

Abstract

The invention discloses a polarization optical detection system polarization measurement error calibration method based on deep learning, and solves the problem that polarization measurement errors introduced by polarization detection optical system instrument polarization and optical system imaging quality cannot be calibrated at the same time in the prior art. Comprising the following steps: 1, obtaining a modulation matrix of a polarization optical detection system to be calibrated; 2, acquiring a clean intensity image matrix and a degraded intensity image matrix; 3, constructing a polarization image restoration network; 4, training a polarization image restoration network; 5, acquiring an actual degradation intensity image matrix; 6, inputting the actual degradation intensity image matrix into the polarization image restoration network to obtain a predicted clean intensity image matrix; 7, calculating a polarization measurement error; 8, obtaining an instrument polarization characteristic measurement error; and 9, calculating the polarization measurement error of the polarization optical detection system to be calibrated according to the polarization measurement error and the instrument polarization characteristic measurement error, and completing calibration.

Description

Technical Field

[0001] The present invention relates to a polarization measurement error calibration method, and in particular to a polarization measurement error calibration method for a polarization optical detection system based on deep learning. Background Art

[0002] Compared with traditional intensity imaging detection technology, polarization optical detection technology carries polarization information of light waves or light waves after the interaction between light and matter. It can provide more complete light field information including intensity and polarization, and obtain higher-dimensional information of the target, such as target geometry, composition, microstructure, etc. It is widely used in industrial monitoring, medical diagnosis, astronomical observation and other fields.

[0003] According to the principles and methods of polarization optical detection, a polarization detection optical system usually consists of a polarization modulation subsystem and an imaging subsystem. When performing polarization detection on a target, polarization measurement errors may be introduced into the polarization measurement results due to differences between the polarization characteristics of the instrument in the actual polarization detection system and the nominal polarization value of the instrument, and between the imaging quality of the optical system and the ideal imaging quality.

[0004] To improve polarization measurement accuracy, the polarization measurement error of the polarization detection optical system needs to be calibrated before effective detection. In addition to building a polarization calibration system to calibrate the polarization detection optical system, some polarization detection optical systems are equipped with a polarization calibration unit to meet high-precision detection requirements. Polarization calibration units vary depending on the polarization calibration method. For example, the commonly used four-point polarization calibration method uses a polarization calibration unit composed of a linear polarizer and a phase delay element (such as a λ / 4 wave plate). Existing calibration methods mostly focus on the instrument polarization error of the polarization detection optical system, lacking technical methods for simultaneously calibrating the imaging quality of the polarization detection optical system. Furthermore, if errors caused by both the instrument polarization and imaging quality of the system are to be calibrated simultaneously, the corresponding polarization calibration system structure is complex and the operation process is cumbersome. This seriously restricts the development of high-precision polarization measurement technology, especially high-precision spatial polarization measurement technology. Summary of the Invention

[0005] In order to solve the technical problems that the existing technology cannot simultaneously calibrate the polarization measurement errors introduced by the polarization of polarization optical detection system instruments and the imaging quality of the optical system, and that the existing polarization calibration system has a complex structure and cumbersome operation process, which seriously restricts the further improvement of the polarization measurement accuracy of the polarization detection optical system, the present invention provides a polarization measurement error calibration method for a polarization optical detection system based on deep learning.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions: A method for calibrating polarization measurement errors of a polarization optical detection system based on deep learning is particularly characterized by comprising the following steps: Step 1: Obtain the modulation matrix of the polarization optical detection system to be calibrated ; Step 2: Get the clean intensity image matrix P and the degraded intensity image matrix Q ; Step 3: Build a polarization image restoration network based on the UNet model; Step 4: Degrade the intensity image matrix Q As input data, it is infinitely close to the clean intensity image matrix P The clean intensity image matrix P t To output data, train the polarization image restoration network; Step 5: Obtain the actual degraded intensity image matrix of the polarization optical detection system to be calibrated Q r ; Step 6: Convert the actual degradation intensity image matrix Q r Input the trained polarization image restoration network to obtain the predicted clean intensity image matrix P r ; Step 7: According to the modulation matrix , actual degradation intensity image matrix Q r and the predicted clean intensity image matrix P r Calculate the polarization measurement error introduced by the polarization optical detection system to be calibrated ; Step 8: Obtain the instrument polarization characteristic measurement error of the polarization optical detection system to be calibrated ; Step 9: Measure the error based on polarization and instrument polarization characteristic measurement error Calculating the polarization measurement error of a polarization optical detection system to be calibrated , complete the calibration.

[0007] Furthermore, step 2 specifically includes: 2.1. Generate a clean intensity image matrix with polarization information P ; The polarization information is described by Stokes vector parameters; 2.2. Generating Zernike polynomials to characterize imaging quality Z ; 2.3、Clean intensity image matrix Pand Zernike polynomials Z Perform convolution to obtain the corresponding degraded intensity image matrix Q .

[0008] Furthermore, in step 3, the polarization image restoration network includes a first residual convolution block to an eighth residual convolution block connected in sequence from the input end to the output end; at the same time, the other output end of the first residual convolution block is also connected to the other input end of the eighth residual convolution block; the other output end of the second residual convolution block is also connected to the other input end of the seventh residual convolution block; the other output end of the third residual convolution block is also connected to the other input end of the sixth residual convolution block; wherein the first residual convolution block to the fourth residual convolution block are encoders, and the fifth residual convolution block to the eighth residual convolution block are decoders.

[0009] Furthermore, in step 3, the sizes of the output images E1, E2, and E3 from the first residual convolution block to the third residual convolution block satisfy: the width and height of the output image E1 are twice that of the output image E2, and the width and height of the output image E2 are twice that of the output image E3.

[0010] Furthermore, step 4 specifically includes: 4.1. Construct aberration guidance maps C1, C2, and C3 of the same size as the output images E1, E2, and E3 of the first to third residual convolution blocks; 4.2. Degraded Intensity Image Matrix Q As input data, the aberration guidance maps C1, C2, and C3 are added to the output images E1, E2, and E3 of the first to third residual convolution blocks respectively to be infinitely close to the clean intensity image matrix P The clean intensity image matrix P t To output data, the polarization image restoration network is trained using a root mean square function, a multi-scale structure loss function, and / or a total variation function as a loss function until the loss function converges, thereby completing the network training.

[0011] Furthermore, step 4.1 is specifically as follows: The aberration distribution principle is used to construct aberration guidance maps C1, C2, and C3 with the same size as the output images E1, E2, and E3 of the first to third residual convolution blocks.

[0012] Furthermore, in step 4.1, the aberration distribution principle is:

[0013] in, is the aberration, which is given by the Zernike polynomial in step 2.2 Z characterization, is the aberration coefficient, 、 and for 、 and The power of , taking a non-negative integer, is the normalized image height, representing different fields of view, and Indicates the position of the light at the exit pupil; when the Cartesian coordinate system is used to represent the coordinates of the exit pupil, the center of the exit pupil is located at the origin of the corresponding coordinate system. Represents the distance between the corresponding light position at the system exit pupil and the center of the exit pupil, The corresponding light position and The angle between the axes; 、 、 and corresponding The selection of is used to construct different types of aberrations , The selection of is used to construct the aberration of different fields of view .

[0014] Furthermore, step 7 specifically includes: 7.1. Calculate the predicted clean intensity image matrix P r Corresponding polarization information :

[0015] 7.2. Calculating the Actual Degraded Intensity Image Matrix Q r Corresponding polarization information :

[0016] 7.3 Polarization Information With polarization information Perform difference calculation to obtain the polarization measurement error introduced by the polarization optical detection system to be calibrated : .

[0017] Furthermore, step 8 is specifically as follows: If the polarization optical detection system to be calibrated includes a polarization calibration unit, the polarization characteristic measurement error of the instrument is obtained through its polarization calibration unit. ; If the polarization optical detection system to be calibrated does not include a polarization calibration unit, the polarization characteristic measurement error of the instrument is obtained through an external polarization calibration system. .

[0018] Furthermore, step 9 is specifically as follows: According to the polarization measurement error and instrument polarization characteristic measurement error Calculating the polarization measurement error of a polarization optical detection system to be calibrated :

[0019] Calibration completed.

[0020] Beneficial effects of the present invention: 1. The present invention provides a method for calibrating the polarization measurement error of a polarization optical detection system based on deep learning. By constructing a polarization image restoration network with a UNet model as the basic architecture, a predicted clean intensity image matrix without aberration influence is obtained through the polarization image restoration network. Combined with the instrument polarization characteristic measurement error of the polarization optical detection system to be calibrated, the calibration accuracy of the polarization detection optical system can be effectively improved, thereby improving the polarization measurement accuracy of the polarization detection optical system.

[0021] 2. The present invention provides a method for calibrating polarization measurement errors of a polarization optical detection system based on deep learning, which uses the clean intensity image matrix to calculate the polarization error of the polarization optical detection system. Z Characterize the degraded intensity image matrix under the effect of aberration Q , construct an aberration guidance map to train the constructed polarization image restoration network, which can enable the trained polarization image restoration network to obtain a predicted clean intensity image matrix without aberration influence, and can decouple the effect of the imaging quality of the polarization optical system on the polarization measurement accuracy, which can effectively improve the polarization measurement accuracy of the polarization detection optical system.

[0022] 3. The present invention provides a polarization measurement error calibration method for a polarization optical detection system based on deep learning. The polarization image restoration network is trained using a root mean square function, a multi-scale structure loss function, and / or a total variation function as a loss function. The method can adaptively select a suitable loss function according to the complexity of the polarization image information, thereby improving the prediction accuracy of the polarization image restoration network.

[0023] 4. The present invention provides a polarization measurement error calibration method for a polarization optical detection system based on deep learning. The aberration distribution principle is used to construct aberration guidance maps C1, C2, and C3 with the same size as the output images E1, E2, and E3 of the first to third residual convolution blocks. When training the network, the aberration guidance maps C1, C2, and C3 are used as input at the same time, which can further improve the prediction accuracy of the high-polarization image restoration network.

[0024] 5. The present invention provides a method for calibrating the polarization measurement error of a polarization optical detection system based on deep learning. Compared with the existing technology, this method does not require the setting of an additional polarization calibration system, which can make the structure of the polarization optical detection system simpler and more compact. It provides a new method for the on-orbit high-precision calibration of spatial polarization optical detection systems with strict envelope requirements such as structural weight, and can effectively promote the development of space detection technology based on polarization optics. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 This is a flow chart of an embodiment of a method for calibrating polarization measurement errors of a polarization optical detection system based on deep learning according to the present invention; Figure 2 3 is a schematic diagram of the polarization image restoration network structure constructed in step 3 of an embodiment of the present invention.

[0026] Reference numerals: 1-first residual convolution block, 2-second residual convolution block, 3-third residual convolution block, 4-fourth residual convolution block, 5-fifth residual convolution block, 6-sixth residual convolution block, 7-seventh residual convolution block, 8-eighth residual convolution block. DETAILED DESCRIPTION

[0027] The following will clearly and completely describe the technical solutions of the present invention in conjunction with the accompanying drawings and embodiments. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0028] The embodiment of the present invention provides a method for calibrating polarization measurement errors of a polarization optical detection system based on deep learning, such as Figure 1 As shown, the method includes the following steps: Step 1: Obtain the modulation matrix of the polarization optical detection system to be calibrated ; Determine the modulation matrix of the polarization optical detection system to be calibrated according to its detection method or modulation method ; The polarization optical detection system to be calibrated usually consists of a polarization modulation unit and an imaging optical system. Some polarization optical detection systems are also equipped with a polarization calibration unit for calibrating the polarization measurement error of the instrument. Step 2: Obtain a clean intensity image matrix with polarization information P and the degraded intensity image matrix Q ; 2.1. Generate a clean intensity image matrix with polarization information P ; The polarization information is described by Stokes vector parameters; 2.2. Generating Zernike polynomials to characterize imaging quality Z ; 2.3、Clean intensity image matrix P and Zernike polynomials Z Perform convolution to obtain the corresponding degraded intensity image matrix Q ; The degraded intensity image matrix Q With polarization information; Step 3: Build a polarization image restoration network based on the UNet model; like Figure 2 As shown, the polarization image restoration network includes a first residual convolution block 1 to an eighth residual convolution block 8 connected in sequence from the input end to the output end; at the same time, the other output end of the first residual convolution block 1 is also connected to the other input end of the eighth residual convolution block 8; the other output end of the second residual convolution block 2 is also connected to the other input end of the seventh residual convolution block 7; the other output end of the third residual convolution block 3 is also connected to the other input end of the sixth residual convolution block 6; wherein the first residual convolution block 1 to the fourth residual convolution block 4 are encoders, and the fifth residual convolution block 5 to the eighth residual convolution block 8 are decoders; the sizes of the output images E1, E2, and E3 of the first residual convolution block 1 to the third residual convolution block 3 satisfy: the width and height of the output image E1 are twice that of the output image E2, and the width and height of the output image E2 are twice that of the output image E3; Step 4: Train the polarization image restoration network; 4.1. Using the aberration distribution principle, construct aberration guidance maps C1, C2, and C3 with the same size as the output images E1, E2, and E3 of the first residual convolution block 1 to the third residual convolution block 3; The aberration distribution principle is: ;in, is the aberration, which is given by the Zernike polynomial in step 2.2 Z characterization, is the aberration coefficient, 、 and for 、 and The power of , taking a non-negative integer, is the normalized image height, representing different fields of view, and Indicates the position of the light at the exit pupil; when the Cartesian coordinate system is used to represent the coordinates of the exit pupil, the center of the exit pupil is located at the origin of the corresponding coordinate system. Represents the distance between the corresponding light position at the system exit pupil and the center of the exit pupil, The corresponding light position and The angle between the axes; By selecting different 、 、 and corresponding Constructing different kinds of aberrations , by choosing different normalized image heights , constructing aberrations of different fields of view ; Thus completing the construction of the aberration guidance map.

[0029] 4.2. Degraded Intensity Image Matrix Q As input data, the aberration guidance maps C1, C2, and C3 are added to the output images E1, E2, and E3 of the first to third residual convolution blocks respectively to be infinitely close to the clean intensity image matrix P The clean intensity image matrix P t To output data, the polarization image restoration network is trained using a root mean square function, a multi-scale structure loss function, and / or a total variation function as a loss function until the loss function converges, thereby completing the network training; That is: input the degraded intensity image matrix into the polarization image restoration network Q , then its output clean intensity image matrix P t , the clean polarization image matrix obtained by the evaluation network P t With the clean intensity image matrix P The similarity is calculated until the loss function converges and the network training is completed.

[0030] Among them, the root mean square function, the multi-scale structure loss function and / or the total variation function are adaptively selected according to the complexity of the polarization image information to improve the prediction accuracy of the polarization image restoration network.

[0031] Step 5: Obtain the actual degraded intensity image matrix of the polarization optical detection system to be calibrated Q r ; Step 6: Convert the actual degradation intensity image matrix Q r Input the trained polarization image restoration network to obtain the predicted clean intensity image matrix P r ; Step 7: According to the modulation matrix , actual degradation intensity image matrix Q r and the predicted clean intensity image matrix P r Calculate the polarization measurement error introduced by the polarization optical detection system to be calibrated ; 7.1. Calculate the predicted clean intensity image matrix P r Corresponding polarization information , that is, the polarization measurement result :

[0032] 7.2. Calculating the Actual Degraded Intensity Image Matrix Q r Corresponding polarization information , that is, the polarization measurement result :

[0033] 7.3 Polarization measurement results Polarization measurement results Perform difference calculation to obtain the polarization measurement error introduced by the polarization optical detection system to be calibrated : ; Step 8: Obtain the instrument polarization characteristic measurement error of the polarization optical detection system to be calibrated ; If the polarization optical detection system to be calibrated includes a polarization calibration unit, the polarization characteristic measurement error of the instrument is obtained through its polarization calibration unit. ; If the polarization optical detection system to be calibrated does not include a polarization calibration unit, the polarization characteristic measurement error of the instrument is obtained through an external polarization calibration system. ; Step 9: Measure the error based on polarization and instrument polarization characteristic measurement error Calculating the polarization measurement error of a polarization optical detection system to be calibrated :

[0034] Calibration completed.

[0035] This method can effectively achieve high-precision calibration of the system polarization measurement error caused by optical imaging quality and instrument polarization in the polarization optical detection system. It can provide a new method for the on-orbit high-precision calibration of the spatial polarization optical detection system and provide technical support for further improving the polarization measurement accuracy of the polarization detection optical system.

[0036] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions within the technical scope disclosed in the present invention shall be covered by the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be subject to the scope of protection of the claims.

Claims

1. A method for calibrating polarization measurement errors in a polarization optical detection system based on deep learning, characterized in that: The following steps are involved: Step 1: Obtain the modulation matrix of the polarization optical detection system to be calibrated ; Step 2: Get the clean intensity image matrix P and the degraded intensity image matrix Q ; Step 3: Build a polarization image restoration network based on the UNet model; Step 4: Degrade the intensity image matrix Q As input data, it is infinitely close to the clean intensity image matrix P The clean intensity image matrix P t To output data, train the polarization image restoration network; Step 5: Obtain the actual degraded intensity image matrix of the polarization optical detection system to be calibrated Q r ; Step 6: Convert the actual degradation intensity image matrix Q r Input the trained polarization image restoration network to obtain the predicted clean intensity image matrix P r ; Step 7: According to the modulation matrix , actual degradation intensity image matrix Q r and the predicted clean intensity image matrix P r Calculate the polarization measurement error introduced by the polarization optical detection system to be calibrated ; Step 8: Obtain the instrument polarization characteristic measurement error of the polarization optical detection system to be calibrated ; Step 9: Measure the error based on polarization and instrument polarization characteristic measurement error Calculating the polarization measurement error of a polarization optical detection system to be calibrated , complete the calibration.

2. The method for calibrating polarization measurement errors of a polarization optical detection system based on deep learning according to claim 1, wherein: Step 2 specifically includes: 2.

1. Generate a clean intensity image matrix with polarization information P ; The polarization information is described by Stokes vector parameters; 2.

2. Generating Zernike polynomials to characterize imaging quality Z ; 2.3、Clean intensity image matrix P and Zernike polynomials Z Perform convolution to obtain the corresponding degraded intensity image matrix Q .

3. The method for calibrating polarization measurement errors of a polarization optical detection system based on deep learning according to claim 2, characterized in that: In step 3, the polarization image restoration network includes a first residual convolution block to an eighth residual convolution block connected in sequence from the input end to the output end; at the same time, the other output end of the first residual convolution block is also connected to the other input end of the eighth residual convolution block; the other output end of the second residual convolution block is also connected to the other input end of the seventh residual convolution block; the other output end of the third residual convolution block is also connected to the other input end of the sixth residual convolution block; wherein the first residual convolution block to the fourth residual convolution block are encoders, and the fifth residual convolution block to the eighth residual convolution block are decoders.

4. The method for calibrating polarization measurement errors of a polarization optical detection system based on deep learning according to claim 3, characterized in that: In step 3, the sizes of the output images E1, E2, and E3 from the first residual convolution block to the third residual convolution block satisfy: the width and height of the output image E1 are twice that of the output image E2, and the width and height of the output image E2 are twice that of the output image E3.

5. The method for calibrating polarization measurement errors of a polarization optical detection system based on deep learning according to claim 4, wherein: Step 4 specifically includes: 4.

1. Construct aberration guidance maps C1, C2, and C3 of the same size as the output images E1, E2, and E3 of the first to third residual convolution blocks; 4.

2. Degraded Intensity Image Matrix Q As input data, the aberration guidance maps C1, C2, and C3 are added to the output images E1, E2, and E3 of the first to third residual convolution blocks respectively to be infinitely close to the clean intensity image matrix P The clean intensity image matrix P t To output data, the polarization image restoration network is trained using a root mean square function, a multi-scale structure loss function, and / or a total variation function as a loss function until the loss function converges, thereby completing the network training.

6. The method for calibrating polarization measurement errors of a polarization optical detection system based on deep learning according to claim 5, characterized in that: Step 4.1 is as follows: The aberration distribution principle is used to construct aberration guidance maps C1, C2, and C3 with the same size as the output images E1, E2, and E3 of the first to third residual convolution blocks.

7. The method for calibrating polarization measurement errors of a polarization optical detection system based on deep learning according to claim 6, wherein: In step 4.1, the aberration distribution principle is: ; in, is the aberration, which is given by the Zernike polynomial in step 2.2 Z characterization, is the aberration coefficient, 、 and for 、 and The power of , taking a non-negative integer, is the normalized image height, representing different fields of view, and Indicates the position of the light at the exit pupil; when the Cartesian coordinate system is used to represent the coordinates of the exit pupil, the center of the exit pupil is located at the origin of the corresponding coordinate system. Represents the distance between the corresponding light position at the system exit pupil and the center of the exit pupil, The corresponding light position and The angle between the axes; 、 、 and corresponding The selection of is used to construct different types of aberrations , The selection of is used to construct the aberration of different fields of view .

8. The method for calibrating polarization measurement errors of a polarization optical detection system based on deep learning according to claim 7, wherein: Step 7 specifically includes: 7.

1. Calculate the predicted clean intensity image matrix P r Corresponding polarization information : ; 7.

2. Calculating the Actual Degraded Intensity Image Matrix Q r Corresponding polarization information : ; 7.3 Polarization Information With polarization information Perform difference calculation to obtain the polarization measurement error introduced by the polarization optical detection system to be calibrated : 。 9. The method for calibrating polarization measurement errors of a polarization optical detection system based on deep learning according to claim 8, characterized in that: Step 8 is as follows: If the polarization optical detection system to be calibrated includes a polarization calibration unit, the polarization characteristic measurement error of the instrument is obtained through its polarization calibration unit. ; If the polarization optical detection system to be calibrated does not include a polarization calibration unit, the polarization characteristic measurement error of the instrument is obtained through an external polarization calibration system. .

10. The method for calibrating polarization measurement errors of a polarization optical detection system based on deep learning according to claim 9, characterized in that: Step 9 is as follows: According to the polarization measurement error and instrument polarization characteristic measurement error Calculating the polarization measurement error of a polarization optical detection system to be calibrated : ; Calibration completed.

Images