Calculation reconstruction spectrometer based on micro-nano deformable structure

Abstract

According to the calculation reconstruction spectrometer based on the micro-nano deformable structure, the topology structure of the metamaterial spectrum encoder is changed by adjusting the magnitude of the sampling voltage, and then the spectral response of the metamaterial spectrum encoder is changed; that is to say, compared with the prior art, the sampling voltage loaded by the spectrum encoder with the micro-nano deformable structure can be changed through the control range and the step length of the sampling voltage, so that the spectrum response corresponding to the spectrum encoder is indirectly changed; a basis is provided for the spectrum reconstruction model to obtain the multi-scale characteristics of the to-be-reconstructed signal under different spectral responses, and the precision and efficiency of spectrum reconstruction are improved; meanwhile, the measurement range and the resolution of the spectrograph can be adjusted more flexibly according to requirements, the minimum on-chip size of the spectrograph is reduced to the magnitude of 2.5 * 2.5 square microns, the on-chip size of the spectrograph is greatly reduced, miniaturization of the spectrograph is achieved under the condition that performance is not sacrificed, and ultra-definition hyper-spectral imaging can be achieved.

Description

Technical Field

[0001] The present invention belongs to the technical field of micro-spectral acquisition, and in particular relates to a computational reconstruction spectrometer based on a micro-nano deformable structure. Background Art

[0002] Spectrometers are the cornerstone of technological development in many disciplines and industries. Traditional spectrometers are usually bulky and contain mechanically movable parts, such as gratings in visible and near-infrared spectrometers, or Michelson interferometers in Fourier transform infrared spectrometers. Miniaturized spectrometers without mechanical moving parts are receiving widespread attention due to their low cost, compactness, and potential for field applications. By utilizing photodetectors with wavelength responses and arrays of additional optical elements, micro-computational reconstruction spectrometers with a wider operating wavelength range have been realized. These micro-computational reconstruction spectrometers consist of arrays of photodetectors with different spectral responses to capture different spectral components of the incident light. However, the resolution of such spectrometers is still limited by the number of photodetectors in the array, and due to their spatial integration, their size in the chip is much larger than their operating wavelength, which restricts the further development of the spectrometer in hyperspectral imaging. Summary of the invention

[0003] In order to overcome the difficulties of existing spectrometer miniaturization technology and break through the size limit of traditional spectrometers, the present invention provides a computational reconstruction spectrometer based on a micro-nano deformable structure, which can efficiently obtain high-precision reconstructed spectra.

[0004] A computational reconstruction spectrometer based on a micro-nano deformable structure, comprising an imaging lens, a spectrum encoder array, a dynamic control circuit, a photodetector, and a spectrum reconstruction model;

[0005] The dynamic control circuit is used to traverse the set voltage interval according to the set step length to obtain different sampling voltages; the dynamic control circuit sequentially loads each sampling voltage to each spectral encoder of the spectral encoder array, wherein the loaded sampling voltage is different, and the spectral response corresponding to each spectral encoder is different;

[0006] The optical signal to be reconstructed with unknown energy is focused onto the spectral encoder array through an imaging lens, and is sequentially coupled with the spectral responses of the spectral encoders under different voltages to obtain coupled optical signals under different sampling voltages.

[0007] The photodetector is used to convert each coupled light signal into a coupled image;

[0008] The coupled optical signal at each sampling voltage is input into a trained spectrum reconstruction model, and the spectrum reconstruction model outputs a reconstructed spectrum corresponding to the optical signal to be reconstructed.

[0009] Furthermore, the spectral reconstruction model is a ResNet101 network, and the loss function used when training the spectral reconstruction model is as follows:

[0010] L=MSE(f1,f2)+λ1GMSE(f1,f2)+λ2Wav(f1,f2)+λ3PERC(f1,f2)

[0011] Among them, MSE represents the mean square error between the true spectrum f1 and the reconstructed spectrum f2 output by the spectral reconstruction model; GMSE represents the root mean square error between the first-order gradient of the true spectrum f1 and the first-order gradient of the reconstructed spectrum f2 output by the spectral reconstruction model; Wav represents the wavelet decomposition error between the spectral characteristic peaks of the true spectrum f1 and the spectral characteristic peaks of the reconstructed spectrum f2 output by the spectral reconstruction model; PERC represents the nonlinear characteristic error between the true spectrum f1 and the reconstructed spectrum f2 output by the spectral reconstruction model, λ1 is the weight corresponding to the root mean square error, λ2 is the weight corresponding to the wavelet decomposition error, and λ3 is the weight corresponding to the nonlinear characteristic error.

[0012] Furthermore, the mean square error MSE is calculated as:

[0013]

[0014] Where n is the total number of wavelength data of the spectrum, f 1,i is the spectral intensity of the i-th wavelength in the real spectrum f1, f 2,i is the spectral intensity of the i-th wavelength in the reconstructed spectrum f2.

[0015] Furthermore, the calculation method of the root mean square error GMSE is:

[0016]

[0017] Where n is the total number of wavelength data of the spectrum, is the first-order gradient of the spectral intensity of the i-th wavelength in the real spectrum f1, is the first-order gradient of the spectral intensity of the i-th wavelength in the reconstructed spectrum f2.

[0018] Furthermore, the calculation method of the wavelet decomposition error Wav is:

[0019]

[0020] Where n is the total number of wavelength data of the spectrum, W g (f 1,i ) is the high-frequency subband of the Harr wavelet transform corresponding to the i-th wavelength in the real spectrum f1, W g (f 2,i) is the high frequency subband of the Harr wavelet transform corresponding to the i-th wavelength in the reconstructed spectrum f2.

[0021] Furthermore, the calculation method of the nonlinear characteristic error PERC is:

[0022]

[0023] Where n is the total number of wavelength data of the spectrum, UNet3(f 1,i ) is the nonlinear feature of the first three layers of the U-Net network output of the i-th wavelength in the real spectrum f1, UNet3(f 2,i ) is the nonlinear feature of the output of the first three layers of the U-Net network for the i-th wavelength in the reconstructed spectrum f2.

[0024] Further, the spectral encoder includes a top layer, a middle layer and a silicon substrate;

[0025] Among them, the top layer is a gold film containing a hollow two-dimensional pattern, the middle layer is a silicon dioxide support column, and the gold film and the silicon substrate serve as the positive electrode and negative electrode for loading the sampling voltage respectively.

[0026] Furthermore, the hollow two-dimensional pattern is four hollow arcs, wherein every two hollow arcs form a group, and the two groups of hollow arcs are centrally symmetrical, and a metamaterial shrapnel that can produce different deformations under different loading voltages is formed between the two arcs belonging to the same group.

[0027] Furthermore, different hollowed-out two-dimensional patterns on the spectral encoder correspond to different optimal spectral response bands. After the hollowed-out two-dimensional pattern of the spectral encoder is determined, the method for obtaining the optimal spectral response band of the spectral encoder under the currently determined hollowed-out two-dimensional image is as follows:

[0028] S1: respectively taking ultraviolet light, visible light, near infrared light, and infrared light as the current light to be measured to perform an average correlation coefficient acquisition operation, and obtaining the average correlation coefficients corresponding to ultraviolet light, visible light, near infrared light, and infrared light, wherein the steps of the average correlation coefficient acquisition operation are as follows:

[0029] S11: Acquire the light signal to be measured generated by the substance to be measured under the current light to be measured;

[0030] S12: the dynamic control circuit traverses the set voltage interval according to the set step length to obtain different sampling voltages; the dynamic control circuit sequentially loads each sampling voltage to each spectral encoder under the currently determined hollow two-dimensional image;

[0031] S13: The optical signal to be measured is focused onto the spectral encoder array through an imaging lens, and is coupled with different spectral responses in sequence to obtain coupled optical signals to be measured under different sampling voltages;

[0032] S14: obtaining a correlation coefficient between two coupled optical signals to be measured at any two different sampling voltages, and then obtaining an average correlation coefficient of the correlation coefficients corresponding to all different sampling voltage combinations;

[0033] S2: taking the band corresponding to the minimum average correlation coefficient as the optimal spectral response band of the spectral encoder under the currently determined hollowed-out two-dimensional image.

[0034] Furthermore, the dynamic control circuit includes a single-chip microcomputer and multiple voltage divider sub-circuits, wherein the number of the voltage divider sub-circuits is the same as the number of spectral encoders in the spectral encoder array, and the single-chip microcomputer controls each voltage divider sub-circuit to output different divided voltages as sampling voltages to the corresponding spectral encoders connected thereto.

[0035] Beneficial effects:

[0036] 1. The present invention provides a computational reconstruction spectrometer based on a micro-nano deformable structure, which changes the topological structure of a metamaterial spectral encoder by adjusting the sampling voltage, thereby changing the spectral response of the metamaterial spectral encoder; that is, compared with the prior art, the present invention can change the sampling voltage loaded by the micro-nano deformable structure spectral encoder by controlling the range and step size of the sampling voltage, thereby indirectly changing the spectral response corresponding to the spectral encoder, thereby providing a basis for high-precision spectral reconstruction; at the same time, the present invention can more flexibly adjust the measurement range and resolution of the spectrometer according to demand, reduce the minimum on-chip size of the spectrometer to the order of 2.5 microns, greatly reduce the on-chip size of the spectrometer, realize miniaturization of the spectrometer without sacrificing performance, and realize ultra-clear hyperspectral imaging.

[0037] 2. The present invention provides a computational reconstruction spectrometer based on micro-nano deformable structures and a spectral encoder array dynamic control circuit, which can control micro-nano deformable structures as small as 2.5×2.5 square microns, apply different bias voltages to adjacent spectral encoders, and use the "time-for-space" technology to achieve snapshot spectroscopy. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 A schematic diagram of the basic structure of a computational reconstruction spectrometer based on a micro-nano deformable structure provided by the present invention.

[0039] Figure 2 The present invention provides a spectral encoder based on a micro-nano deformable structure.

[0040] Figure 3 A schematic diagram of the deformation of the micro-nano deformable structure provided by the present invention.

[0041] Figure 4 A schematic diagram of the time domain control of the micro-nano deformable structure provided by the present invention.

[0042] Figure 5 A partial diagram of the dynamic control circuit provided by the present invention.

[0043] Figure 6 Comparison of single peak spectrum acquisition results of the spectrometer provided by the present invention.

[0044] Figure 7 Comparison of the collection results of the spectrometer provided by the present invention for the praseodymium-neodymium filter standard material. DETAILED DESCRIPTION

[0045] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0046] like Figure 1 As shown, a computational reconstruction spectrometer based on a micro-nano deformable structure includes an imaging lens 101, a spectrum encoder array 102, a dynamic control circuit 105, a photodetector 103, and a spectrum reconstruction model 104;

[0047] The dynamic control circuit is used to traverse the set voltage interval according to the set step length to obtain different sampling voltages; the dynamic control circuit sequentially loads each sampling voltage to each spectral encoder of the spectral encoder array, wherein the loaded sampling voltage is different, and the spectral response corresponding to each spectral encoder is different;

[0048] The optical signal to be reconstructed with unknown spectral energy is focused onto the spectral encoder array through an imaging lens, and coupled with different spectral responses in turn to obtain coupled optical signals under different sampling voltages;

[0049] The photodetector is used to convert each coupled light signal into a coupled image;

[0050] The coupled optical signal at each sampling voltage is input into a trained spectrum reconstruction model, and the spectrum reconstruction model outputs an energy reconstruction spectrum corresponding to the optical signal to be reconstructed.

[0051] It should be noted that the coupled optical signals at each sampling voltage are input into the spectrum reconstruction model in the form of multi-channel image data. Specifically, the number of channels of the image data is determined by the number of different sampling voltages output by the dynamic control circuit.

[0052] like Figure 2 As shown, the spectral encoder includes a top layer 201, a middle layer 202 and a silicon substrate 203; wherein the top layer is a gold film containing a hollow two-dimensional pattern, the middle layer is a silicon dioxide support column, and the gold film and the silicon substrate are respectively used as the positive electrode and the negative electrode for loading the sampling voltage, as shown in FIG. Figure 3As shown, based on this, the present invention can achieve the coupling of the incident light signal to be reconstructed and the spectral response of the micro-nano deformable spectral encoder by transmission. For example, the top gold film provided in this embodiment has a thickness of 60nm, the thickness of the middle silicon dioxide layer is 300nm, the thickness of the base layer is 300nm, and the period of a single spectral encoder is 2500nm.

[0053] Meanwhile, the hollow two-dimensional pattern is four hollow arcs, wherein every two hollow arcs form a group, and the two groups of hollow arcs are centrally symmetrical, and a metamaterial that can produce different deformations under different loading voltages is formed between the two arcs belonging to the same group. Figure 4 As shown, it is a schematic diagram of the deformation comparison of the metamaterial under different sampling voltages. The greater the loaded sampling voltage, the greater the deformation perpendicular to the gold film surface.

[0054] It should be noted that the present invention uses etching to obtain the two-dimensional pattern of the top gold film, and the middle silicon dioxide layer is hollowed out to reserve deformation space for downward deformation and serve as a local self-supporting column to prevent the gold film from collapsing.

[0055] Furthermore, the dynamic control circuit includes a single chip microcomputer and a plurality of voltage divider sub-circuits, wherein the number of the voltage divider sub-circuits is the same as the number of spectral encoders in the spectral encoder array, and the single chip microcomputer controls each voltage divider sub-circuit to output different voltage dividers as sampling voltages to the spectral encoders connected to them. Figure 5 As shown in FIG. 1 , a voltage divider circuit composed of a resistor and a triode is realized. At this time, the suspended section of the resistor R567 is connected to the gold film as the positive electrode, and the gold film as the negative electrode can be directly grounded. The suspended end of the resistor R579 is connected to the microcontroller. Under the control signal of the microcontroller, the conduction and shutdown of each triode are controlled to realize different VCC voltage division of the resistor R567, thereby changing the size of the sampling voltage loaded on the spectrum encoder.

[0056] It should be noted that the spectral encoder is glued to the PCB board with conductive silver glue, the top gold film positive electrode is set outside the arc area, and the bottom base negative electrode is set at the pin of the socket on the PCB board. The voltage divider sub-circuit directly applies voltage by pressing the probe against the top pattern layer, and the voltage divider sub-circuit and the bottom silicon substrate are indirectly applied voltage through the socket connection on the PCB board.

[0057] It should be noted that the single-chip microcomputer outputs multiple drive signals, and the specific number is composed of integrated micro-nano deformable spectral encoders. The power monitoring reset circuit of the single-chip microcomputer provides a reset signal. The single-chip microcomputer includes a 74HC595 for column direction control and a 74HC154 for row direction control, and each spectral encoder is independently controlled by positioning the rows and columns.

[0058] Furthermore, different hollowed-out two-dimensional patterns on the spectral encoder correspond to different optimal spectral response bands. After the hollowed-out two-dimensional pattern of the spectral encoder is determined, the method for obtaining the optimal spectral response band of the spectral encoder under the currently determined hollowed-out two-dimensional image is as follows:

[0059] S1: respectively taking ultraviolet light, visible light, near infrared light, and infrared light as the current light to be measured to perform an average correlation coefficient acquisition operation, and obtaining the average correlation coefficients corresponding to ultraviolet light, visible light, near infrared light, and infrared light, wherein the steps of the average correlation coefficient acquisition operation are as follows:

[0060] S11: Acquire the light signal to be measured generated by the substance to be measured under the current light to be measured;

[0061] S12: the dynamic control circuit traverses the set voltage interval according to the set step length to obtain different sampling voltages; the dynamic control circuit sequentially loads each sampling voltage to each spectral encoder under the currently determined hollow two-dimensional image;

[0062] S13: The optical signal to be measured is focused onto the spectral encoder array through an imaging lens, and is coupled with different spectral responses in sequence to obtain coupled optical signals to be measured under different sampling voltages;

[0063] S14: obtaining a correlation coefficient between two coupled optical signals to be measured at any two different sampling voltages, and then obtaining an average correlation coefficient of the correlation coefficients corresponding to all different sampling voltage combinations;

[0064] S2: taking the band corresponding to the minimum average correlation coefficient as the optimal spectral response band of the spectral encoder under the currently determined hollowed-out two-dimensional image.

[0065] It should be noted that the spectrum reconstruction model of the present invention is ResNet101, which is used to extract multi-scale features in multi-channel image data. For the spectrum reconstruction of the computational reconstruction spectrometer, the importance of low-frequency coupling signals is much higher than the high-frequency detail information, so the average pooling layer of ResNet101 is removed.

[0066] Furthermore, the present invention provides a loss function that integrates the spectral reconstruction error and the characteristic peak positioning error. Specifically, the loss function used when training the spectral reconstruction model is as follows:

[0067] L=MSE(f1,f2)+λ1GMSE(f1,f2)+λ2Wav(f1,f2)+λ3PERC(f1,f2)

[0068] Among them, MSE represents the mean square error between the true spectrum f1 and the reconstructed spectrum f2 output by the spectrum reconstruction model, which reflects the error of spectrum reconstruction at each wavelength point; GMSE represents the root mean square error between the first-order gradient of the true spectrum f1 and the first-order gradient of the reconstructed spectrum f2 output by the spectrum reconstruction model, which reflects the local reconstruction error of the spectrum wavelength point; Wav represents the wavelet decomposition error between the spectral characteristic peaks of the true spectrum f1 and the spectral characteristic peaks of the reconstructed spectrum f2 output by the spectrum reconstruction model; PERC represents the nonlinear characteristic error between the true spectrum f1 and the reconstructed spectrum f2 output by the spectrum reconstruction model. PERC also belongs to the perception error, λ1 is the weight corresponding to the root mean square error, λ2 is the weight corresponding to the wavelet decomposition error, and λ3 is the weight corresponding to the nonlinear characteristic error.

[0069] The calculation method of mean square error MSE is:

[0070]

[0071] Where n is the total number of wavelength data of the spectrum, f 1,i is the spectral energy intensity of the i-th wavelength in the real spectrum f1, f 2,i is the spectral energy intensity of the i-th wavelength in the energy reconstruction spectrum f2.

[0072] The calculation method of the root mean square error GMSE is:

[0073]

[0074] Where n is the total number of wavelength data of the spectrum, is the first-order gradient of the spectral intensity of the i-th wavelength in the real spectrum f1, is the first-order gradient of the spectral intensity of the i-th wavelength in the reconstructed spectrum f2.

[0075] The calculation method of wavelet decomposition error Wav is:

[0076]

[0077] Where n is the total number of wavelength data of the spectrum, W g (f 1,i ) is the high-frequency subband of the Harr wavelet transform corresponding to the i-th wavelength in the real spectrum f1, W g (f 2,i ) is the high frequency subband of the Harr wavelet transform corresponding to the i-th wavelength in the reconstructed spectrum f2.

[0078] The calculation method of nonlinear characteristic error PERC is:

[0079]

[0080] Where n is the total number of wavelength data of the spectrum, UNet3(f 1,i ) is the nonlinear feature of the first three layers of the U-Net network output of the i-th wavelength in the real spectrum f1, UNet3(f 2,i ) is the nonlinear feature of the output of the first three layers of the U-Net network for the i-th wavelength in the reconstructed spectrum f2.

[0081] It should be noted that the input of the spectrum reconstruction model of the present invention is the coupling signal of the unknown input light collected by the photodetector and the spectrum encoder based on the micro-nano deformable structure under different voltages, and these coupling signals are input into the spectrum reconstruction model in the form of multi-channel image data. Specifically, the number of channels of the image data is determined by the number of different voltages output by the dynamic control circuit.

[0082] Furthermore, the specific measurement method steps of the present invention are as follows:

[0083] Step 1: inputting incident light to the computational reconstruction spectrometer based on micro-nano deformable structure provided by the present invention;

[0084] Step 2: Turn on the control circuit and input different voltages;

[0085] Step 3: Collect image data under different voltages on the photodetector;

[0086] Step 4: Run the deep learning-based spectral reconstruction model provided by the present invention on a computing device to obtain a high-quality incident light spectrum.

[0087] Figure 6 and Figure 7 The acquisition results of the computational reconstruction spectrometer based on the micro-nano deformable structure provided by the present invention are demonstrated. Figure 6 It mainly shows the collection effect of a single spectral peak. Figure 7 The praseodymium-neodymium filter standard material recommended by the National Metrology Verification Regulations JJG1034-2008 "Spectrophotometer Standard Filter" is used as the measured material. Compared with the standard spectrum, the computational reconstruction spectrometer provided by the present invention can collect high-quality spectra while significantly reducing the size of the spectrometer to 2.5 microns.

[0088] In summary, the present invention relates to a micro-spectrometer based on a micro-nano deformable structure, which utilizes deformable micro-nano manufacturing technology, that is, by applying different bias sampling voltages on the top gold film and the bottom silicon substrate to achieve three-dimensional deformation of the micro-nano structure, so that the micro-nano metamaterial under different bias sampling voltages has different topological morphologies, indirectly realizing dynamic regulation of the incident light field, and combining with deep learning technology to reconstruct the incident light spectrum to achieve high-precision spectral detection; the present invention can break through the size limitation of tens of microns of the spatial integration of traditional computational reconstruction spectrometers, and further reduce the spatial size of the computational reconstruction spectrometer in the form of time-for-space, and combine with deep learning technology to achieve high-precision spectral detection at the scale of 2.5×2.5 square microns, which provides a completely different novel technical route for the miniaturization of high-performance spectrometers and even the field of hyperspectral imaging.

[0089] Of course, the present invention may have many other embodiments. Without departing from the spirit and essence of the present invention, those skilled in the art may certainly make various corresponding changes and modifications based on the present invention, but these corresponding changes and modifications should all fall within the scope of protection of the claims attached to the present invention.

Claims

1. A computational reconstruction spectrometer based on micro-nano deformable structures, characterized in that: Including imaging lens, spectral encoder array, dynamic control circuit, photodetector, spectral reconstruction model; The dynamic control circuit is used to traverse the set voltage interval according to the set step length to obtain different sampling voltages; the dynamic control circuit sequentially loads each sampling voltage to each spectral encoder of the spectral encoder array, wherein the loaded sampling voltage is different, and the spectral response corresponding to each spectral encoder is different; The optical signal to be reconstructed with unknown energy is focused onto the spectral encoder array through an imaging lens, and is sequentially coupled with the spectral responses of the spectral encoders under different voltages to obtain coupled optical signals under different sampling voltages. The photodetector is used to convert each coupled light signal into a coupled image; The coupled optical signal at each sampling voltage is input into a trained spectrum reconstruction model, and the spectrum reconstruction model outputs a reconstructed spectrum corresponding to the optical signal to be reconstructed.

2. A computational reconstruction spectrometer based on micro-nano deformable structure as claimed in claim 1, characterized in that: The spectral reconstruction model is a ResNet101 network, and the loss function used when training the spectral reconstruction model is as follows: L=MSE(f1,f2)+λ1GMSE(f1,f2)+λ2Wav(f1,f2)+λ3PERC(f1,f2) Among them, MSE represents the mean square error between the true spectrum f1 and the reconstructed spectrum f2 output by the spectral reconstruction model; GMSE represents the root mean square error between the first-order gradient of the true spectrum f1 and the first-order gradient of the reconstructed spectrum f2 output by the spectral reconstruction model; Wav represents the wavelet decomposition error between the spectral characteristic peaks of the true spectrum f1 and the spectral characteristic peaks of the reconstructed spectrum f2 output by the spectral reconstruction model; PERC represents the nonlinear characteristic error between the true spectrum f1 and the reconstructed spectrum f2 output by the spectral reconstruction model, λ1 is the weight corresponding to the root mean square error, λ2 is the weight corresponding to the wavelet decomposition error, and λ3 is the weight corresponding to the nonlinear characteristic error.

3. A computational reconstruction spectrometer based on micro-nano deformable structure as claimed in claim 2, characterized in that: The calculation method of mean square error MSE is: Where n is the total number of wavelength data of the spectrum, f 1,i is the spectral intensity of the i-th wavelength in the real spectrum f1, f 2,i is the spectral intensity of the i-th wavelength in the reconstructed spectrum f2.

4. The computational reconstruction spectrometer based on micro-nano deformable structure as claimed in claim 2, characterized in that: The calculation method of the root mean square error GMSE is: Where n is the total number of wavelength data of the spectrum, is the first-order gradient of the spectral intensity of the i-th wavelength in the real spectrum f1, is the first-order gradient of the spectral intensity of the i-th wavelength in the reconstructed spectrum f2.

5. The computational reconstruction spectrometer based on micro-nano deformable structure as claimed in claim 2, characterized in that: The calculation method of wavelet decomposition error Wav is: Where n is the total number of wavelength data of the spectrum, W g (f 1,i ) is the high-frequency subband of the Harr wavelet transform corresponding to the i-th wavelength in the real spectrum f1, W g (f 2,i ) is the high frequency subband of the Harr wavelet transform corresponding to the i-th wavelength in the reconstructed spectrum f2.

6. The computational reconstruction spectrometer based on micro-nano deformable structure according to claim 2, characterized in that: The calculation method of nonlinear characteristic error PERC is: Where n is the total number of wavelength data of the spectrum, UNet3(f 1,i ) is the nonlinear feature of the first three layers of the U-Net network output of the i-th wavelength in the real spectrum f1, UNet3(f 2,i ) is the nonlinear feature of the output of the first three layers of the U-Net network for the i-th wavelength in the reconstructed spectrum f2.

7. The computational reconstruction spectrometer based on micro-nano deformable structure according to claim 1, characterized in that: The spectral encoder includes a top layer, a middle layer, and a silicon substrate; Among them, the top layer is a gold film containing a hollow two-dimensional pattern, the middle layer is a silicon dioxide support column, and the gold film and the silicon substrate serve as the positive electrode and negative electrode for loading the sampling voltage respectively.

8. The computational reconstruction spectrometer based on micro-nano deformable structure according to claim 7, characterized in that: The hollow two-dimensional pattern is four hollow arcs, wherein every two hollow arcs form a group, and the two groups of hollow arcs are centrally symmetrical, and a metamaterial shrapnel that can produce different deformations under different loading voltages is formed between the two arcs belonging to the same group.

9. The computational reconstruction spectrometer based on micro-nano deformable structure according to claim 7, characterized in that: Different hollow two-dimensional patterns on the spectral encoder correspond to different optimal spectral response bands. After the hollow two-dimensional pattern of the spectral encoder is determined, the method for obtaining the optimal spectral response band of the spectral encoder under the currently determined hollow two-dimensional image is: S1: respectively taking ultraviolet light, visible light, near infrared light, and infrared light as the current light to be measured to perform an average correlation coefficient acquisition operation, and obtaining the average correlation coefficients corresponding to ultraviolet light, visible light, near infrared light, and infrared light, wherein the steps of the average correlation coefficient acquisition operation are as follows: S11: Acquire the light signal to be measured generated by the substance to be measured under the current light to be measured; S12: the dynamic control circuit traverses the set voltage interval according to the set step length to obtain different sampling voltages; the dynamic control circuit sequentially loads each sampling voltage to each spectral encoder under the currently determined hollow two-dimensional image; S13: The optical signal to be measured is focused onto the spectral encoder array through an imaging lens, and is coupled with different spectral responses in sequence to obtain coupled optical signals to be measured under different sampling voltages; S14: obtaining a correlation coefficient between two coupled optical signals to be measured at any two different sampling voltages, and then obtaining an average correlation coefficient of the correlation coefficients corresponding to all different sampling voltage combinations; S2: taking the band corresponding to the minimum average correlation coefficient as the optimal spectral response band of the spectral encoder under the currently determined hollowed-out two-dimensional image.

10. The computational reconstruction spectrometer based on micro-nano deformable structure according to claim 1, characterized in that: The dynamic control circuit includes a single-chip microcomputer and multiple voltage divider sub-circuits, wherein the number of the voltage divider sub-circuits is the same as the number of spectral encoders in the spectral encoder array, and the single-chip microcomputer controls each voltage divider sub-circuit to output different divided voltages as sampling voltages to the corresponding spectral encoders.

Images