Dynamic interference detection device and method based on generative adversarial neural network

Abstract

The invention discloses a dynamic interference detection device and method based on a generative adversarial neural network. According to the invention, deep learning is integrated into an interference measurement technology, a phase shift interference fringe generator is constructed through a generative adversarial network, an interference fringe set with a specified phase shift amount is obtained, and to-be-measured phase information is obtained through a two-step phase shift phase recovery algorithm. According to the dynamic interference detection device and method based on the generative adversarial neural network, dynamic interference measurement of single-frame fringes is achieved. The device disclosed by the invention has the advantages of high efficiency, simplicity and convenience in operation; the method has the advantages of high measurement precision and strong anti-interference capability, and is suitable for high-precision dynamic optical detection.

Description

Technical Field

[0001] The present invention belongs to the field of optical detection technology, and in particular relates to a dynamic interference detection device and method based on a generative adversarial neural network. Background Art

[0002] As the basic component of the optical system, the surface characteristics of the optical element largely determine the performance of the optical system. From spherical to aspherical surfaces, their use has greatly improved the performance of the optical system. With the development of science and technology, people have higher and higher requirements for the performance of optical systems and surface accuracy. In the fields of national defense, aerospace, space telemetry, etc., high-tech fields such as lithography chips play an increasingly important role. Therefore, the detection of high-precision optical components has become the focus of competition in the high-tech field of countries around the world.

[0003] Interferometry is a test and measurement method that uses the interference principle of light and the wavelength of light as the scale. The principle is to use an interferometer to emit a laser with a fixed wavelength, and use the reference light reflected from the reference sample surface and the measured light reflected from the sample surface to form interference fringes on the detector. The interference fringes are solved by the phase shift algorithm to obtain the surface information of the sample to be tested. It has the characteristics of non-contact, high sensitivity, and high precision. It can quickly and effectively obtain the surface information of optical components and other advantages. It is widely used in industrial production. The multi-step phase shift method is the phase extraction demodulation method adopted by the mainstream of interferometry. This method involves collecting multiple phase-shifted interference fringe patterns at different phases, and then calculating the phase surface through the phase shift algorithm. The more steps, the higher the processing accuracy; however, the lower the efficiency. In addition, system errors are easily accumulated in multi-step operations. At different phase shift time points, this method will introduce time errors and environmental errors, which have a significant impact on high-precision interferometry.

[0004] The inventors of the present application have discovered that since the error frequency caused by the vibration of the measuring platform may be twice the fringe frequency, realizing dynamic interferometric measurement of single-frame fringes is of great significance for solving the above-mentioned problem. Summary of the invention

[0005] The purpose of the present invention is to provide a dynamic interference detection device and method based on a generative adversarial neural network. The dynamic interference detection device based on the generative adversarial neural network, whose interference measurement device does not require a phase shifting device, includes but is not limited to a mechanical phase shifting device, a wavelength phase shifting device, and a polarization phase shifting device. The dynamic interference detection method based on a generative adversarial neural network uses a generative adversarial network to construct a phase-shifting interference fringe generator to generate virtual phase-shifting interference fringes, and obtains the phase information to be measured through a two-step phase-shifting phase recovery algorithm. The dynamic interference detection device and method based on a generative adversarial neural network realize dynamic interference measurement of single-frame fringes. The device in the present invention has the advantages of high efficiency, simplicity, and easy operation; the method has the advantages of high measurement accuracy and strong anti-interference ability. It is suitable for use in high-precision dynamic optical detection.

[0006] The technical solution for achieving the purpose of the present invention is:

[0007] A dynamic interference detection device based on a generative adversarial neural network comprises an interferometer, a reference mirror, a measured mirror, a displacement adjustment platform and a dynamic interference detection software system of a generative adversarial neural network; after a laser in the interferometer emits incident light of a fixed wavelength and passes through the reference mirror, a portion of the light returns to the interferometer and is received by a detector to form a reference light, and a portion of the light passes through the reference mirror to reach the surface of the measured mirror and is reflected back to the interferometer and is received by the detector to form a measured light; by adjusting the displacement adjustment platform, when the reference light and the measured light coincide, interference fringes are formed and displayed on the dynamic interference detection software system of the generative adversarial neural network and collected; the dynamic interference detection software system of the generative adversarial neural network comprises an inter-frame fitting algorithm module for extracting fringe background and fringe modulation; a generative adversarial network constructs a phase shift interference fringe generator for collecting single-frame fringes, generating interference fringes with a fixed phase shift, and obtaining the phase information of the incident light to be measured through a two-step phase shift phase recovery algorithm.

[0008] A dynamic interference detection method based on a generative adversarial neural network is performed by the above-mentioned dynamic interference detection device based on a generative adversarial neural network, comprising: step 1, inputting the collected single-frame first interference fringe into the first generative adversarial network to construct a phase-shift interference fringe generator, and obtaining a The second interference fringe of the phase shift amount; step two, extracting the fringe background and fringe modulation under the measurement conditions through the inter-frame fitting algorithm module; step three, inputting the first interference fringe, the second interference fringe, the fringe background and the fringe modulation into the two-step phase shift phase recovery algorithm to obtain the phase information of the incident light to be measured.

[0009] Compared with the prior art, the present invention has the following beneficial effects:

[0010] (1) The interferometric measurement device does not require a phase shifting device, including but not limited to a mechanical phase shifting device, a wavelength phase shifting device, and a polarization phase shifting device, thereby ensuring that the device structure is simple and efficient.

[0011] (2) A phase-shift interference fringe generator is constructed using a generative adversarial network to generate virtual phase-shift interference fringes, which can achieve accurate prediction of the phase-shift fringes and avoid the phase shift error introduced by the phase shifter.

[0012] (3) It can realize dynamic interferometric measurement of single-frame fringes with high precision and strong anti-interference ability, avoiding the influence of time error and environmental error on the measurement results. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 is a diagram of a dynamic interference detection device based on a generative adversarial neural network according to an embodiment of the present invention;

[0014] Figure 2 is a flow chart of a dynamic interference detection method based on a generative adversarial neural network according to an embodiment of the present invention;

[0015] Figure 3 It is a schematic diagram of a phase-shift interference fringe generator constructed by a generative adversarial network according to an embodiment of the present invention, namely, an interference pattern generative adversarial network (IGA-Net) structure;

[0016] Figure 4 A single-frame interference fringe pattern is collected by interference detection software according to an embodiment of the present invention;

[0017] Figure 5 is a virtual phase-shift interference fringe pattern generated according to an embodiment of the present invention, having The second interference fringe of the phase shift amount;

[0018] Figure 6 It is the final surface shape measurement result in the embodiment of the present invention.

[0019] The meanings of the reference numerals in the figure are: 1 is an interferometer; 2 is a reference mirror; 3 is a measured mirror; 4 is a displacement adjustment table; 5 is a dynamic interference detection software system that generates adversarial neural networks. DETAILED DESCRIPTION

[0020] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0021] like Figure 1As shown, a dynamic interference detection device based on a generative adversarial neural network includes an interferometer 1, a reference mirror 2, a measured mirror 3, a displacement adjustment stage 4, and a dynamic interference detection software system 5 of a generative adversarial neural network; after a fixed wavelength incident light emitted by a laser in the interferometer 1 passes through the reference mirror 2, a portion of the light returns to the interferometer 1 and is received by the detector to form a reference light, and a portion of the light passes through the reference mirror to reach the surface of the measured mirror 3 and is reflected back to the interferometer 1 and is received by the detector to form a measured light. By adjusting the displacement adjustment stage 4, when the reference light and the measured light coincide, interference fringes are formed and displayed on the dynamic interference detection software system 5 of the generative adversarial neural network and collected. The dynamic interference detection software system 5 of the generative adversarial neural network includes an inter-frame fitting algorithm module for extracting the fringe background and the fringe modulation degree; a phase-shift interference fringe generator constructed by a generative adversarial network is used to collect single-frame fringes, generate interference fringes with a fixed phase shift, and obtain the phase information to be measured of the incident light through a two-step phase shift phase recovery algorithm. In one embodiment, the dynamic interference detection software system 5 of the generative adversarial neural network includes a phase-shift interference fringe generator constructed by a generative adversarial network.

[0022] Figure 3 It is a schematic diagram of a phase-shift interference fringe generator constructed by a generative adversarial network according to an embodiment of the present invention, namely, an interference pattern generative adversarial network (IGA-Net) structure. IGA-Net is divided into a generator and a discriminator, both of which adopt the U-Net (U-type network) architecture, which includes an encoder and a decoder, wherein Conv is a convolutional layer, CAB (Channel Attention Block) is a channel attention module, SAM (Supervised Attention Module) is a supervised attention module, Sigmoid is an activation function, GAPM (Global Average Pooling Module) is a global pooling module, and DL (Dense Layer) is a dense layer. This design allows the model to capture global and local features at the same time, thereby generating high-quality phase-shift interference patterns. In order to ensure the convergence of the generative adversarial network, the following modules are introduced into the network structure: 1. Gradient normalization: The introduction of gradient normalization effectively controls the amplitude of the gradient, prevents problems such as gradient explosion or disappearance, and improves the stability of training; 2. Spectral normalization: Spectral normalization is applied after each convolutional layer of the discriminator to limit the spectral norm of the weight matrix to 1; 3. Batch normalization: Batch normalization is used in the generator and discriminator to accelerate training convergence and improve the quality of generated results.

[0023] like Figure 2 As shown, the dynamic interference detection method based on the generative adversarial neural network, using the dynamic interference detection device based on the generative adversarial neural network, the steps of realizing single-frame dynamic interference measurement are as follows:

[0024] Step 1: Input the collected single-frame first interference fringe into the first generative adversarial network to construct a phase-shift interference fringe generator, and obtain The second interference fringe of the phase shift amount;

[0025] Step 2: extracting the fringe background and fringe modulation under the measurement conditions through an inter-frame fitting algorithm module;

[0026] Step 3: Input the first interference fringe, the second interference fringe, the fringe background and the fringe modulation into a two-step phase shift phase recovery algorithm to obtain the phase information of the incident light to be measured.

[0027] Example:

[0028] This example is based on a device for dynamic interference detection using a generative adversarial neural network. The laser in the interferometer 1 emits incident light with a fixed wavelength of 633nm. After passing through a reference mirror 2 with a diameter of 200mm, part of the light returns to the interferometer 1 and is received by the detector to form the reference light. Part of the light passes through the reference mirror 2 and reaches the surface of the measured mirror 3 with a diameter of 150mm. It is reflected back to the interferometer and is received by the detector to form the measured light. By adjusting the displacement adjustment stage 4, when the reference light and the measured light coincide, interference fringes are formed and displayed and collected on the dynamic interference detection software system 5 of the generative adversarial neural network. Figure 4 shown.

[0029] The dynamic interference detection method based on generative adversarial neural network has the following detection steps:

[0030] Step 1: Input the collected single-frame first interference fringe into the first generative adversarial network to construct a phase-shift interference fringe generator, and obtain The second interference fringe of the phase shift amount, such as Figure 5 As shown;

[0031] Step 2: extracting the fringe background and fringe modulation under the measurement conditions through an inter-frame fitting algorithm module;

[0032] Step 3: Input the first interference fringe, the second interference fringe, the fringe background and the fringe modulation into the two-step phase shift phase recovery algorithm to obtain the phase information to be measured. The root mean square RMS value is 2.6nm, and the peak-to-valley PV value is 54.0nm. Figure 6 shown.

[0033] In the description provided herein, a large number of specific details are described. However, it is understood that embodiments of the present invention can be practiced without these specific details. In some instances, well-known methods, structures and techniques are not shown in detail so as not to obscure the understanding of this description.

[0034] Although the present invention has been described according to a limited number of embodiments, it will be apparent to those skilled in the art, with the benefit of the above description, that other embodiments may be envisioned within the scope of the invention thus described. In addition, it should be noted that the language used in this specification is primarily selected for readability and instructional purposes, rather than for the purpose of explaining or limiting the subject matter of the present invention.

Claims

1. A dynamic interference detection device based on generative adversarial neural network, characterized in that: include: An interferometer (1), a reference mirror (2), a measured mirror (3), a displacement adjustment table (4), and a dynamic interference detection software system (5) for generating adversarial neural networks; a laser in the interferometer (1) emits incident light of a fixed wavelength, passes through the reference mirror (2), a portion of the light returns to the interferometer (1) and is received by a detector to form a reference light, and a portion of the light passes through the reference mirror to reach the measured mirror (3), is reflected from the surface and returns to the interferometer (1) and is received by the detector to form a measured light; by adjusting the displacement adjustment table (4), when the reference light and the measured light coincide, interference fringes are formed, which are displayed on the dynamic interference detection software system (5) for generating adversarial neural networks and collected; the dynamic interference detection software system (5) for generating adversarial neural networks includes an inter-frame fitting algorithm module for extracting fringe background and fringe modulation; a phase shift interference fringe generator constructed by a generation adversarial network is used to collect single-frame fringes, generate interference fringes with a fixed phase shift, and obtain the phase information of the incident light to be measured through a two-step phase shift phase recovery algorithm.

2. The dynamic interference detection device based on generative adversarial neural network according to claim 1 is characterized in that: The phase-shift interference fringe generator adopts a generative adversarial network structure, which consists of a generator and a discriminator. It captures global and local features at the same time to generate high-quality phase-shift interference patterns. At the same time, it uses gradient normalization, spectral normalization, and batch normalization to ensure the convergence of the network.

3. The dynamic interference detection device based on generative adversarial neural network according to claim 1 is characterized in that: The inter-frame fitting algorithm module uses the inter-frame fitting method to extract the fringe background and fringe modulation, and based on this, extracts the phase information through a two-step phase shifting phase recovery algorithm.

4. A dynamic interference detection method based on a generative adversarial neural network, performed by a dynamic interference detection device based on a generative adversarial neural network according to any one of claims 1-2, characterized in that: include: Step 1: Input the collected single-frame first interference fringe into the first generative adversarial network to construct a phase-shift interference fringe generator, and obtain The second interference fringe of the phase shift amount; Step 2: extracting the fringe background and fringe modulation under the measurement conditions through an inter-frame fitting algorithm module; Step 3: Input the first interference fringe, the second interference fringe, the fringe background and the fringe modulation into a two-step phase shift phase recovery algorithm to obtain the phase information of the incident light to be measured.

Images