Spectral super-resolution model training method, spectral super-resolution measurement method and device

Abstract

The invention discloses a spectral super-resolution model training method, a spectral super-resolution measurement method and a spectral super-resolution measurement device. The method comprises the following steps: acquiring multiple groups of low-resolution spectral data and corresponding high-resolution spectral data; constructing a spectral super-resolution model fusing the three-dimensional convolutional layer and the spectral physical constraint term; and training the spectral super-resolution model by taking each group of low-resolution spectral data as input and taking the group of high-resolution spectral data as a label as a training sample in sequence until an error between the high-resolution spectral data output by the spectral super-resolution model and the high-resolution spectral data as the label meets a preset threshold value. Through the spectrum super-resolution model, the efficiency of spectrum measurement can be improved in a disguised manner, and the accuracy of parameter inversion can be ensured.

Description

Technical Field

[0001] The present application relates to, but is not limited to, the field of spectral analysis technology, and in particular to a spectral super-resolution model training method, a spectral super-resolution measurement method and a device. Background Art

[0002] In the field of spectral measurement, especially in applications such as plasma diagnostics, environmental monitoring, and biomedical imaging, high-precision three-dimensional spectral data is crucial for analyzing key parameters such as material composition, temperature distribution, and electron density. However, existing spectral measurement technology still has significant technical bottlenecks, such as: Low measurement efficiency: Traditional 3D spectral scanning methods typically require dense sampling of the target area (e.g., a 100×100 spatial grid) to ensure the spatial resolution of the data. This dense sampling method causes a single measurement to take up to several hours, severely limiting its application in dynamic process monitoring or high-throughput experiments.

[0003] Insufficient physical consistency: Existing super-resolution methods do not take into account the physical laws of spectra, resulting in large errors in the inversion of physical parameters. Summary of the Invention

[0004] The following is a summary of the subject matter described in detail herein. This summary is not intended to limit the scope of the claims.

[0005] The present application provides a spectral super-resolution model training method, a spectral super-resolution measurement method and a device, which can improve the efficiency of spectral measurement in disguise and ensure the accuracy of parameter inversion.

[0006] An embodiment of the present application provides a spectral super-resolution model training method, comprising: collecting multiple sets of low-resolution spectral data and their corresponding high-resolution spectral data; constructing a spectral super-resolution model that integrates a three-dimensional convolutional layer and a spectral physical constraint term; and training the spectral super-resolution model using each set of low-resolution spectral data as input and the high-resolution spectral data of the set as labels as training samples until the error between the high-resolution spectral data output by the spectral super-resolution model and the high-resolution spectral data used as the label meets a preset threshold.

[0007] An embodiment of the present application also provides a spectral super-resolution measurement method, including: using a trained spectral super-resolution model to super-reconstruct the collected low-resolution spectral data and output high-resolution spectral data; performing a parameter inversion process based on the output high-resolution spectral data to obtain the three-dimensional spatial distribution of plasma parameters.

[0008] An embodiment of the present application also provides a data security analysis device, comprising: a memory and a processor, wherein the memory is used to store a program for spectral super-resolution measurement; the processor is used to read the program for spectral super-resolution measurement and execute the spectral super-resolution measurement method as described in any embodiment of the present application.

[0009] Compared with the related art, the embodiment of the present application provides a spectral super-resolution model training method, spectral super-resolution measurement method and device. The scheme collects multiple sets of low-resolution spectral data and their corresponding high-resolution spectral data, and constructs a spectral super-resolution model that integrates a three-dimensional convolution layer and a spectral physical constraint term. Then, each set of low-resolution spectral data is used as input, and the high-resolution spectral data of the set is used as a label as a training sample to train the spectral super-resolution model until the error between the high-resolution spectral data output by the spectral super-resolution model and the high-resolution spectral data as the label meets a preset threshold. In this way, a spectral super-resolution model that can convert low-resolution spectral data into high-resolution spectral data and considers physical constraints is obtained. Since the spectral super-resolution model can convert low-resolution spectral data into high-resolution spectral data, only low-resolution spectral data needs to be collected during spectral measurement sampling, and no sampling is required through dense sampling. Therefore, the efficiency of spectral measurement can be improved in disguise by using the model. In addition, since the model considers physical constraints in the process of converting low-resolution spectral data into high-resolution spectral data, the high-resolution spectral data obtained after conversion conforms to the spectral physical laws, which can reduce the inversion error of physical parameters.

[0010] Other features and advantages of the present application will be described in the following description, and in part will become apparent from the description, or will be understood by practicing the present application. Other advantages of the present application can be realized and obtained by the solutions described in the description and the drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] The accompanying drawings are used to provide an understanding of the technical solution of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the technical solution of the present application and do not constitute a limitation on the technical solution of the present application.

[0012] Figure 1 This is a flow chart of the spectral super-resolution model training method according to an embodiment of the present application; Figure 2 This is a flow chart of the spectral super-resolution measurement method according to an embodiment of the present application; Figure 3 Schematic diagram of a spectral super-resolution measurement device according to an embodiment of the present application. DETAILED DESCRIPTION

[0013] This application describes multiple embodiments, but this description is exemplary rather than restrictive, and it is obvious to those skilled in the art that there may be more embodiments and implementations within the scope of the embodiments described in this application. Although many possible feature combinations are shown in the drawings and discussed in the detailed description, many other combinations of the disclosed features are also possible. Unless specifically limited, any feature or element of any embodiment may be used in combination with any other feature or element in any other embodiment, or may replace any other feature or element in any other embodiment.

[0014] The present application includes and contemplates combinations of features and elements known to those of ordinary skill in the art. The embodiments, features, and elements disclosed in this application may also be combined with any conventional features or elements to form a unique inventive solution. Any features or elements of any embodiment may also be combined with features or elements from other inventive solutions to form another unique inventive solution. Therefore, it should be understood that any feature shown and / or discussed in this application may be implemented individually or in any appropriate combination. Therefore, except for the limitations made according to the appended claims and their equivalents, the embodiments are not subject to other limitations. In addition, various modifications and changes may be made within the scope of protection of the appended claims.

[0015] In addition, when describing representative embodiments, the specification may have presented the method and / or process as a specific sequence of steps. However, to the extent that the method or process does not rely on the specific order of the steps described herein, the method or process should not be limited to the steps in the specific order described. As will be understood by those skilled in the art, other orders of steps are also possible. Therefore, the specific order of the steps set forth in the specification should not be interpreted as a limitation to the claims. In addition, the claims for the method and / or process should not be limited to performing their steps in the order written, and those skilled in the art can readily understand that these orders can be changed and still remain within the spirit and scope of the embodiments of the present application.

[0016] An embodiment of the present application provides a spectral super-resolution model training method, such as Figure 1 As shown, the following steps may be included: Step S110: collecting multiple sets of low-resolution spectral data and their corresponding high-resolution spectral data; Step S120: constructing a spectral super-resolution model integrating a three-dimensional convolutional layer and a spectral physical constraint term; Step S130: sequentially train the spectral super-resolution model using each set of low-resolution spectral data as input and the high-resolution spectral data of the set as labels as training samples until the error between the high-resolution spectral data output by the spectral super-resolution model and the high-resolution spectral data used as labels meets a preset threshold.

[0017] Exemplarily, the low-resolution spectral data refers to data with a spatial resolution much greater than the Debye length (such as spectral data with a spatial resolution greater than 10 times the Debye length), and the high-resolution spectral data refers to data with a spatial resolution of the Debye length or slightly greater than the Debye length (such as spectral data with a spatial resolution between 1 times the Debye length and 2 times the Debye length).

[0018] The spectral super-resolution model training method of this embodiment collects multiple sets of low-resolution spectral data and their corresponding high-resolution spectral data, and constructs a spectral super-resolution model that integrates a three-dimensional convolutional layer and spectral physical constraints. Then, each set of low-resolution spectral data is used as input, and the high-resolution spectral data of the set is used as a label as a training sample to train the spectral super-resolution model until the error between the high-resolution spectral data output by the spectral super-resolution model and the high-resolution spectral data used as the label meets a preset threshold. In this way, a spectral super-resolution model that can convert low-resolution spectral data into high-resolution spectral data and considers physical constraints is obtained. Because the spectral super-resolution model can convert low-resolution spectral data into high-resolution spectral data, only low-resolution spectral data needs to be collected during spectral measurement sampling, without the need for sampling through dense sampling. Therefore, the efficiency of spectral measurement can be improved in a disguised manner. In addition, because the model considers physical constraints in the process of converting low-resolution spectral data into high-resolution spectral data, the high-resolution spectral data obtained after conversion conforms to the laws of spectral physics, which can reduce the error in physical parameter inversion.

[0019] In an exemplary embodiment, the spectral super-resolution model includes an input layer, a three-dimensional convolutional encoder, a spectral attention module, and a physical constraint decoder; The input layer is used to receive the low-resolution spectral data input into the spectral super-resolution model; The three-dimensional convolutional encoder is used to downsample the low-resolution spectral data and extract high-level features, wherein the high-level features include but are not limited to one or more of the following options: characteristic peak position, spectral line broadening, and peak height ratio; The spectral attention module is used to compress the high-level features; The physical constraint decoder is used to upsample the compressed high-level features, reconstruct them into high-resolution spectral data, and output them.

[0020] In this embodiment, the characteristic peak position can be used to correspond to the element type, the spectral line broadening can be used to reflect the electron density, and the peak height ratio can be used to correlate with the temperature.

[0021] In one example of this embodiment, the three-dimensional convolutional encoder performs downsampling using a convolution operation with a convolution kernel size of 3×3×5 and a stride of 1×1×2, and the number of channels is sequentially increased from 1 to 64 and then to 128 to enhance feature expression capability; The compression ratio of the spectral attention module is 8:1, and the activation function adopts SiLU; The physical constraint decoder performs upsampling using a transposed convolution operation with a transposed convolution kernel size of 3×3×5 and a step size of 2×2×2 until the error between the reconstructed high-resolution spectral data and the high-resolution spectral data meets a preset threshold.

[0022] Exemplarily, the size of the input layer can be B×1×H×W×C; where B represents the batch size, 1 represents only one spatial dimension (H×W for a plane array), H and W represent the spatial dimensions (in pixels), and C represents the number of low-resolution spectral channels.

[0023] For example, a three-dimensional convolutional encoder can use a three-dimensional convolution operation to simultaneously process information in the spatial dimension and the spectral dimension, which enables the network to automatically learn the complex coupling relationship between spatial position and spectral features and capture the spectral change pattern in the local spatial region.

[0024] For example, the spectral attention module can dynamically enhance the representation of important spectral peaks (such as the center of a plasma emission line) by learning weights, while suppressing noise or non-critical background signals, improving the model's ability to capture weak but important spectral features. By integrating the spectral attention module into the network of a spectral super-resolution model, the sensitivity to characteristic peaks can be enhanced.

[0025] For example, the physical constraint decoder is responsible for upsampling the features extracted by the encoder and reconstructing them into high-resolution output, using a transposed convolution kernel (or deconvolution) with a size of 3×3×5 and a step size of 2×2×2 to match the downsampling step size of the encoder, thereby achieving a significant improvement in spatial resolution (theoretically, a step size of 2 can increase the spatial resolution by a factor of 2, and combined with sparse sampling, the overall improvement can exceed 4 times).

[0026] Exemplarily, the output channel number of the physical constraint decoder may be 1, indicating that the final output is a high-resolution spectral channel (usually representing intensity).

[0027] The spectral super-resolution model training method of this embodiment designs a spectral super-resolution model including an input layer, a three-dimensional convolutional encoder, a spectral attention module and a physical constraint decoder. Through these designs, low-resolution spectral data can be converted into high-resolution spectral data, and the effects of "improving the model's ability to capture weak but important spectral features and enhancing the sensitivity of characteristic peaks" and "significantly improving resolution" can be achieved.

[0028] In an exemplary embodiment, the spectral super-resolution model further includes a residual block group, wherein the residual block group is located between the spectral attention module and the physical constraint decoder; The residual block group includes 6 residual units, each of the residual units includes two 3×3×3 three-dimensional convolutional layers, and spectral normalization is applied after each of the three-dimensional convolutional layers.

[0029] The spectral super-resolution model training method of this embodiment employs a spectral attention module and a physical constraint decoder, with six residual units (RMUs) placed between the RMU and the physical constraint decoder. Each RMU contains two 3×3×3 3D convolutional layer RMUs, and spectral normalization is applied after each 3D convolutional layer. Spectral normalization helps control model gradients, improves training stability, and may also provide a certain regularization effect to prevent overfitting, thereby improving model training efficiency and enhancing the performance and robustness of the trained model.

[0030] In an exemplary embodiment, the method further includes: optimizing the parameters of the spectral super-resolution model based on a preset loss function, wherein the loss function includes a data fidelity term, a Boltzmann distribution constraint term, a Stark broadening constraint term, and a spatial continuity constraint term.

[0031] In an example of this embodiment, the loss function is as follows: ; in, represents the value of the loss function, is the data fidelity item, represents the high-resolution spectral data collected experimentally, represents the high-resolution spectral data output by the model, N represents the total number of pixels involved in the calculation (i.e. the total number of spatial pixels obtained based on the spatial dimensions H and W of the input layer), represents the norm; is the Boltzmann distribution constraint, represents the central wavelength of the i-th wavelength channel (the channel in the three-dimensional convolutional encoder), represents the statistical weight of the upper energy level (determined by the atomic energy level structure), represents the probability of spontaneous emission transition (characterizing the intensity of the spectral line), represents the upper energy level energy, represents the Boltzmann constant, represents the plasma electron temperature; is the Stark broadening constraint term, represents the full width at half maximum of an emission line in the reconstructed spectrum, represents the broadening value calculated by the Stark broadening theory formula, represents the electron number density; It is a spatial continuity constraint term, which means the sum of the gradient changes of the reconstructed data in the spatial dimension. 、 、 Indicates weight, 0.05< <0.3, 0.01< <0.1,0.005< <0.02.

[0032] Among them, the above loss function N 、 Extracted from the low-resolution spectral data collected experimentally, 、 、 Obtained from publicly available databases such as NIST ASD.

[0033] In this embodiment: Data fidelity can be used to measure the difference between reconstructed data and real data (such as mean square error MSE); Boltzmann distribution constraint, which can force the reconstructed spectrum to conform to the Boltzmann distribution; The Stark broadening constraint term, based on the theoretical relationship between electron density and spectral line broadening, can improve the accuracy of electron density inversion; The spatial continuity constraint helps suppress high-frequency noise and spatial artifacts in the reconstruction results, making the results smoother and more consistent with the continuity of actual physical scenes.

[0034] In summary, the spectral super-resolution model training method of this embodiment optimizes the parameters of the spectral super-resolution model based on a loss function including a data fidelity term, a Boltzmann distribution constraint term, a Stark broadening constraint term, and a spatial continuity constraint term, forcing the model to learn a mapping that conforms to physical laws, meeting physical consistency guarantees, and making the obtained high-resolution spectral data conform to the plasma radiation laws. Furthermore, spectral measurements using the high-resolution spectral data output by the spectral super-resolution model are more consistent with physical laws, ensuring the accuracy of parameter inversion (if the strategy of reducing the sampling density or reducing the number of spectral channels is adopted to shorten the measurement time, key spectral feature information will be lost, such as spectral line broadening or weak characteristic peaks, and the lack of this information will directly affect the accuracy of subsequent parameter inversion).

[0035] In an exemplary embodiment, the low-resolution spectral data is collected based on a non-uniform sampling strategy, and collecting multiple sets of low-resolution spectral data may include: Sampling is performed based on a preset coarse sampling density to obtain coarse sampling data; Performing fast parameter inversion based on the coarse sampling data and calculating the gradient field; Determine whether the gradient value in the gradient field exceeds a preset gradient threshold; if the gradient value is greater than the preset gradient threshold, increase the local sampling density in the area corresponding to the gradient value to obtain the low-resolution spectral data; if the gradient value is less than or equal to the preset gradient threshold, perform sampling based on the preset coarse sampling density to obtain the low-resolution spectral data.

[0036] The spectral super-resolution model training method of this embodiment uses a non-uniform sampling strategy to reduce the amount of sampled data, ensuring a positive correlation between the sampling density and the plasma parameter gradient. Compared to traditional fixed sampling methods, this sampling method achieves dynamic optimization through a closed-loop feedback mechanism, reducing the number of sampling points while ensuring high-density coverage of critical areas (such as gradient abrupt changes). Experimental results show that this spectral super-resolution model training method can reduce the number of sampling points by over 70%.

[0037] In an exemplary embodiment, collecting multiple sets of high-resolution spectral data may include: collecting high-resolution three-dimensional spectral data cubes as the multiple sets of high-resolution spectral data via a precision displacement platform.

[0038] Illustratively, the spectral super-resolution model of this embodiment can improve the spatial resolution by more than 4 times.

[0039] In summary, the spectral super-resolution model training method of this embodiment can achieve spatial resolution improvement (converting low-resolution spectral data into high-resolution spectral data) and spectral channel reconstruction.

[0040] Exemplarily, the promotion process and the reconstruction process may include: (1) Data preprocessing: Spectral normalization to eliminate instrument response differences; use a phase correlation algorithm to correct for displacement platform positioning errors to ensure accurate coordinate correspondence of each sampling point; annotate known emission line parameters in the spectrum to provide anchor points for physical constraints; (2) Three-dimensional feature extraction (spectral feature extraction): Slide the convolution kernel along the wavelength dimension to identify the characteristic peak position (corresponding to the element type), spectral line broadening (reflecting the electron density), and peak height ratio (related to temperature); (3) Spectral attention weighting: The model automatically enhances the signal in the key spectral line area and suppresses continuous background radiation (4) Physical constraint reconstruction, including at least one of the following: light intensity-temperature relationship constraint based on Boltzmann distribution, and spectral line shape constraint based on Stark broadening theory.

[0041] An embodiment of the present application provides a spectral super-resolution measurement method, such as Figure 2 As shown, the following steps may be included: Step S210: using the trained spectral super-resolution model to perform super-resolution reconstruction on the collected low-resolution spectral data and output high-resolution spectral data; Step S220: performing a parameter inversion process based on the output high-resolution spectral data to obtain the three-dimensional spatial distribution of plasma parameters.

[0042] The spectral super-resolution measurement method of this embodiment utilizes a trained spectral super-resolution model to super-reconstruct collected low-resolution spectral data, outputting high-resolution spectral data. This output is then used to perform parameter inversion to obtain the three-dimensional spatial distribution of plasma parameters. This approach allows spectral measurement sampling to be performed with only low-resolution spectral data, eliminating the need for dense sampling, significantly improving spectral measurement efficiency.

[0043] Through actual experiments, it was found that the solution of this embodiment can shorten the measurement time by more than 80% while maintaining accuracy; for example, in actual experiments, the measurement time can be shortened from 5 hours to 45 minutes.

[0044] In an exemplary embodiment, the parameter inversion process includes: The rolling Fourier ring correlation algorithm is introduced to quantify the local reconstruction error; The inversion results are corrected using the quantified error information.

[0045] The spectral super-resolution measurement method of this embodiment quantifies the local reconstruction error by introducing the rolling Fourier ring correlation algorithm, thereby introducing the spatial correlation learned by the spectral super-resolution model into the inversion, thereby improving the accuracy of parameter inversion.

[0046] An embodiment of the present application provides a data security analysis device, such as Figure 3 As shown, it includes: a memory and a processor, the memory is used to store a program for spectral super-resolution measurement; the processor is used to read the program for spectral super-resolution measurement and execute the spectral super-resolution measurement method as described in any embodiment of the present application.

[0047] In summary, this application implements a spectral super-resolution model training method, spectral super-resolution measurement method and device, which can be applied to but not limited to scenarios such as plasma monitoring and combustion diagnosis. This solution has the following beneficial effects: (1) Improved measurement efficiency: a. Sampling optimization: A non-uniform sampling strategy reduces data volume by 70%, and combined with motion control algorithms, it shortens scanning time. b. Hardware simplification: Single-channel designs (such as plasma modulators) replace multi-antenna arrays to reduce system complexity. c. Accuracy assurance principle: An information compensation mechanism, in which the model learns the intrinsic manifold structure of plasma radiation through training, can recover high-frequency details even with low-resolution input.

[0048] (2) Physical consistency guarantee: The reconstructed spectrum is made to conform to the plasma radiation law through the physical constraint loss function.

[0049] (3) Strong system compatibility: It can be adapted to various detection devices such as fiber optic spectrometers and CCD spectrometers.

[0050] Those skilled in the art will appreciate that all or some of the steps, systems, and functional modules / units in the methods, systems, and devices disclosed above may be implemented as software, firmware, hardware, or any combination thereof. In hardware implementations, the division between functional modules / units described above does not necessarily correspond to the division between physical components; for example, a physical component may have multiple functions, or a function or step may be performed collaboratively by several physical components. Some or all components may be implemented as software executed by a processor, such as a digital signal processor or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit. Such software may be distributed on computer-readable media, which may include computer storage media (or non-transitory media) and communication media (or transient media). As is well known to those skilled in the art, the term "computer storage media" encompasses volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information, such as computer-readable instructions, data structures, program modules, or other data. Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information and can be accessed by a computer. In addition, as is well known to those skilled in the art, communication media typically embodies computer-readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and may include any information delivery media.

[0051] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example" or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and features of different embodiments or examples without contradiction.

[0052] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations on the present application. Ordinary technicians in this field can change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. A spectral super-resolution model training method, characterized in that: include: Collecting multiple sets of low-resolution spectral data and their corresponding high-resolution spectral data; Construct a spectral super-resolution model that integrates three-dimensional convolutional layers and spectral physical constraints; The spectral super-resolution model is trained with each set of low-resolution spectral data as input and the high-resolution spectral data of the set as labels as training samples until the error between the high-resolution spectral data output by the spectral super-resolution model and the high-resolution spectral data used as labels meets a preset threshold.

2. The spectral super-resolution model training method according to claim 1, characterized in that The spectral super-resolution model includes an input layer, a three-dimensional convolutional encoder, a spectral attention module and a physical constraint decoder; The input layer is used to receive the low-resolution spectral data input into the spectral super-resolution model; The three-dimensional convolutional encoder is used to downsample the low-resolution spectral data and extract high-level features, wherein the high-level features include one or more of the following: characteristic peak position, spectral line broadening, and peak height ratio; The spectral attention module is used to compress the high-level features; The physical constraint decoder is used to upsample the compressed high-level features, reconstruct them into high-resolution spectral data, and output them.

3. The spectral super-resolution model training method according to claim 2, characterized in that: The 3D convolutional encoder uses a convolution operation with a kernel size of 3×3×5 and a stride of 1×1×2 for downsampling, and the number of channels is set to increase from 1 to 64 and then to 128; The compression ratio of the spectral attention module is 8:1, and the activation function adopts SiLU; The physical constraint decoder performs upsampling using a transposed convolution operation with a transposed convolution kernel size of 3×3×5 and a step size of 2×2×2 until the error between the reconstructed high-resolution spectral data and the high-resolution spectral data meets a preset threshold.

4. The spectral super-resolution model training method according to claim 2, characterized in that The spectral super-resolution model further includes a residual block group, wherein the residual block group is located between the spectral attention module and the physical constraint decoder; The residual block group includes 6 residual units, each of the residual units includes two 3×3×3 three-dimensional convolutional layers, and spectral normalization is applied after each of the three-dimensional convolutional layers.

5. The spectral super-resolution model training method according to claim 1, characterized in that: The method further comprises: The parameters of the spectral super-resolution model are optimized based on a preset loss function, wherein the loss function includes a data fidelity term, a Boltzmann distribution constraint term, a Stark broadening constraint term, and a spatial continuity constraint term.

6. The spectral super-resolution model training method according to claim 5, characterized in that: The loss function is as follows: ； in, represents the value of the loss function, is the data fidelity item, represents the high-resolution spectral data collected experimentally, represents the high-resolution spectral data output by the model, N Indicates the total number of pixels involved in the calculation, represents the norm; is the Boltzmann distribution constraint, represents the central wavelength of the i-th wavelength channel, represents the statistical weight of the upper energy level, represents the probability of spontaneous emission transition, represents the upper energy level energy, represents the Boltzmann constant, represents the plasma electron temperature; is the Stark broadening constraint term, represents the full width at half maximum of an emission line in the reconstructed spectrum, represents the broadening value calculated by the Stark broadening theory formula, represents the electron number density; It is a spatial continuity constraint term, which means the sum of the gradient changes of the reconstructed data in the spatial dimension. 、 、 Indicates weight, 0.05< <0.3, 0.01< <0.1,0.005< <0.

02.

7. The spectral super-resolution model training method according to claim 1, characterized in that: The low-resolution spectral data is collected based on a non-uniform sampling strategy, and the collecting of multiple sets of low-resolution spectral data includes: Sampling is performed based on a preset coarse sampling density to obtain coarse sampling data; Performing rapid parameter inversion based on the coarse sampling data and calculating the gradient field; Determine whether the gradient value in the gradient field exceeds a preset gradient threshold; if the gradient value is greater than the preset gradient threshold, increase the local sampling density in the area corresponding to the gradient value to obtain the low-resolution spectral data; if the gradient value is less than or equal to the preset gradient threshold, perform sampling based on the preset coarse sampling density to obtain the low-resolution spectral data.

8. A spectral super-resolution measurement method, characterized in that: include: The trained spectral super-resolution model is used to perform super-resolution reconstruction on the collected low-resolution spectral data and output high-resolution spectral data; The parameter inversion process is performed based on the output high-resolution spectral data to obtain the three-dimensional spatial distribution of plasma parameters.

9. The spectral super-resolution measurement method according to claim 8, characterized in that: The parameter inversion process includes: The rolling Fourier ring correlation algorithm is introduced to quantify the local reconstruction error; The inversion results are corrected using the quantified error information.

10. A spectral super-resolution measurement device, comprising: A memory and a processor, characterized in that: The memory is used to store a program for spectral super-resolution measurement; The processor is configured to read the program for spectral super-resolution measurement and execute the spectral super-resolution measurement method according to claim 8 or 9.

Images