Fourier transform spectrometer structure design method and system based on deep learning

Abstract

The invention provides a Fourier transform spectrometer structure design method and system based on deep learning, and relates to the technical field of spectrometer design, and the method comprises the steps: achieving the parameterization of an interferometer optical structure through the decomposition of an optical path structure of an interferometer and the design of a rule; the method comprises the following steps: parameterizing actual design requirements and a plurality of common interferometer optical structures to realize training sample configuration required by a primary structure derivation model; deducing a new interferometer structure according to the trained structure deduction model; all interferometer structures are evaluated, and interferometers with better imaging quality are screened out; putting the interferometer with better imaging quality into a training sample, retraining and deducing the structure deduction model, and finally designing an interferometer optical structure meeting the requirement; the scheme can be quickly given according to the design requirement of the spectrograph, and the design cost is saved.

Description

Technical Field

[0001] The present disclosure relates to the technical field of spectrometer design, and in particular to a method and system for structural design of a Fourier transform spectrometer based on deep learning. Background Art

[0002] The statements in this section merely provide background information related to the present disclosure and do not necessarily constitute prior art.

[0003] Spectroscopy is a basic property of matter and an important means to study atomic energy levels, material structure, chemical composition, and determine the elemental composition of matter. Spectrometers have the advantages of fast analysis speed, high accuracy, and non-destructive testing. They are widely used in many fields such as chemistry, environment, materials, and life sciences. Spectrometers can be divided into filter-type spectrometers, diffraction grating spectrometers, and Fourier transform spectrometers. Compared with the other two spectrometers, Fourier transform spectrometers have the advantages of multi-channels, large radiation flux, and low stray light. With the emergence of fast Fourier transform algorithms and the rapid development of computers, Fourier transform spectrometers have gradually become the focus of current spectrometer research. Fourier transform spectrometers use interferometers to generate interference fringes and use the Fourier transform relationship between the interference pattern and the spectrum to obtain the spectral information of the target.

[0004] Fourier transform spectrometers can be divided into time-modulated spectrometers and space-modulated spectrometers. At present, the research on time-modulated spectrometers is relatively mature, with high spectral resolution, and many scholars are studying it. The time-modulated Fourier transform spectrometer uses the Michelson interferometer as the core component. The incident light beam is divided into two beams by the beam splitter: a reflected light beam and a transmitted light beam; the reflected light beam is reflected by a static mirror and transmitted by a beam splitter to reach the focusing mirror; the transmitted light beam is reflected by a moving mirror and reflected by a beam splitter to reach the focusing mirror; the two light beams are focused on the detector to form an image and show interference fringes, and then the spectral information is obtained using Fourier transform.

[0005] The existing schemes for improving the design of spectrometers under different requirements are mainly through changes in their optical structures, such as rotating interferometers, double-moving mirror interferometers, etc., but the above changes require a certain amount of design time. In addition, or improving the data design of the interferometer, improving the interferometer through data calculation, all of which are through calculating the position and tilt angle of each optical device. However, due to different requirements for the size, shape, resolution and measurement spectral range of the Fourier transform spectrometer in different test environments, it is more troublesome to calculate the position and tilt angle of each optical device separately, and the practicality and real-time performance are not high, and it is impossible to quickly design a suitable spectrometer optical path. Summary of the invention

[0006] In order to solve the above problems, the present disclosure proposes a Fourier change spectrometer structure design method and system based on deep learning, which realizes the parameterization of the interferometer optical structure by decomposing the optical path structure of the interferometer and designing rules, and derives a new interferometer structure based on the structure derivation model. All the derived interferometer structures are evaluated to screen out interferometers with better imaging quality. Based on the interferometers with better imaging quality, the structure derivation model is retrained and deduced, and finally the optimal interferometer design scheme that meets the requirements is obtained.

[0007] According to some embodiments, the present disclosure adopts the following technical solutions: The structural design method of Fourier transform spectrometer based on deep learning includes: Obtain the target spectrometer model and decompose the interferometer optical path structure in the target spectrometer; According to the design rules, the decomposed optical path structure is parameterized and the interferometer optical path structure is expressed in the form of a symbol sequence; Input the symbolic sequence expression of the interferometer optical path structure into the structure deduction model to obtain the interferometer optical path structure design scheme, select the feasible optical path structure design scheme by design evaluation index, and re-input the feasible optical path structure design scheme into the structure deduction model for deduction to obtain the final optimal optical path structure design scheme; Among them, the structure derivation model is a long short-term memory network model. During the training process, the interferometer size and shape, resolution, measurement spectral range and interferometer parameterization results are used as the input of the long short-term memory network. The loss functions are constructed with the difference between the deduced results and the actual results and the difference between the interferometer size and the required size as constraints, so that the long short-term memory network can gradually derive the interferometer optical path structure.

[0008] Furthermore, a target spectrometer is obtained, and an interferometer optical path structure in the target spectrometer is decomposed, including: decomposing the optical path structure of the interferometer, wherein the interferometer is composed of a beam splitter and a reflector. In the decomposition result, the beam splitter is represented by its tilt angle, and the reflector has two properties, namely, a tilt angle and an installation distance, and the installation distance will change due to the movement of the reflector, that is, the symbols of the reflector are arranged in the order of installation distance, tilt angle, and installation distance after movement.

[0009] Furthermore, the decomposed optical path structure is parameterized according to the design rules, and the interferometer optical path structure is expressed in the form of a symbol sequence, including: converting the interferometer optical path structure into a symbol sequence expression, converting the incident light beam into a transmitted light beam and a reflected light beam, and parameterizing the interferometer optical path structure according to the reflectors through which the two light beams pass. Based on the beam splitter, the reflectors through which the transmitted light beam passes are arranged on the left side of the beam splitter, and the arrangement order is arranged from right to left according to the order of the reflectors through which the transmitted light beam passes, and the reflectors through which the reflected light beam passes are arranged on the right side of the beam splitter, and the arrangement order is arranged from left to right according to the order of the reflectors through which the transmitted light beam passes.

[0010] Furthermore, the structure derivation model is a long short-term memory network model, the long short-term memory network model is trained, the long short-term memory network model is initialized, the initial unit state is set to zero, the input of the training sample is the interferometer size, resolution, measurement spectral range and interferometer parameterization result, the comparison result of the number of reflectors of the interferometer and the input is used as the output of the long short-term memory network, the interferometer parameterization results in the sample are arranged from small to large, and are sequentially put into the input and output of the long short-term memory network for training.

[0011] Furthermore, the loss function of the long short-term memory network is constructed. The loss function consists of two parts. The first part is the difference between the calculated optical structure result and the actual result. The second part is the difference between the calculated interferometer size and the required size. The construction method of the second part of the loss function is as follows: According to the required resolution, the maximum optical path difference is calculated, which is expressed as the reciprocal of the required resolution. The maximum optical path difference determines the moving distance of the moving mirror. According to the derived interferometer parameterization results, the position of the beam splitter tilt angle is first found, and the symbol sequence on the left side of the beam splitter tilt angle is arranged from right to left to form sequence 1, and the symbol sequence on the right side of the beam splitter tilt angle is arranged from left to right to form sequence 2, thereby reconstructing the two optical paths of the interferometer and obtaining the coordinates of each reflector.

[0012] Furthermore, the position of each reflector in sequence 1 and sequence 2 is calculated according to the coordinates of each reflector, and the total distance difference between the reflectors in the two sequences is set as the maximum optical path difference, and the derived interferometer size is calculated. Assume that the length and width of the calculated size are x max and y max , assuming the required dimensions are x need and y need， Therefore, the loss function of the second part is: .

[0013] According to some embodiments, the present disclosure adopts the following technical solutions: The Fourier transform spectrometer structure design system based on deep learning includes: A decomposition module is used to obtain the target spectrometer model and decompose the interferometer optical path structure in the target spectrometer; A parameterization module is used to parameterize the decomposed optical path structure according to the design rules and express the interferometer optical path structure in the form of a symbol sequence; A derivation module is used to input the symbol sequence expression of the interferometer optical path structure into the structure derivation model to obtain the interferometer optical path structure design scheme, design evaluation indicators to screen the feasible schemes of the optical path structure design, and use the feasible schemes of the optical path structure design to re-input the structure derivation model for derivation to obtain the final optimal optical path structure design scheme; Among them, the structure derivation model is a long short-term memory network model. During the training process, the interferometer size and shape, resolution, measurement spectral range, and interferometer parameterization results are used as inputs of the long short-term memory network. The difference between the deduced results and the actual results and the difference between the interferometer size and the required size are used as constraints to construct loss functions, so that the long short-term memory network can gradually derive the interferometer optical path structure.

[0014] According to some embodiments, the present disclosure adopts the following technical solutions: A computer program product includes a computer program, and when the computer program is executed by a processor, the method for designing a Fourier change spectrometer structure based on deep learning is implemented.

[0015] According to some embodiments, the present disclosure adopts the following technical solutions: A non-transitory computer-readable storage medium, wherein the non-transitory computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by a processor, the method for designing a Fourier transform spectrometer structure based on deep learning is implemented.

[0016] According to some embodiments, the present disclosure adopts the following technical solutions: An electronic device comprises: a processor, a memory and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory so that the electronic device executes the method for designing a Fourier change spectrometer structure based on deep learning.

[0017] Compared with the prior art, the present invention has the following beneficial effects: The deep learning-based structural design method of the Fourier change spectrometer disclosed in the present invention parameterizes the interferometer optical path structure in the spectrometer, and then uses the long short-term memory network to derive a variety of interferometer structures. Then, the formed interferometers are evaluated to select the appropriate interferometer, and the long short-term memory network is further trained and evaluated to finally achieve the optimal design of the spectrometer. By inputting the required size, shape, resolution and measurement spectral range of the spectrometer, a Fourier change spectrometer with suitable size and good imaging effect can be quickly designed by computer, saving a lot of time.

[0018] The deep learning-based Fourier change spectrometer structure design method disclosed in the present invention uses a parameterized method to express the interferometer structure in the form of a symbol sequence, so that the long short-term memory network can effectively recognize it; by modifying the loss function, the loss function consists of two parts, the first part is the difference between the calculated optical structure result and the actual result, and the second part is the difference between the calculated interferometer size and the required size, the interferometer size is taken into account in the long short-term memory network, so that it can derive an interferometer structure with a better size; through cyclic training, a better interferometer structure can be further derived on the basis of the interferometer structure with a better size. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The accompanying drawings constituting a part of the present disclosure are used to provide a further understanding of the present disclosure. The illustrative embodiments of the present disclosure and their descriptions are used to explain the present disclosure and do not constitute an improper limitation on the present disclosure.

[0020] Figure 1 The fully automatic design flow chart of the interferometer of the embodiment of the present disclosure; Figure 2 Michelson interferometer parameterization result of the embodiment of the present disclosure; Figure 3 This is a diagram of the interferometer structure derived from an embodiment of the present disclosure. DETAILED DESCRIPTION

[0021] The present disclosure is further described below in conjunction with the accompanying drawings and embodiments.

[0022] It should be noted that the following detailed descriptions are all illustrative and are intended to provide further explanation of the present disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present disclosure belongs.

[0023] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present disclosure. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0024] Example 1 In one embodiment of the present disclosure, a method for designing a Fourier transform spectrometer structure based on deep learning is provided. The method parameterizes the internal optical structure of the interferometer to identify the structure derivation model, and the structure derivation model is a long short-term memory network model. The long short-term memory network is used to derive a feasible solution, and finally a better interferometer solution for specific needs is screened out through evaluation indicators, and the long short-term memory network is retrained with the solution as a training sample to further derive a better interferometer, and finally the optimal spectrometer design solution is obtained. The method saves manpower and designs a spectrometer with better performance. The method steps include: Step 1: Obtain the target spectrometer model and decompose the interferometer optical path structure in the target spectrometer; Step 2: parameterize the decomposed optical path structure according to the design rules, and express the interferometer optical path structure in the form of a symbol sequence; Step 3: Input the symbolic sequence expression of the interferometer optical path structure into the structure deduction model to obtain the interferometer optical path structure design scheme, select the feasible optical path structure design scheme by design evaluation index, and re-input the feasible optical path structure design scheme into the structure deduction model for deduction to obtain the final optimal optical path structure design scheme; Among them, the structure derivation model is a long short-term memory network model. During the training process, the interferometer size and shape, resolution, measurement spectral range and interferometer parameterization results are used as the input of the long short-term memory network. The loss functions are constructed with the difference between the deduced results and the actual results and the difference between the interferometer size and the required size as constraints, so that the long short-term memory network can gradually derive the interferometer optical path structure.

[0025] As an embodiment, the present disclosure discloses a method for designing a Fourier transform spectrometer structure based on deep learning, wherein the design requirements of the Fourier transform spectrometer are as follows: the target size is less than 450 mm in length and width, and the resolution is 0.045 cm -1 The measured spectral range is 350~7500 cm -1 The overall implementation process of the Fourier transform spectrometer structure design method based on deep learning is as follows: Figure 1 The specific implementation process is as follows: Step 1: Obtain the target spectrometer model and decompose the interferometer optical path structure in the target spectrometer; Specifically, the optical path structure of the interferometer is decomposed, where the interferometer is composed of a beam splitter and a reflector. In the decomposition result, the beam splitter is represented by its tilt angle β. Since the reflector has two properties, the tilt angle α and the installation distance l, and the installation distance l may change due to the movement of the reflector, the symbol arrangement order of the reflector is the installation distance l1, the tilt angle α, and the installation distance l2 after the movement, and the reflector expression is l1αl2.

[0026] Step 2: parameterize the decomposed optical path structure according to the design rules, and express the interferometer optical path structure in the form of a symbol sequence; Specifically, the design rules parameterize the optical path structure of the interferometer. The result of parameterization is to convert the interferometer structure into a symbolic sequence expression. The interferometer can convert the incident light beam into a transmitted light beam and a reflected light beam, and the optical path structure of the interferometer is parameterized according to the reflectors through which the two light beams pass. Based on the beam splitter β, the reflectors (l1αl2) through which the transmitted light beam passes are arranged on the left side of β, and the arrangement order is arranged from right to left according to the order of the reflectors through which the transmitted light beam passes. The reflectors l1αl2 through which the reflected light beam passes are arranged on the right side of β, and the arrangement order is arranged from left to right according to the order of the reflectors through which the transmitted light beam passes.

[0027] As an example. Take the Michelson interferometer as an example. The interferometer consists of a beam splitter, a fixed reflector and a movable reflector. Assume that the beam splitter tilt angle is β; the fixed reflector tilt angle is α, and the installation distance is l1; the movable reflector tilt angle is α, the installation distance is l1, and the distance after movement is l2. The result of the Michelson interferometer parameterization is l1αl2βl1αl2. Since the fixed reflector does not move, one of the l2 symbols in the parameterization result is empty. The parameterization result of the Michelson interferometer is as follows: Figure 2 This method can parameterize all translational interferometers.

[0028] Step 3: Input the symbolic sequence expression of the interferometer optical path structure into the structure deduction model to obtain the interferometer optical path structure design scheme, select the feasible scheme of the optical path structure design by design evaluation index, and re-input the feasible scheme of the optical path structure design into the structure deduction model for deduction to obtain the final optimal optical path structure design scheme; Specifically, the structure derivation model is a long short-term memory network model. First, the long short-term memory network model is trained. During the training process, the interferometer size, shape, resolution, measurement spectrum range and interferometer parameterization results are used as inputs of the long short-term memory network. The difference between the derivation result and the actual result and the difference between the interferometer size and the required size are used as constraints to construct loss functions respectively, so that the long short-term memory network gradually derives the interferometer optical path structure. Specifically including: 1) Design training samples, that is, the interferometer size is 450mm and the resolution is 0.045 cm -1 The parameterization and common interferometer parameterization results are used as the input of the training samples; the common interferometer parameterization results are used as sample outputs. The interferometer parameterization results are used as both the input and output of the training samples, but in specific performance, the two are not interferometers with the same optical structure. The common interferometer parameterization results are specifically selected as: l1αl2βl1αl2, l1αl2l1αl2βl1αl2, l1αl2βl1αl2l1αl2, l1αl2l1αl2βl1αl2l1αl2, l1αl2l1αl2βl1αl2l1αl2.

[0029] 2) Construct a long short-term memory network model, and its specific output expression h i for:

[0030]

[0031] in, W and b is the weight matrix and bias required for training, [ ] represents connecting two vectors, σ is the sigmoid function, h i-1 For the i -1 output value of the LSTM network, x i is the input parameter, C i For the i The unit state of a neural network.

[0032] Among them, the unit state c i The expression is:

[0033] in, f i It is the forget gate of the neural network. The unit state is the element-wise multiplication of the forget gate by the previous unit state, and then the element-wise multiplication of the input gate by the current input state, and then the two products are added together.

[0034] 3) Initialize the LSTM network model, set the initial unit state c0 to zero, and use the interferometer size, resolution, measurement spectrum range and interferometer parameterization results in the training sample as the input of the LSTM network. x i , compare the number of interferometer mirrors and the input result (the number of interferometer mirrors and the input x i The same or more) as the output of the long short-term memory network h i The common interferometer parameterization results are arranged from few to many and put into the input and output of the long short-term memory network in sequence.

[0035] 4) The difference between the derived results and the actual results and the difference between the interferometer size and the required size are used as constraints to construct loss functions. The loss function consists of two parts. The first part is to calculate the difference between the derived results of the network model and the actual results. The second part is to calculate the difference between the derived interferometer size and the required size.

[0036] Furthermore, the loss function of the first part is constructed as follows:

[0037] Among them, | V | is the total number of interferometers derived by the LSTM network, y i =1 or 0, 1 when the LSTM network predicts correctly, otherwise 0. p i is the difference between the true structure and the derived structure.

[0038] Furthermore, the loss function construction method of the second part is as follows: First, according to the required resolution, the maximum optical path difference L can be calculated, which is expressed as the reciprocal of the required resolution, specifically 112 mm. The maximum optical path difference determines the moving distance of the moving mirror. According to the interferometer parameterization results predicted by the model, the position of the beam splitter tilt angle β is first found by computer, and the symbol sequence to the left of β is arranged from right to left to form sequence 1, and the symbol sequence to the right of β is arranged from left to right to form sequence 2, thereby reconstructing the two optical paths of the interferometer. The coordinate axis is set at the beam splitter position, and the coordinates of each reflector can be expressed as:

[0039]

[0040] in, ll1 or l2, which are the minimum installation distance of the reflector, the minimum installation distance is selected as 100 mm, and the movable distance of the reflector is 212 mm; n represents the number of times the beam has passed n Therefore, the position of each reflector in sequence 1 and sequence 2 can be calculated according to the above formula, and the total distance difference between the two sequence reflectors is set as the maximum optical path difference L, so as to calculate the derived interferometer size. Assume that the length and width of the calculated size are x max and y max , the required dimensions are 450mm in length and 450mm in width respectively. Therefore, the loss function for the second part is:

[0041] 5) Train the long short-term memory network model and update the parameters in the network through gradient descent. The update rules are as follows:

[0042] in, is the training parameter of the LSTM network ( W and b ), the subscripts new and old represent the parameters after the network update and before the update, respectively. is the learning rate, is the gradient of the loss function with respect to the parameters.

[0043] As an embodiment, a new interferometer structure is derived using a trained long short-term memory network. The design requirements and the parameterization results of the training samples are input into the trained long short-term memory network, so that the long short-term memory network gradually derives a new interferometer structure, and the derived interferometer structure is l1αl2l1αl2l1αl2βl1αl2l1αl2, l1αl2l1αl2βl1αl2l1αl2l1αl2, l1αl2l1αl2l1αl2βl1αl2l1αl2l1αl2, l1αl2l1αl2l1αl2l1αl2βl1αl2l1αl2l1αl2.

[0044] As an embodiment, the design evaluation indexes are used to select feasible solutions for optical path structure design, including: Zemax software is used to evaluate the derived interferometer structure and the interferometer structure of the training sample. The best interferometer is selected according to the root mean square radius of the interferometer imaging.

[0045] The RMS radius of all interferometer images is shown in the following table:

[0046] Among them, the root mean square radius is an indicator used to describe the beam size. The smaller the root mean square radius, the better the imaging quality. It can be found that the interferometer structure is l1αl2l1αl2l1αl2βl1αl2l1αl2l1αl2 with the smallest root mean square radius and the best imaging quality. The specific optical path diagram is as follows: Figure 3 shown.

[0047] Furthermore, the derived interferometer optical structure is used to retrain the LSTM network, so that the LSTM network is re-derived into a similar new interferometer structure based on the sample, and the new interferometer structure is re-evaluated. If the new interferometer structure has better imaging quality, the interferometer is used as a training sample and returns to the training process again, until the LSTM network is unable to form an updated interferometer structure due to size limitations, and the obtained interferometer is used as the final interferometer.

[0048] If the derived interferometer structure is l1αl2l1αl2l1αl2βl1αl2l1αl2l1αl2, it is put into the training sample of the long short-term memory network model, and the long short-term memory network is retrained. During the training process, it is found that due to size limitations, the long short-term memory network cannot form an updated interferometer structure. Therefore, the interferometer with the structure of l1αl2l1αl2l1αl2βl1αl2l1αl2l1αl2 is used as the final interferometer. Finally, the interferometer is equipped with the corresponding housing and auxiliary equipment to form a complete spectrometer structure, completing the design of the spectrometer.

[0049] Example 2 In one embodiment of the present disclosure, a Fourier transform spectrometer structure design system based on deep learning is provided, comprising: A decomposition module is used to obtain the target spectrometer model and decompose the interferometer optical path structure in the target spectrometer; A parameterization module is used to parameterize the decomposed optical path structure according to the design rules and express the interferometer optical path structure in the form of a symbol sequence; A derivation module is used to input the symbol sequence expression of the interferometer optical path structure into the structure derivation model to obtain the interferometer optical path structure design scheme, design evaluation indicators to screen the feasible schemes of the optical path structure design, and use the feasible schemes of the optical path structure design to re-input the structure derivation model for derivation to obtain the final optimal optical path structure design scheme; Among them, the structure derivation model is a long short-term memory network model. During the training process, the interferometer size and shape, resolution, measurement spectral range, and interferometer parameterization results are used as inputs of the long short-term memory network. The difference between the deduced results and the actual results and the difference between the interferometer size and the required size are used as constraints to construct loss functions, so that the long short-term memory network can gradually derive the interferometer optical path structure.

[0050] Example 3 A computer program product includes a computer program, and when the computer program is executed by a processor, the method for designing a Fourier change spectrometer structure based on deep learning is implemented.

[0051] Example 4 A non-transitory computer-readable storage medium, wherein the non-transitory computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by a processor, the method for designing a Fourier transform spectrometer structure based on deep learning is implemented.

[0052] Example 5 An electronic device comprises: a processor, a memory and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory so that the electronic device executes the method for designing a Fourier change spectrometer structure based on deep learning.

[0053] The present disclosure is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present disclosure. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0054] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0055] Although the above describes the specific implementation methods of the present disclosure in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present disclosure. Technical personnel in the relevant field should understand that on the basis of the technical solution of the present disclosure, various modifications or variations that can be made by those skilled in the art without creative work are still within the scope of protection of the present disclosure.

Claims

1. A Fourier transform spectrometer structure design method based on deep learning, characterized in that: include: Obtain the target spectrometer model and decompose the interferometer optical path structure in the target spectrometer; According to the design rules, the decomposed optical path structure is parameterized and the interferometer optical path structure is expressed in the form of a symbol sequence; The symbolic sequence expression of the interferometer optical path structure is input into the structure deduction model to obtain the interferometer optical path structure design scheme, the design evaluation index is used to screen the feasible optical path structure design scheme, and the feasible optical path structure design scheme is re-input into the structure deduction model for deduction to obtain the final optimal optical path structure design scheme; Among them, the structure derivation model is a long short-term memory network model. During the training process, the interferometer size and shape, resolution, measurement spectral range and interferometer parameterization results are used as the input of the long short-term memory network. The loss functions are constructed with the difference between the deduced results and the actual results and the difference between the interferometer size and the required size as constraints, so that the long short-term memory network can gradually derive the interferometer optical path structure.

2. The method for designing a Fourier transform spectrometer structure based on deep learning according to claim 1, characterized in that: Obtain a target spectrometer and decompose the interferometer optical path structure in the target spectrometer, including: decomposing the optical path structure of the interferometer, wherein the interferometer is composed of a beam splitter and a reflector. In the decomposition result, the beam splitter is represented by its tilt angle, and the reflector has two properties, namely, the tilt angle and the installation distance, and the installation distance will change due to the movement of the reflector, that is, the symbols of the reflector are arranged in the order of installation distance, tilt angle and installation distance after movement.

3. The method for designing a Fourier transform spectrometer structure based on deep learning according to claim 1, characterized in that: The decomposed optical path structure is parameterized according to the design rules, and the interferometer optical path structure is expressed in the form of a symbol sequence, including: converting the interferometer optical path structure into a symbol sequence expression, converting the incident light beam into a transmitted light beam and a reflected light beam, and parameterizing the interferometer optical path structure according to the reflectors through which the two light beams pass. Based on the beam splitter, the reflectors through which the transmitted light beam passes are arranged on the left side of the beam splitter, and the arrangement order is arranged from right to left according to the order of the reflectors through which the transmitted light beam passes, and the reflectors through which the reflected light beam passes are arranged on the right side of the beam splitter, and the arrangement order is arranged from left to right according to the order of the reflectors through which the transmitted light beam passes.

4. The method for designing a Fourier transform spectrometer structure based on deep learning according to claim 1, characterized in that: The structure derivation model is a long short-term memory network model. The long short-term memory network model is trained and initialized, and the initial unit state is set to zero. The input of the training sample is the interferometer size, resolution, measurement spectral range and interferometer parameterization result. The comparison result of the number of reflectors of the interferometer and the input is used as the output of the long short-term memory network. The interferometer parameterization results in the sample are arranged from few to many and are sequentially put into the input and output of the long short-term memory network for training.

5. The method for designing a Fourier transform spectrometer structure based on deep learning according to claim 1, characterized in that: Construct the loss function of the long short-term memory network. The loss function consists of two parts. The first part is the difference between the calculated optical structure result and the actual result. The second part is the difference between the calculated interferometer size and the required size. The construction method of the loss function of the second part is as follows: According to the required resolution, the maximum optical path difference is calculated, which is expressed as the reciprocal of the required resolution. The maximum optical path difference determines the moving distance of the moving mirror. According to the derived interferometer parameterization results, the position of the beam splitter tilt angle is first found, and the symbol sequence on the left side of the beam splitter tilt angle is arranged from right to left to form sequence 1, and the symbol sequence on the right side of the beam splitter tilt angle is arranged from left to right to form sequence 2, thereby reconstructing the two optical paths of the interferometer and obtaining the coordinates of each reflector.

6. The method for designing a Fourier transform spectrometer structure based on deep learning as claimed in claim 5, characterized in that: The position of each reflector in sequence 1 and sequence 2 is calculated according to the coordinates of each reflector, and the total distance difference between the reflectors in the two sequences is set as the maximum optical path difference. The derived interferometer size is calculated, assuming that the length and width of the calculated size are x max and y max , assuming the required dimensions are x need and y need， Therefore, the loss function of the second part is: 。 7. A Fourier transform spectrometer structure design system based on deep learning, characterized in that: include: A decomposition module is used to obtain the target spectrometer model and decompose the interferometer optical path structure in the target spectrometer; A parameterization module is used to parameterize the decomposed optical path structure according to the design rules and express the interferometer optical path structure in the form of a symbol sequence; A derivation module is used to input the symbol sequence expression of the interferometer optical path structure into the structure derivation model to obtain the interferometer optical path structure design scheme, design evaluation indicators to screen the feasible schemes of the optical path structure design, and use the feasible schemes of the optical path structure design to re-input the structure derivation model for derivation to obtain the final optimal optical path structure design scheme; Among them, the structure derivation model is a long short-term memory network model. During the training process, the interferometer size and shape, resolution, measurement spectral range, and interferometer parameterization results are used as inputs of the long short-term memory network. The difference between the deduced results and the actual results and the difference between the interferometer size and the required size are used as constraints to construct loss functions, so that the long short-term memory network can gradually derive the interferometer optical path structure.

8. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method for designing a Fourier transform spectrometer structure based on deep learning according to any one of claims 1 to 6 is implemented.

9. A non-transitory computer-readable storage medium, characterized in that: The non-transitory computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by the processor, the deep learning-based Fourier change spectrometer structure design method as described in any one of claims 1 to 6 is implemented.

10. An electronic device, characterized in that: include: A processor, a memory and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory so that the electronic device executes the Fourier change spectrometer structure design method based on deep learning as described in any one of claims 1-6.

Images