Single-frame high-quality sheared speckle interferometric phase extraction method and device

Abstract

According to the single-frame high-quality shearing speckle interference phase extraction method and device, the quality of the obtained phase is comparable to that of a traditional multi-frame phase solution algorithm, and single-frame high-quality phase extraction of shearing speckle interference is completed. The method comprises the following steps: (1) making a data set; (2) constructing a neural network; (3) neural network training; (4) detecting the detected piece: when the detected piece is static, carrying out N-step phase shift method shooting to obtain N frames of speckle interferograms when the detected piece is static; applying an external force to the measured piece, and recording light intensity information by using a camera to obtain a frame of interferogram when the measured piece is deformed; (5) single-frame phase extraction: phase extraction is carried out according to an N + 1 time phase shift method, and on the basis of N frames of phase shift interferograms when the measured piece is static, phase extraction is carried out by using one frame of interferograms after the measured piece is deformed, so that low-quality phase distribution is obtained; and (6) phase repairing.

Description

Technical Field

[0001] The present invention relates to the technical field of photoelectric detection, and in particular to a single-frame high-quality shearing speckle interferometry phase extraction method, which aims to obtain high-quality phase comparable to that of traditional multi-frame phase deconvolution algorithms and complete single-frame high-quality extraction of the phase of shearing speckle interferometry. It also relates to a single-frame high-quality shearing speckle interferometry phase extraction device. Background Art

[0002] With the rapid advancement of material processing technology, composite materials that combine lightweight characteristics with customizable mechanical properties have become an important direction in the field of material research and development, and are widely used in high-end industrial fields such as aerospace. However, composite materials are prone to defects during processing, and some components are often subjected to extreme working conditions (such as high temperature and high pressure environments) for a long time, resulting in frequent internal damage problems in the materials. Currently, the demand for quantitative testing, noise-resistant testing, and transient testing of composite materials in the field of industrial testing is increasing, and dynamic testing and high-precision testing are required for material testing.

[0003] Shear speckle interferometry (SSI) is a highly sensitive optical interferometry technique capable of detecting minute deformation gradients in a test piece. It requires recording speckle patterns before and after deformation, then performing phase extraction for quantitative analysis. With increasing industrial demand for dynamic and high-precision detection, quantitative accuracy and dynamic detection capabilities have become two key development goals for SSI. Therefore, as one of the core steps in SSI, achieving high-quality dynamic phase extraction simultaneously has become a key research topic.

[0004] The most classic phase extraction method for shear speckle interferometry is the time-shifting method. Because shear speckle interferometry requires phase extraction both before and after the DUT deforms, this classic phase-shifting method is known as the N+N (N=3, 4, 5, etc.) time-shifting method. Piezoelectric ceramics, due to their simple structure and precise control of the phase-shifting step size, have become the mainstream phase-shifting device for this method. The N+N time-shifting method for shear speckle interferometry achieves high phase extraction quality, resulting in clear fringe boundaries and facilitating subsequent unwrapping processing. However, this method requires at least three speckle image frames to calculate the phase, making it difficult to measure the dynamic deformation of the DUT.

[0005] To achieve dynamic detection, some researchers have conducted research based on spatial carrier methods. Spatial carrier technology extracts phase information using a fixed spatial carrier frequency for single-frame dephasing. However, due to spectral aliasing and the low signal-to-noise ratio caused by the aperture, the phase extraction quality of spatial carrier algorithms is poor. These factors often lead to phase ambiguity and discontinuity in the phase extraction results of spatial carriers, which directly leads to errors in subsequent phase unwrapping and limits the application of shearing speckle interferometry for quantitative analysis.

[0006] Another group of scholars have improved the dynamic detection capability of the time phase shifting method by improving it. Some scholars have realized dynamic phase extraction based on N (N = 3, 4, 5, etc.) frames of speckle images before the deformation of the test piece and 1 frame of speckle image after deformation. This method is called the N+1 time phase shifting method. This type of method sacrifices phase quality while achieving dynamic detection, and the calculated phase is ambiguous and broken. In addition, some scholars have proposed the N+2 time phase shifting method, but this method only makes a trade-off between accuracy and dynamic detection capability between the N+N and N+1 methods, and still cannot achieve dynamic high-precision phase extraction at the same time.

[0007] However, the existing shearing speckle interferometry phase extraction technology cannot achieve single-frame and high-quality phase extraction at the same time. Summary of the Invention

[0008] To overcome the shortcomings of the prior art, the technical problem to be solved by the present invention is to provide a single-frame high-quality shearing speckle interferometry phase extraction method. The method uses a neural network to repair the low-quality phase obtained by the traditional single-frame phase deconvolution algorithm. The quality of the obtained phase is comparable to that of the traditional multi-frame phase deconvolution algorithm. The single-frame advantage of the single-frame phase deconvolution method and the high-quality advantage of the multi-frame phase deconvolution method are strongly combined to complete the single-frame high-quality extraction of the phase of shearing speckle interferometry.

[0009] The technical solution of the present invention is: this single-frame high-quality shearing speckle interferometry phase extraction method comprises the following steps:

[0010] (1) Dataset preparation: In the shear speckle interferometry system, piezoelectric ceramics are used for phase shifting. The N+N time phase shifting method is used to obtain high-quality phases. Noise is added to obtain the corresponding low-quality phases to complete the dataset preparation.

[0011] (2) Neural network construction: Build a neural network to perform phase repair with low-quality phase as input and high-quality phase as output;

[0012] (3) Neural network training: Use the data set obtained in step (1) to train the network so that the neural network has the ability to perform phase repair.

[0013] (4) Detection of the test piece: When the test piece is stationary, perform N-step phase shifting method to obtain N frames of speckle interferograms when the test piece is stationary; apply external force to the test piece and use a camera to record the light intensity information to obtain 1 frame of interferogram when the test piece is deformed;

[0014] (5) Single-frame phase extraction: Phase extraction is performed based on the N+1 time phase shifting method. Based on the N-frame phase shifting interferogram when the DUT is stationary, the phase is extracted using the 1-frame interferogram after the DUT is deformed to obtain a low-quality phase distribution.

[0015] (6) Phase repair: Use the phase repair network obtained in step (3) to repair the phase of step

[0016] The low-quality phase obtained in (5) is repaired to finally obtain a high-quality phase distribution.

[0017] This paper establishes a single-frame, high-quality shearing speckle interferometry phase extraction method through the steps of dataset preparation, neural network construction, neural network training, DUT inspection, single-frame phase extraction, and phase repair. This method utilizes a neural network to repair the low-quality phase obtained by traditional single-frame dephasing algorithms. The resulting phase quality is comparable to that of traditional multi-frame dephasing algorithms. This method combines the single-frame advantages of single-frame dephasing methods with the precision advantages of time dephasing methods to achieve high-quality single-frame phase extraction using shearing speckle interferometry. The entire process is convenient and efficient, with a low detection threshold, achieving single-frame, high-quality phase extraction, which is more beneficial for industrial inspection and material maintenance.

[0018] A single-frame high-quality shearing speckle interferometry phase extraction device is also provided, comprising:

[0019] The dataset creation module is configured to use piezoelectric ceramics for phase shifting in a shear speckle interferometry system, using the N+N time phase shifting method to obtain high-quality phases, and adding noise to obtain the corresponding low-quality phases to complete the dataset creation;

[0020] A neural network building module configured to build a neural network to meet the phase repair requirements of low-quality phase input and high-quality phase output;

[0021] A neural network training module is configured to use the data set obtained in the data set preparation module to perform network training, so that the neural network has the ability to perform phase repair;

[0022] The DUT detection module is configured to perform N-step phase shifting when the DUT is stationary.

[0023] Take pictures to obtain N frames of speckle interference patterns when the DUT is stationary; apply external force to the DUT and use a camera to record light intensity information to obtain 1 frame of interference pattern when the DUT is deformed;

[0024] The single frame phase extraction module is configured to extract phase according to the N+1 time phase shifting method.

[0025] Taking the N frames of phase-shifted interferograms when the DUT is stationary as the basis, the phase is extracted using the 1 frame of interferogram after the DUT is deformed to obtain a low-quality phase distribution;

[0026] The phase repair module is configured to use the phase repair network obtained by the neural network training module to repair the low-quality phase obtained in the single-frame phase extraction module, and ultimately obtain a high-quality phase distribution. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 This is a flow chart of the single-frame high-quality shearing speckle interferometry phase extraction method of the present invention.

[0028] Figure 2 It is a structural schematic diagram of the single-frame high-quality shearing speckle interferometry phase extraction system of the present invention.

[0029] Figure 3 It is the high-quality phase distribution obtained by the N+N time phase shifting method.

[0030] Figure 4 is the low-quality phase distribution obtained after adding noise.

[0031] Among them: 1- shearing speckle interferometry module, 2- CMOS camera, 3- first lens, 4- second lens, 5- imaging lens, 6- test piece, 7- laser, 8- beam splitter, 9- piezoelectric ceramic, 10- reflector, 11- shearing mirror, 12- computer. DETAILED DESCRIPTION

[0032] like Figure 1 As shown, this single-frame high-quality shearing speckle interferometry phase extraction method includes the following steps:

[0033] (1) Dataset preparation: In the shear speckle interferometry system, piezoelectric ceramics are used for phase shifting. The N+N time phase shifting method is used to obtain high-quality phases. Noise is added to obtain the corresponding low-quality phases to complete the dataset preparation.

[0034] (2) Neural network construction: Build a neural network to perform phase repair with low-quality phase as input and high-quality phase as output;

[0035] (3) Neural network training: Use the data set obtained in step (1) to train the network so that the neural network has the ability to perform phase repair.

[0036] (4) Detection of the test piece: When the test piece is stationary, perform N-step phase shifting method to obtain N frames of speckle interferograms when the test piece is stationary; apply external force to the test piece and use a camera to record the light intensity information to obtain 1 frame of interferogram when the test piece is deformed;

[0037] (5) Single-frame phase extraction: Phase extraction is performed based on the N+1 time phase shifting method. Based on the N-frame phase shifting interferogram when the DUT is stationary, the phase is extracted using the 1-frame interferogram after the DUT is deformed to obtain a low-quality phase distribution.

[0038] (6) Phase repair: Use the phase repair network obtained in step (3) to repair the phase of step (5)

[0039] The low-quality phase obtained in the experiment is repaired to finally obtain a high-quality phase distribution.

[0040] This paper establishes a single-frame, high-quality shearing speckle interferometry phase extraction method through the steps of dataset preparation, neural network construction, neural network training, DUT inspection, single-frame phase extraction, and phase repair. This method utilizes a neural network to repair the low-quality phase obtained by traditional single-frame dephasing algorithms. The resulting phase quality is comparable to that of traditional multi-frame dephasing algorithms. This method combines the single-frame advantages of single-frame dephasing methods with the precision advantages of time dephasing methods to achieve high-quality single-frame phase extraction using shearing speckle interferometry. The entire process is convenient and efficient, with a low detection threshold, achieving single-frame, high-quality phase extraction, which is more beneficial for industrial inspection and material maintenance.

[0041] Preferably, in step (1), the N+N time phase shifting method selects the 4+4 time phase shifting method to ensure high quality of phase extraction, and Gaussian noise is added to all 4+4 frame interference patterns to truly simulate the noise interference during actual single-frame phase extraction.

[0042] Preferably, in step (2), a graph-based neural network with a Unet architecture is used.

[0043] Preferably, in step (3), the neural network is fully trained and overfitting is avoided.

[0044] Preferably, in step (4), piezoelectric ceramics are used to perform four-step phase shift shooting before the test piece is deformed, and single-frame detection is performed during the deformation of the test piece:

[0045]

[0046] Among them, I1, I2, I3, I4 are the interference patterns before the DUT is loaded, I, is the interference pattern after the DUT is loaded, A is the background of the interference pattern, B is the interference modulation term, It is the basic phase

[0047] Cloth, Δ is the phase difference caused by the external force, which is also the final measurement result of shear speckle interferometry.

[0048] Preferably, in step (5), the phase is extracted using the 4+1 time phase shifting method formula:

[0049]

[0050] Preferably, in step (6), the image size and value range of the single-frame de-phased result are kept consistent with the neural network.

[0051] Those skilled in the art will appreciate that all or part of the steps in the above-described method can be implemented by instructing the relevant hardware through a program. The program can be stored in a computer-readable storage medium. When executed, the program includes the steps of the above-described method. The storage medium can be ROM / RAM, a magnetic disk, an optical disk, a memory card, etc. Therefore, corresponding to the method of the present invention, the present invention also includes a single-frame high-quality shearing speckle interferometry phase extraction device, which is generally represented in the form of functional modules corresponding to the steps of the method. The device includes:

[0052] The dataset creation module is configured to use piezoelectric ceramics for phase shifting in a shear speckle interferometry system, using the N+N time phase shifting method to obtain high-quality phases, and adding noise to obtain the corresponding low-quality phases to complete the dataset creation;

[0053] A neural network building module configured to build a neural network to meet the phase repair requirements of low-quality phase input and high-quality phase output;

[0054] A neural network training module is configured to use the data set obtained in the data set preparation module to perform network training, so that the neural network has the ability to perform phase repair;

[0055] The DUT detection module is configured to perform N-step phase shifting when the DUT is stationary.

[0056] Take pictures to obtain N frames of speckle interference patterns when the DUT is stationary; apply external force to the DUT and use a camera to record light intensity information to obtain 1 frame of interference pattern when the DUT is deformed;

[0057] The single frame phase extraction module is configured to extract phase according to the N+1 time phase shifting method.

[0058] Taking the N frames of phase-shifted interferograms when the DUT is stationary as the basis, the phase is extracted using the 1 frame of interferogram after the DUT is deformed to obtain a low-quality phase distribution;

[0059] The phase repair module is configured to use the phase repair network obtained by the neural network training module to repair the low-quality phase obtained in the single-frame phase extraction module, and ultimately obtain a high-quality phase distribution.

[0060] Preferably, in the neural network building module, a graph-based neural network with a Unet architecture is used; in the neural network training module, the neural network is fully trained and overfitting is avoided.

[0061] Preferably, in the test piece detection module, piezoelectric ceramics are used to perform four-step phase shift shooting before the test piece is deformed, and single-frame detection is performed during the deformation of the test piece:

[0062]

[0063] Among them, I1, I2, I3, I4 are the interference patterns before the DUT is loaded, I, is the interference pattern after the DUT is loaded, A is the background of the interference pattern, B is the interference modulation term, It is the basic phase

[0064] Cloth, Δ is the phase difference caused by the external force, which is also the final measurement result of shear speckle interferometry;

[0065] In the single-frame phase extraction module, the 4+1 time phase shifting method formula is used for phase extraction:

[0066]

[0067] In the phase repair module, the image size and value range of the single-frame dephasing result are kept consistent with the neural network.

[0068] The following is a detailed description of a specific embodiment of the present invention. A single-frame high-quality shearing speckle interferometry phase extraction method is implemented in the following manner:

[0069] The process of establishing a single-frame high-quality shearing speckle interferometry phase extraction method is as follows: Figure 1 As shown, the specific implementation steps are:

[0070] Step 1: Dataset creation

[0071] In the example, Figure 2 In the shear speckle interferometry system shown in the figure, 1031 groups of 4+4 time phase shifting method phase extraction experiments were carried out, and the results were obtained. Figure 3 The high-quality phase image shown. After adding Gaussian noise to each group, we get Figure 4 The corresponding low-quality phase image is shown. A total of 20,620 data sets were collected. The data size was 224×224, and the dataset was divided into training, validation, and test sets in a ratio of 7:2:1 for network training.

[0072] Step 2: Neural Network Construction

[0073] In this example, the swin-unet network structure is built using pytorch, and the input and output dimensions are both 224×224.

[0074] Step 3: Neural Network Training

[0075] In this example, RMS is used as the loss function. For the first 400 epochs, the learning rate is set to 1e-4 to accelerate optimization. During epochs 400-500, the learning rate is reduced to 5e-5 for fine-tuning the network. Adam is used as the optimizer.

[0076] Step 4: DUT testing

[0077] In this example, piezoelectric ceramics are used to capture four phase-shifted images before the DUT deforms, and real-time single-frame images are captured during deformation. The camera model is BFS-U3-200S6C-C, with a resolution of 2736*1824 and a pixel size of 2.4μm.

[0078] Step 5: Single frame phase extraction

[0079] In this example, a 4+1 time phase shifting method is used for phase extraction.

[0080] Step 6: Phase Repair

[0081] In this example, the low-quality phase obtained by the 4+1 temporal phase shifting method is cropped and scaled to ensure that the image dimension of the input network is 224×224. The network directly outputs a high-quality phase with an accuracy comparable to that of the four-step phase shifting method.

[0082] The beneficial effects of the present invention are as follows:

[0083] 1. The single-frame high-quality shearing speckle interferometry phase extraction method disclosed in the present invention does not require complex optical path devices or complex post-processing algorithms, and can combine the advantages of single-frame phase extraction and multi-frame phase extraction to achieve single-frame high-quality phase extraction.

[0084] 2. The single-frame high-quality shearing speckle interferometry phase extraction device disclosed in the present invention has a simple structure and only relies on the traditional shearing speckle interferometry optical path structure. It can complete single-frame high-quality phase extraction in conjunction with the phase repair neural network, and the hardware modification cost is low.

[0085] The above description is merely a preferred embodiment of the present invention and does not constitute any form of limitation to the present invention. Any simple modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention are still within the scope of protection of the technical solution of the present invention.

Claims

1. Single-frame high-quality shearing speckle interferometry phase extraction method, characterized by: It includes the following steps: (1) Dataset preparation: In the shear speckle interferometry system, piezoelectric ceramics are used for phase shifting. The N+N time phase shifting method is used to obtain high-quality phases. Noise is added to obtain the corresponding low-quality phases to complete the dataset preparation. (2) Neural network construction: Build a neural network to perform phase repair with low-quality phase as input and high-quality phase as output; (3) Neural network training: Use the data set obtained in step (1) to train the network so that the neural network has the ability to perform phase repair. (4) Detection of the test piece: When the test piece is stationary, perform N-step phase shifting method to obtain N frames of speckle interferograms when the test piece is stationary; apply external force to the test piece and use a camera to record the light intensity information to obtain 1 frame of interferogram when the test piece is deformed; (5) Single-frame phase extraction: Phase extraction is performed based on the N+1 time phase shifting method. Based on the N-frame phase shifting interferogram when the DUT is stationary, the phase is extracted using the 1-frame interferogram after the DUT is deformed to obtain a low-quality phase distribution. (6) Phase repair: Use the phase repair network obtained in step (3) to repair the low-quality phase obtained in step (5) and finally obtain a high-quality phase distribution.

2. The single-frame high-quality shearing speckle interferometry phase extraction method according to claim 1, characterized in that: In step (1), the N+N time phase shifting method selects the 4+4 time phase shifting method to ensure high quality of phase extraction, and Gaussian noise is added to all 4+4 frame interference patterns to truly simulate the noise interference during actual single-frame phase extraction.

3. The single-frame high-quality shearing speckle interferometry phase extraction method according to claim 2, characterized in that: In the step (2), a graph-based neural network with a Unet architecture is used.

4. The single-frame high-quality shearing speckle interferometry phase extraction method according to claim 3, characterized in that: In the step (3), the neural network is fully trained and overfitting is avoided.

5. The single-frame high-quality shearing speckle interferometry phase extraction method according to claim 4, characterized in that: In step (4), piezoelectric ceramics are used to perform four-step phase shifting photography before the test piece is deformed, and single-frame detection is performed during the deformation of the test piece: Among them, I1, I2, I3, I4 are the interference patterns before the test piece is loaded, I ， is the interference pattern after the DUT is loaded, A is the background of the interference pattern, B is the interference modulation term, is the basic phase distribution, Δ is the phase difference caused by external force, and is also the final measurement result of shear speckle interferometry.

6. The single-frame high-quality shearing speckle interferometry phase extraction method according to claim 5, characterized in that: In step (5), the phase is extracted using the 4+1 time phase shifting method formula:

7. The single-frame high-quality shearing speckle interferometry phase extraction method according to claim 6, characterized in that: In the step (6), the image size and value range of the single-frame de-phased result are kept consistent with the neural network.

8. Single-frame high-quality shearing speckle interferometry phase extraction device, characterized by: It includes: The dataset creation module is configured to use piezoelectric ceramics for phase shifting in a shearing speckle interferometry system, using the N+N time phase shifting method to obtain high-quality phases, and adding noise to obtain the corresponding low-quality phases to complete the dataset creation; A neural network building module configured to build a neural network to meet the phase repair requirements of low-quality phase input and high-quality phase output; A neural network training module is configured to use the data set obtained in the data set preparation module to perform network training, so that the neural network has the ability to perform phase repair; The DUT detection module is configured to perform N-step phase shifting to capture images when the DUT is stationary, thereby obtaining N frames of speckle interferograms when the DUT is stationary; apply external force to the DUT and use a camera to record light intensity information to obtain one frame of interferogram when the DUT is deformed; The single-frame phase extraction module is configured to perform phase extraction based on the N+1 time phase shifting method. Based on the N-frame phase shifting interferogram when the DUT is stationary, the single-frame interferogram after the DUT is deformed is used for phase extraction to obtain a low-quality phase distribution. The phase repair module is configured to use the phase repair network obtained by the neural network training module to repair the low-quality phase obtained in the single-frame phase extraction module, and ultimately obtain a high-quality phase distribution.

9. The single-frame high-quality shearing speckle interferometry phase extraction device according to claim 8, characterized in that: In the neural network building module, a graph-based neural network with a Unet architecture is used; In the neural network training module, the neural network is fully trained and overfitting is avoided.

10. The single-frame high-quality shearing speckle interferometry phase extraction method according to claim 9, characterized in that: In the DUT detection module, piezoelectric ceramics are used to perform four-step phase shifting photography before the DUT is deformed, and single-frame detection is performed during the DUT deformation process: Among them, I1, I2, I3, I4 are the interference patterns before the test piece is loaded, I ， is the interference pattern after the DUT is loaded, A is the background of the interference pattern, B is the interference modulation term, is the basic phase distribution, Δ is the phase difference caused by the external force, and is also the final measurement result of shear speckle interferometry; in the single-frame phase extraction module, the 4+1 time phase shifting method formula is used for phase extraction: In the phase repair module, the image size and value range of the single-frame dephasing result are kept consistent with the neural network.

Images