Spectral reconstruction method, device, equipment and medium based on self-reference spectral constraints

Abstract

A spectrum reconstruction method, device, equipment and medium based on self-reference spectrum constraint, including obtaining a total transmission matrix obtained by calibration; obtaining each observation speckle generated after the light to be measured passes through each filtering interval of the spectrum filter and enters the scattering medium, all the observation speckles constitute the total observation speckle; the total intensity of the speckle based on each observation speckle constitutes a self-reference spectrum; constructing a spectrum reconstruction model based on a neural network and the physical forward process of the total transmission matrix, constructing a composite loss function based on the speckle intensity estimation error and the self-reference spectrum error, optimizing the neural network parameters by using the composite loss function until convergence, and obtaining an optimized neural network to realize spectrum reconstruction for a given total observation speckle. The present invention does not require a large-scale data set and a pre-training process of a neural network, and at the same time, it simply relies on the neural network to provide implicit priors, does not need to understand the prior information of the spectrum to be measured, and eliminates the tedious process of manually adjusting the regularization coefficient.

Description

Technical Field

[0001] The present invention mainly relates to the technical field of spectrum reconstruction, and in particular to a spectrum reconstruction method, device, equipment and medium based on self-reference spectrum constraints. Background Art

[0002] Spectroscopy is called "optical gene". Spectrometers are the cornerstone instruments of modern scientific research and industry, playing an important role in environmental monitoring, biochemistry, and laser characterization. However, traditional spectrometers using gratings, prisms or resonant cavities have some shortcomings and cannot meet the needs of special scenarios. These shortcomings include the contradiction between resolution and compactness, the upper limit of a single measurement bandwidth (free spectral range), and the long measurement time caused by scanning elements.

[0003] In recent years, speckle reconstruction spectrometers have attracted widespread attention due to their advantages of high resolution, miniaturization, freedom from free spectral range restrictions, and high-speed measurement. The basic principle of speckle reconstruction spectrometers is to inject a light beam into a disordered medium, and each wavelength will produce a unique light intensity distribution in the detector array as a spectral "fingerprint". This "frequency-space" mapping relationship based on linear scattering allows people to reconstruct the spectrum through various inverse problem solving algorithms when new speckles are observed.

[0004] Due to the diversity of spectral morphologies, traditional spectral reconstruction algorithms usually require prior information about the spectrum to be measured (such as sparsity), so that the algorithm parameters can be carefully adjusted manually, otherwise the accuracy of the spectral reconstruction structure will be reduced. In recent years, supervised learning models based on neural networks can directly learn this "frequency-space" mapping relationship from a large number of data pairs, and then infer the spectrum based on the observed speckle. However, this solution requires the collection of a large number of multi-type, high-quality training data sets; and purely data-driven networks have problems with the generalization of prediction results and weak physical interpretability. Summary of the invention

[0005] In view of the technical problems existing in the prior art, the present invention proposes a spectrum reconstruction method, device, equipment and medium based on self-reference spectrum constraints. The scheme reconstructs the spectrum based on the "frequency-space" mapping in the disordered scattering medium, and the reconstruction method is a combination of a neural network and a "frequency-space" mapping physical model.

[0006] To achieve the above purpose, the technical solution adopted by the present invention is as follows:

[0007] In one aspect, the present invention provides a speckle reconstruction spectrometer based on self-reference constraint and physical enhanced neural network, comprising:

[0008] Obtain the calibrated total transmission matrix;

[0009] Obtain the i-th observation speckle generated after the light to be measured passes through the i-th filtering interval of the spectral filter and enters the scattering medium, where i=1,2,…,K, K filtering intervals correspond to K observation speckles, and the K observation speckles constitute the total observation speckle;

[0010] The intensity of the i-th observed speckle is summed to obtain the total speckle intensity of the i-th observed speckle, and the total speckle intensity of K observed speckles constitutes a self-reference spectrum;

[0011] A spectral reconstruction model based on a neural network and a total transmission matrix physical forward process is constructed, a composite loss function is constructed based on a speckle intensity estimation error and a self-reference spectrum error, and the composite loss function is used to optimize the neural network parameters until convergence, thereby obtaining an optimized neural network for spectral reconstruction;

[0012] Based on the optimized neural network, spectral reconstruction is performed for a given total observed speckle.

[0013] Furthermore, the total transmission matrix is ​​obtained by the following calibration method:

[0014] Build a speckle-spectral calibration optical path, a high-resolution tunable narrow linewidth calibration light source, a spectral filter, a scattering medium, an objective lens, and a detector; the high-resolution tunable narrow linewidth calibration light source is connected to the spectral filter, and the spectral filter is connected to the scattering medium to generate speckles. The speckles output by the scattering medium are imaged to the detector through the objective lens. The typical tunable resolution of the narrow linewidth calibration light source is pm level, the 3dB linewidth of the narrow linewidth calibration light source is required to be less than its tuning resolution, and the typical filtering interval bandwidth of the spectral filter is nm level.

[0015] The high-resolution tunable narrow linewidth calibration light source performs high-resolution wavelength step scanning M times in the i-th filter interval in turn, and records the two-dimensional speckle image corresponding to each wavelength behind the scattering medium. The two-dimensional speckle image corresponding to each wavelength is straightened into an N×1 vector, and the M wavelengths are arranged in columns in turn to construct the narrowband transmission matrix corresponding to the i-th filter interval, where i=1,2,…,K;

[0016] The narrowband transmission matrices corresponding to the K filter intervals are concatenated diagonally to obtain a total transmission matrix T(K×N)×(K×M).

[0017] During measurement, the high-resolution tunable narrow-linewidth calibration light source in the above-mentioned speckle-spectrum calibration optical path is replaced with the light source to be measured, and the light source to be measured outputs the light to be measured, and the light to be measured passes through the i-th filtering interval of the spectral filter and enters the scattering medium to generate the i-th observation speckle, where i=1,2,…,K, and K filtering intervals correspond to K observation speckles, and the K observation speckles constitute the total observation speckle.

[0018] On the other hand, the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above-mentioned spectral reconstruction method based on self-reference spectral constraints when executing the computer program.

[0019] On the other hand, the present invention provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the above-mentioned spectrum reconstruction method based on self-reference spectrum constraints are implemented.

[0020] On the other hand, the present invention provides a computer program product, which is stored on a computer-readable storage medium and includes computer instructions, which, when executed by a processor, enable a computer device to implement the steps of the above-mentioned spectral reconstruction method based on self-reference spectral constraints.

[0021] Compared with the prior art, the present invention has the following beneficial effects:

[0022] Compared with the traditional spectral reconstruction model based on truncated singular value inversion, Lasso regularization and Tikhonov regularization, the method of the present invention does not require prior information of the spectrum to be measured, but simply relies on a neural network to provide implicit priors, while also eliminating the tedious process of manual parameter adjustment.

[0023] Compared with data-driven neural networks, the method of the present invention does not require pre-training of the network, eliminating the reliance on large-scale, multi-type data sets. Preparing and collecting such data sets in experiments often requires a lot of manpower and material resources, and has high requirements for the stability of the system and environment. In particular, it is very difficult to accurately generate and annotate multiple types of spectra. On the one hand, the maturity of devices with spectral programmable filtering performance is not high; on the other hand, marking the precise reference values ​​of the spectrum often requires high-precision measuring instruments, which greatly increases the cost of establishing the data set.

[0024] Compared with data-driven neural networks, the method of the present invention has higher spectral reconstruction generalization due to the physical model embedded in the spectral reconstruction process, and its reconstruction results have strong physical interpretability. In addition, due to the introduction of self-reference spectrum, the accuracy and credibility of the reconstruction results have been greatly improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the structures shown in these drawings without paying creative work.

[0026] Figure 1 is a flow chart of a spectrum reconstruction method based on self-reference spectrum constraints provided by an embodiment;

[0027] Figure 2 is a diagram of a speckle-spectrum calibration optical path structure provided in an embodiment;

[0028] Numbers in the figure: 1, narrow linewidth calibration light source, 2, spectral filter, 3, scattering medium, 4, objective lens, 5, detector. DETAILED DESCRIPTION

[0029] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0030] In one embodiment of the present invention, referring to Figure 1 The present invention provides a spectrum reconstruction method based on self-reference spectrum constraint, comprising:

[0031] Obtain the calibrated total transmission matrix;

[0032] Acquire each observation speckle generated after the light to be measured passes through each filtering interval of the spectral filter and enters the scattering medium, and all the observation speckles constitute the total observation speckle;

[0033] A self-reference spectrum is constructed based on the total intensity of each observed speckle;

[0034] A spectral reconstruction model based on a neural network and a total transmission matrix physical forward process is constructed, a composite loss function is constructed based on a speckle intensity estimation error and a self-reference spectrum error, and the composite loss function is used to optimize the neural network parameters until convergence, thereby obtaining an optimized neural network for spectral reconstruction;

[0035] Based on the optimized neural network, spectral reconstruction is performed for a given total observed speckle.

[0036] In a specific embodiment, it is proposed that the total transmission matrix can be obtained by the following calibration method:

[0037] Reference Figure 2 , Figure 2The speckle-spectral calibration optical path structure diagram provided in an embodiment includes a high-resolution tunable narrow linewidth calibration light source 1, a spectral filter 2, a scattering medium 3, an objective lens 4, and a detector 5; the high-resolution tunable narrow linewidth calibration light source 1 is connected to the spectral filter 2, and the spectral filter 2 is connected to the scattering medium 3 to generate speckles, and the scattering medium 3 outputs speckles and is imaged to the detector 5 through the objective lens 4. The i filter intervals of the spectral filter 2 are set in sequence, i=1,2,…,K. The typical tunable resolution of the narrow linewidth calibration light source is pm level, the 3dB linewidth of the narrow linewidth calibration light source is required to be less than its tuning resolution, and the typical filter interval bandwidth of the spectral filter is nm level.

[0038] The high-resolution tunable narrow linewidth calibration light source performs high-resolution wavelength step scanning M times in the i-th filter interval in turn, and records the speckle two-dimensional map corresponding to each wavelength behind the scattering medium. The speckle two-dimensional map corresponding to each wavelength is straightened into an N×1 vector, and the M wavelengths are arranged in columns in turn to construct the narrowband transmission matrix corresponding to the i-th filter interval. , where i=1,2,…,K;

[0039] ;

[0040] The narrowband transmission matrices corresponding to the K filter intervals are spliced ​​diagonally to obtain the total transmission matrix T(K×N)×(K×M):

[0041] ;

[0042] The measurement phase is used to obtain the observed speckles generated after the light to be measured passes through each filtering interval of the spectral filter and enters the scattering medium. All observed speckles constitute the total observed speckle. Specifically, the high-resolution tunable narrow linewidth calibration light source in the above-mentioned speckle-spectrum calibration optical path is replaced with the light source to be measured, and the speckle spectrum measurement optical path is obtained. The light source to be measured outputs the light to be measured, and the light to be measured enters the scattering medium after passing through the i-th filtering interval of the spectral filter, and the corresponding speckle is recorded after the scattering medium to constitute the i-th observed speckle. , where i=1,2,…,K, K filter intervals correspond to K observed speckles. K observed speckles Connect in order to form the total observed speckle .

[0043] For the i-th observed speckle The intensities of K speckles are summed to obtain the total intensities of K speckles, which constitute the self-reference spectrum SSR.

[0044] K observation speckles and K narrowband transmission matrices Corresponding to K segments of narrow-band spectrum to be measured with a length of M×1 . K segments of narrow-band spectrum to be measured with a length of M×1 Connect in order to get the total spectrum to be measured .

[0045] The relationship between the total observed speckle, the total transmission matrix and the total spectrum to be measured can be described by the following "frequency-space" mapping forward physical model, that is, the relationship model between the total observed speckle, the total transmission matrix and the total spectrum to be measured is constructed as follows:

[0046] ;

[0047] The total observed speckle , represents the i-th observed speckle;

[0048] Total transmission matrix ;

[0049] represents the narrowband transmission matrix corresponding to the i-th filtering interval, ;

[0050] Total spectrum to be measured , Represents the i-th narrow-band spectrum to be measured with a length of M×1.

[0051] The present invention constructs a spectral reconstruction model based on a neural network, the weights of the neural network are randomly initialized, and the neural network structure includes but is not limited to a convolutional neural network and a multi-layer perceptron. The measured total observed speckle is input into the neural network, and an estimate of the total spectrum to be measured is output. In the initial stage, the weights of the neural network are not optimized, so the output of the initial estimate of the total spectrum to be measured is not accurate. Based on the above-constructed relationship model between the total observed speckle, the total transmission matrix and the total spectrum to be measured, the total observed speckle estimate is further calculated. The present invention uses a composite loss function constructed by the speckle intensity estimation error and the self-reference spectrum error to update the weight of the neural network until the composite loss function converges to obtain an optimized spectral reconstruction neural network.

[0052] Wherein: the method for obtaining the speckle light intensity estimation error includes:

[0053] Based on the physical model between the total observed speckle, the total transmission matrix and the total measured spectrum , calculate the total observed speckle estimate corresponding to the total measured spectrum estimate output by the neural network. That is, substitute the total measured spectrum estimate output by the neural network as the total measured spectrum into the physical model to obtain the total observed speckle estimate corresponding to the total measured spectrum estimate.

[0054] The mean square error calculation is performed on the total observed speckle estimation and the total observed speckle to obtain the speckle intensity estimation error: .

[0055] The method for obtaining the self-reference spectrum error comprises:

[0056] The total observed speckle is input into the neural network, and the total spectrum to be measured is estimated as output;

[0057] The total spectrum to be measured is estimated and then downsampled to perform mean square error calculation with the self-reference spectrum to obtain the self-reference spectrum error .

[0058] The composite loss function is: , is the weight coefficient, which can be obtained by adjusting the generalized cross validation method.

[0059] The present invention can ensure that the network output results meet the constraints of the physical model by embedding the physical model of "frequency-space" mapping into the neural network. However, because there is still a gap between the model and the actual physical process, in order to improve the accuracy and credibility of spectral measurement, the actual spectral measured value should also be reflected in the loss function of the neural network. Existing commercial timing filter devices can usually provide wide-range, nm-level spectral filtering, and have good compatibility with most speckle reconstruction spectral measurement systems. Adding low-cost commercial timing filter devices can conveniently provide a rough but highly accurate self-reference value, thereby correcting possible distorted reconstructed spectral results.

[0060] In another embodiment, a spectrum reconstruction device based on self-reference spectrum constraint is provided, comprising:

[0061] A calibration module, used for obtaining a calibrated total transmission matrix;

[0062] A measurement module is used to obtain the i-th observation speckle generated after the light to be measured passes through the i-th filtering interval of the spectral filter and enters the scattering medium, where i=1,2,…,K, K filtering intervals correspond to K observation speckles, and the K observation speckles constitute the total observation speckle;

[0063] A self-reference spectrum calculation module is used to sum the intensity of the i-th observed speckle to obtain the total speckle intensity of the i-th observed speckle, and the total speckle intensity of K observed speckles constitutes a self-reference spectrum;

[0064] A network optimization module is used to reconstruct the total spectrum to be measured, construct a spectrum reconstruction model based on a neural network and a physical forward process of a total transmission matrix, construct a composite loss function based on a speckle intensity estimation error and a self-reference spectrum error, and optimize the neural network parameters using the composite loss function until convergence, thereby obtaining an optimized neural network for spectrum reconstruction;

[0065] The spectral reconstruction module realizes spectral reconstruction based on the optimized neural network and total observed speckle.

[0066] First, the total observed speckle is input into the neural network to obtain the total measured spectrum estimate. On the one hand, the total measured spectrum estimate itself is downsampled and then compared with the self-reference spectrum to perform a mean square error operation to obtain the self-reference spectrum error On the other hand, the total measured spectrum is estimated by the physical model After the operation, the total observed speckle estimation is obtained. The total observed speckle estimation and the total observed speckle are subjected to mean square error calculation to obtain the speckle intensity estimation error: A loss function is constructed based on the speckle intensity estimation error and the self-reference spectrum error, and the loss function is used to guide the optimization of the internal parameters of the neural network until convergence, and an optimized neural network for spectral reconstruction is obtained;

[0067] Similarly, the calibration module includes a speckle-spectral calibration optical path, which includes a high-resolution tunable narrow linewidth calibration light source, a spectral filter, a scattering medium, an objective lens, and a detector; the high-resolution tunable narrow linewidth calibration light source is connected to the spectral filter, the spectral filter is connected to the scattering medium to generate speckles, and the scattering medium outputs speckles that are imaged to the detector through the objective lens;

[0068] The high-resolution tunable narrow linewidth calibration light source performs high-resolution wavelength step scanning M times in the i-th filter interval in turn, and records the two-dimensional speckle image corresponding to each wavelength behind the scattering medium. The two-dimensional speckle image corresponding to each wavelength is straightened into an N×1 vector, and the M wavelengths are arranged in columns in turn to construct the narrowband transmission matrix corresponding to the i-th filter interval, where i=1,2,…,K;

[0069] The narrowband transmission matrices corresponding to the K filter intervals are spliced ​​diagonally to obtain the total transmission matrix T (K×N)×(K×M) .

[0070] Similarly, the measurement module includes a speckle-spectrum measurement optical path, which is obtained by replacing the high-resolution tunable narrow-linewidth calibration light source in the speckle-spectrum calibration optical path with the light source to be measured, and the light source to be measured outputs the light to be measured.

[0071] In the network optimization module: the method for obtaining the speckle intensity estimation error includes: based on the physical model between the total observed speckle, the total transmission matrix and the total spectrum to be measured , calculate the total observed speckle estimate corresponding to the total measured spectrum estimate output by the neural network; perform mean square error calculation on the total observed speckle estimate and the total observed speckle to obtain the speckle intensity estimation error The method for obtaining the self-reference spectrum error comprises: inputting the total observed speckle into a neural network, and outputting an estimate of the total spectrum to be measured; performing a mean square error operation on the total spectrum estimate to be measured and the self-reference spectrum after downsampling, and obtaining the self-reference spectrum error. .

[0072] The constructed composite loss function is: , is the weight coefficient;

[0073] On the other hand, the present invention provides a computer device, including a memory and a processor, the memory storing a computer program, and the processor implementing the steps of the spectral reconstruction method based on self-reference spectral constraints provided in any of the above embodiments when executing the computer program. The computer device can be a server. The computer device includes a processor, a memory, a network interface and a database connected via a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store sample data. The network interface of the computer device is used to communicate with an external terminal via a network connection.

[0074] On the other hand, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the spectral reconstruction method based on self-reference spectral constraints provided in any of the above embodiments.

[0075] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0076] Matters not covered by the present invention are known technologies.

[0077] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0078] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention. It should be pointed out that, for a person of ordinary skill in the art, several modifications and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

[0079] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A spectrum reconstruction method based on self-reference spectrum constraints, characterized in that: include: Obtain the calibrated total transmission matrix; Obtain the i-th observation speckle generated after the light to be measured passes through the i-th filtering interval of the spectral filter and enters the scattering medium, where i=1,2,…,K, K filtering intervals correspond to K observation speckles, and the K observation speckles constitute the total observation speckle; The intensity of the i-th observed speckle is summed to obtain the total speckle intensity of the i-th observed speckle, and the total speckle intensity of K observed speckles constitutes a self-reference spectrum; A spectral reconstruction model based on a neural network and a total transmission matrix physical forward process is constructed, a composite loss function is constructed based on a speckle intensity estimation error and a self-reference spectrum error, and the composite loss function is used to optimize the neural network parameters until convergence, thereby obtaining an optimized neural network for spectral reconstruction. The method for obtaining the speckle intensity estimation error includes: based on a physical model between the total observed speckle, the total transmission matrix, and the total spectrum to be measured , calculate the total observed speckle estimate corresponding to the total measured spectrum estimate output by the neural network; perform mean square error calculation on the total observed speckle estimate and the total observed speckle to obtain the speckle intensity estimation error The method for obtaining the self-reference spectrum error comprises: inputting the total observed speckle into a neural network, and outputting an estimate of the total spectrum to be measured; performing a mean square error operation on the total spectrum estimate after downsampling and the self-reference spectrum to obtain the self-reference spectrum error ; Based on the optimized neural network, spectral reconstruction is performed for a given total observed speckle.

2. The spectrum reconstruction method based on self-reference spectrum constraint according to claim 1, characterized in that: The total transmission matrix is ​​obtained by the following calibration method: A high-resolution tunable narrow linewidth calibration light source is connected to a spectral filter, and a scattering medium is connected to the spectral filter to generate speckles. The typical tunable resolution of the narrow linewidth calibration light source is pm level, and the 3dB linewidth of the narrow linewidth calibration light source is required to be less than its tuning resolution. The typical filtering interval bandwidth of the spectral filter is nm level. The high-resolution tunable narrow linewidth calibration light source performs high-resolution wavelength step scanning M times in the i-th filter interval in turn, and records the two-dimensional speckle image corresponding to each wavelength behind the scattering medium. The two-dimensional speckle image corresponding to each wavelength is straightened into an N×1 vector, and the M wavelengths are arranged in columns in turn to construct the narrowband transmission matrix corresponding to the i-th filter interval, where i=1,2,…,K; The narrowband transmission matrices corresponding to the K filter intervals are spliced ​​diagonally to obtain the total transmission matrix T (K×N)×(K×M) .

3. The spectrum reconstruction method based on self-reference spectrum constraint according to claim 2, characterized in that: Based on the physical model between the total observed speckle, the total transmission matrix and the total measured spectrum , described by the following formula: The total observed speckle , represents the i-th observed speckle; Total transmission matrix , represents the narrowband transmission matrix corresponding to the i-th filtering interval; Total spectrum to be measured , Indicates that the length of the i-th segment is M ×1 narrow-band spectrum to be measured.

4. The spectrum reconstruction method based on self-reference spectrum constraint according to claim 1, 2 or 3, characterized in that: The composite loss function is: , is the weight coefficient.

5. A spectral reconstruction device based on self-reference spectral constraints, characterized in that: include: A calibration module, used for calibrating and obtaining a total transmission matrix; A measurement module is used to obtain the i-th observation speckle generated after the light to be measured passes through the i-th filtering interval of the spectral filter and enters the scattering medium, where i=1,2,…,K, K filtering intervals correspond to K observation speckles, and the K observation speckles constitute the total observation speckle; A self-reference spectrum calculation module is used to sum the intensity of the i-th observed speckle to obtain the total speckle intensity of the i-th observed speckle, and the total speckle intensity of K observed speckles constitutes a self-reference spectrum; The network optimization module is used to reconstruct the total spectrum to be measured, construct a spectrum reconstruction model based on the physical forward process of the neural network and the total transmission matrix, construct a composite loss function based on the speckle intensity estimation error and the self-reference spectrum error, and optimize the neural network parameters using the composite loss function until convergence to obtain an optimized neural network for spectrum reconstruction. The method for obtaining the speckle intensity estimation error includes: based on the physical model between the total observed speckle, the total transmission matrix and the total spectrum to be measured , calculate the total observed speckle estimate corresponding to the total measured spectrum estimate output by the neural network; perform mean square error calculation on the total observed speckle estimate and the total observed speckle to obtain the speckle intensity estimation error The method for obtaining the self-reference spectrum error comprises: inputting the total observed speckle into a neural network, and outputting an estimate of the total spectrum to be measured; performing a mean square error operation on the total spectrum estimate after downsampling and the self-reference spectrum to obtain the self-reference spectrum error ; The spectral reconstruction module realizes spectral reconstruction based on the optimized neural network and total observed speckle.

6. The spectral reconstruction device based on self-reference spectral constraints according to claim 5, characterized in that: The calibration module comprises a speckle-spectral calibration optical path, which comprises a high-resolution tunable narrow linewidth calibration light source, a spectral filter, a scattering medium, an objective lens, and a detector; the high-resolution tunable narrow linewidth calibration light source is connected to the spectral filter, which is then connected to the scattering medium to generate speckles, and the speckles output by the scattering medium are imaged to the detector through the objective lens; the typical tunable resolution of the narrow linewidth calibration light source is pm level, the 3dB linewidth of the narrow linewidth calibration light source is required to be less than its tuning resolution, and the typical filtering interval bandwidth of the spectral filter is nm level; The high-resolution tunable narrow linewidth calibration light source performs high-resolution wavelength step scanning M times in the i-th filter interval in turn, and records the two-dimensional speckle image corresponding to each wavelength behind the scattering medium. The two-dimensional speckle image corresponding to each wavelength is straightened into an N×1 vector, and the M wavelengths are arranged in columns in turn to construct the narrowband transmission matrix corresponding to the i-th filter interval, where i=1,2,…,K; The narrowband transmission matrices corresponding to the K filter intervals are spliced ​​diagonally to obtain the total transmission matrix T (K×N)×(K×M) .

7. The spectral reconstruction device based on self-reference spectral constraints according to claim 6, characterized in that: The measurement module comprises a speckle-spectrum measurement optical path, which is obtained by replacing a high-resolution tunable narrow-linewidth calibration light source in a speckle-spectrum calibration optical path with a light source to be measured, and the light source to be measured outputs light to be measured.

8. The spectral reconstruction device based on self-reference spectral constraints according to claim 6, characterized in that: The composite loss function is: , is the weight coefficient.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the spectrum reconstruction method based on self-reference spectrum constraints as claimed in claim 1 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the spectrum reconstruction method based on self-reference spectrum constraints as claimed in claim 1 are implemented.

Images