A high-precision noise suppression and feature enhancement method suitable for strong non-periodic ultraviolet light signals

CN122817637APending Publication Date: 2026-09-25SONGYUAN POWER SUPPLY COMPANY OF STATE GRID JILINSHENG ELECTRIC POWER SUPPLY
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202610988913.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-07-03
Publication Date
2026-09-25

AI Technical Summary

Technical Problem

[0011]本发明进一步的目的在于克服现有方法中噪声识别不准、特征误削弱、参数依赖性强和复杂噪声适应能力不足等问题,提供一种更适用于强非周期、强非平稳紫外光信号的噪声抑制与特征增强技术方案,为后续紫外检测中的特征提取、状态识别、参数反演和故障诊断提供更高质量的数据基础

Benefits of technology

[0065]1.本发明针对复杂环境下紫外光信号中存在的设备噪声、环境噪声、局部突变噪声以及随机波动噪声,能够实现更准确的噪声提取与抑制。与现有采用ATM、HTM、STM及常规平滑方法的技术方案相比,本发明在模型信号和实测信号处理中均表现出更高的去噪精度,说明本发明能够更有效地提高复杂环境下紫外光信号的信噪比和可用性。

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122817637A_ABST
    Figure CN122817637A_ABST
Patent Text Reader

Abstract

The application discloses a high-precision noise suppression and feature enhancement method suitable for strong non-periodic ultraviolet light signals and relates to the technical fields of signal processing and ultraviolet spectrum analysis. The application comprises the following steps: S1: performing adaptive complete set empirical mode decomposition on an original spectrum signal to be processed to obtain a plurality of intrinsic mode functions; S2: identifying noise-dominant components from the first several intrinsic mode function components according to statistical characteristics; S3: constructing upper envelopes and lower envelopes of the identified noise-dominant intrinsic mode function components, respectively, generating a median envelope based on the upper and lower envelopes, performing smoothing processing on the median envelope, and extracting noise components in the intrinsic mode functions by combining the convergence and divergence relationship between the median envelopes before and after smoothing; and S4: superimposing the extracted noise components in the intrinsic mode functions to obtain total noise and deducting the total noise from the original spectrum signal.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention belongs to the field of signal processing and ultraviolet spectroscopy analysis technology, specifically a high-precision noise suppression and feature enhancement method suitable for strong aperiodic ultraviolet light signals. Background Technology

[0002] Currently, methods for noise suppression and feature enhancement of ultraviolet (UV) signals under complex environments include the following: 1. Using signal smoothing methods such as Savitzky-Golay smoothing and Rlowess smoothing to suppress random noise in UV spectral signals or UV-correlated optical signals. This type of method is simple to implement and widely used, typically reducing signal fluctuations through local fitting or sliding smoothing. 2. Using frequency domain methods such as Fast Fourier Transform (FFT), Discrete Cosine Transform (DCT), Butterworth filtering, Gauss-Hermite linear filtering, and maximum entropy nonlinear filtering to filter UV signals and reduce high-frequency noise components. This type of method mainly achieves noise reduction based on the idea of ​​frequency component separation. 3. Using adaptive decomposition methods such as Empirical Mode Decomposition (EMD), Ensemble Empirical Mode Decomposition (EEMD), and Complete Ensemble Empirical Mode Decomposition (CEEMDAN) to decompose UV signals into multiple Intrinsic Mode Functions (IMFs), and then reconstructing different mode components to achieve noise suppression and effective information extraction. 4. Hard thresholding (HTM), soft thresholding (STM), adaptive thresholding (ATM), and IEMDATD methods are used to perform threshold denoising on the IMF components obtained by CEEMDAN or correlation decomposition methods, thereby reconstructing the denoised signal. These methods represent implementations that are currently closest to this invention.

[0003] When using methods such as Savitzky-Golay smoothing and rlowess smoothing to denoise ultraviolet light signals, although random fluctuations can be reduced to some extent, the original signal is easily flattened as a whole, resulting in the weakening of local peaks, valleys, edge changes and weak feature information, which is not conducive to the preservation of fine features of ultraviolet light signals in complex environments.

[0004] When using frequency domain methods such as Fast Fourier Transform, Butterworth filtering, Gauss-Hermite linear filtering, and maximum entropy nonlinear filtering to process ultraviolet light signals, traditional frequency domain methods struggle to accurately separate the signal from noise due to the strong non-periodicity, nonlinearity, and complex spectral changes of such ultraviolet correlated optical signals. This is especially true when there are significant nonlinear changes and multiple interference extrema, resulting in poor denoising performance.

[0005] When using decomposition methods such as EMD, EEMD, and CEEMDAN, combined with threshold denoising methods such as HTM, STM, and ATM to process ultraviolet light signals, although a certain degree of mode decomposition and noise suppression can be achieved, existing methods are prone to problems such as inaccurate identification of noise features and real signal features, which leads to some noise not being effectively removed or the real signal features being mistakenly weakened, affecting the denoising accuracy and feature fidelity.

[0006] When using existing decomposition methods such as CEEMDAN and IEMDATD, it is usually necessary to set parameters such as the amplitude of auxiliary white noise and the number of noise sets. Different parameter selections will affect the decomposition results and denoising effect, thereby affecting the stability, repeatability and consistency of the algorithm in engineering applications.

[0007] When using the existing methods described above to process ultraviolet light signals in complex environments, there is still a problem of insufficient adaptability to different types of noise. The noise distribution patterns of different samples are not the same. Some noise is concentrated near the extreme values ​​of spectral lines, while others show downward jumps across the entire spectrum. Moreover, the noise distribution has peak and long tail characteristics, which indicates that the noise in complex environments is multi-source, non-uniform, and non-Gaussian. Existing uniform smoothing or uniform threshold methods are difficult to meet the noise suppression requirements of different regions and scales.

[0008] To address the aforementioned shortcomings, we propose a high-precision noise suppression and feature enhancement method suitable for strong aperiodic ultraviolet light signals, in order to solve the problems mentioned above.

[0009] The information disclosed above in this background section is only for enhancing the understanding of the background section of this invention, and therefore may include prior art that is not known to those skilled in the art. Summary of the Invention

[0010] The purpose of this invention is to provide a high-precision noise suppression and feature enhancement method suitable for strong aperiodic ultraviolet light signals. By performing adaptive decomposition, noise region identification and feature fidelity processing on the ultraviolet light signal, it can effectively suppress equipment noise, environmental noise and local abrupt noise, while preserving as much of the peak value, valley value, edge change and weak feature information in the original signal as possible, thereby improving the accuracy, stability and reliability of ultraviolet light signal processing in complex environments.

[0011] The further objective of this invention is to overcome the problems of inaccurate noise identification, feature attenuation, strong parameter dependence, and insufficient adaptability to complex noise in existing methods, and to provide a noise suppression and feature enhancement technology solution that is more suitable for strong non-periodic and strong non-stationary ultraviolet light signals, so as to provide a higher quality data foundation for feature extraction, state recognition, parameter inversion and fault diagnosis in subsequent ultraviolet detection.

[0012] To achieve the above objectives, the present invention provides the following technical solution: a high-precision noise suppression and feature enhancement method suitable for strong aperiodic ultraviolet light signals, comprising the following steps:

[0013] S1: Perform adaptive complete ensemble empirical mode decomposition on the original spectral signal to be processed to obtain several eigenmode functions;

[0014] S2: Identify the noise-dominated component from the first few eigenmode functions based on statistical characteristics;

[0015] S3: Construct upper and lower envelopes for the identified noise-dominated intrinsic mode function components of each order, and generate a median envelope based on the upper and lower envelopes; extract the noise components in each order intrinsic mode function by smoothing the median envelope and combining the convergence and divergence relationship between the median envelopes before and after smoothing.

[0016] S4: The noise components extracted from each order of intrinsic mode function are superimposed to obtain the total noise, and the total noise is subtracted from the original spectral signal to obtain the denoised target spectral signal.

[0017] Preferably, the step of identifying the noise-dominated intrinsic mode function components in S2 specifically includes:

[0018] Define the number of noise-dominant intrinsic mode functions. Its determination method satisfies the following formula:

[0019]

[0020] in,

[0021]

[0022]

[0023] In the formula, Denotes the k-th order eigenmode function; This represents the average value of the k-th eigenmode function over the entire wavelength range; This represents the estimated standard deviation of noise in the k-th eigenmode function; This represents the standard deviation of the k-th eigenmode function; Indicates the total number of sampling wavelength points; This represents the total number of intrinsic mode functions; This represents the number of noise-dominant intrinsic mode functions determined according to statistical criteria.

[0024] Preferably, in step S3, the noise components in each order intrinsic mode function are extracted, and calculations are performed based on the relationship that the noise energy attenuates with increasing order. Specifically:

[0025] Energy of noise components in the first-order eigenmode function Represented as:

[0026]

[0027] Energy of the noise component in the k-th order eigenmode function The attenuation is approximated by the following formula:

[0028]

[0029] in, This represents the energy of the noise component in the first-order eigenmode function. This represents the energy of the noise component in the k-th eigenmode function; this relationship reflects that as the order of the eigenmode function increases, the contribution of the noise component in each eigenmode function to the total noise gradually decreases.

[0030] Preferably, the method for determining the range of processing for each order of intrinsic mode function components in S3 is as follows:

[0031] Determine a maximum order ,Should This represents the highest order among all intrinsic mode functions that contain at least one distinctly flat segment with low-amplitude noise; only for all satisfying The intrinsic mode functions perform the operations of constructing the envelope and extracting noise.

[0032] Preferably, for all that satisfy The eigenmode functions are used to construct their local maximum envelopes. and local minimum envelope And further, the median envelope was obtained. :

[0033]

[0034] in, Let represent the upper envelope of the k-th eigenmode function; The lower envelope of the k-th eigenmode function is represented. Let represent the median envelope of the k-th eigenmode function;

[0035] Subsequently, based on the relationship between the upper and lower envelopes and the median envelope of the intrinsic mode functions, flat regions in the k-th order intrinsic mode functions are identified; the criteria for identifying flat regions are:

[0036]

[0037] in, A threshold is used to distinguish between flat noise regions and fluctuating signal regions; segments satisfying this inequality are considered low-amplitude flat noise regions; the boundary wavelength between adjacent regions is denoted as... ;Right now This indicates the boundary between the flat noise region and the fluctuating signal region.

[0038] Preferably, the step of extracting noise components in S3 by combining the convergence and divergence relationship of the median envelope before and after smoothing is specifically as follows:

[0039] A robust locally weighted regression smoothing method is used to smooth the median envelope. Smoothing is performed to obtain the smoothed median envelope. ;

[0040] Determine the boundary wavelengths of the convergence and divergence regions between the median envelope and the smooth median envelope. It satisfies:

[0041]

[0042] in, This indicates the boundary positions of the convergence and divergence regions between the median envelope and the smoothed median envelope;

[0043] Construct a continuous reference curve based on the boundary wavelength. : Take the region where the median envelope converges to the smooth median envelope Take within the region where the two diverge .

[0044] Preferably, after obtaining the continuous reference curve, the noise component corresponding to the k-th eigenmode function can be extracted:

[0045] For the identified flat region, we have:

[0046]

[0047] For non-flat undulating signal regions, let:

[0048]

[0049] in, This represents the noise component extracted from the k-th eigenmode function; This represents the continuous reference curve obtained from the reconstruction.

[0050] Preferably, in step S4, the noise components extracted from each order of intrinsic mode function are superimposed to obtain the total noise. Then, by subtracting the total noise from the original spectrum, the denoised target signal can be obtained, and its expression is:

[0051]

[0052]

[0053] In the formula, Indicates total noise; This represents the target spectrum after denoising. This represents the original input spectrum; when reconstructing the total noise using SMEDM, the following can be used: Alternative That is, the maximum order of the intrinsic mode function in the actual region containing obvious flat noise is used for reconstruction, so as to improve the adaptability to complex non-stationary signals.

[0054] This invention also provides a high-precision noise suppression and feature enhancement system suitable for strong aperiodic ultraviolet light signals, comprising:

[0055] The raw spectrum input module is used to acquire the spectral signal to be processed;

[0056] The CEEMDAN adaptive decomposition module is used to decompose the original signal into eigenmode function components at different scales.

[0057] The noise-dominated IMF discrimination module is used to determine the order range of the intrinsic mode functions involved in noise extraction based on statistical characteristics.

[0058] The median envelope construction module is used to generate the upper envelope, lower envelope, and median envelope of the target intrinsic mode function.

[0059] A smoothed median envelope generation module is used to smooth the median envelope to obtain a reference trend;

[0060] The noise component extraction module is used to separate and extract noise components from the intrinsic mode functions of each order based on the flat region identification results and the convergence / divergence relationship between the median envelope and the smooth median envelope.

[0061] The total noise reconstruction module is used to superimpose the noise components extracted from each order to generate the total noise;

[0062] The denoising result output module is used to subtract the total noise from the original spectral signal and output the final denoised spectral result.

[0063] Preferably, the noise-dominant IMF discrimination module determines the order of the highest intrinsic mode function containing a distinctly flat noise region based on statistical and waveform characteristics. And the total noise reconstruction module is based on this Multi-order noise superposition and reconstruction are performed to improve adaptability to complex non-stationary signals.

[0064] Compared with the prior art, the present invention has the following beneficial effects:

[0065] 1. This invention addresses the equipment noise, environmental noise, local abrupt change noise, and random fluctuation noise present in ultraviolet light signals under complex environments, achieving more accurate noise extraction and suppression. Compared with existing technical solutions employing ATM, HTM, STM, and conventional smoothing methods, this invention demonstrates higher denoising accuracy in both model and measured signal processing, indicating that it can more effectively improve the signal-to-noise ratio and usability of ultraviolet light signals under complex environments.

[0066] 2. Existing technologies often suffer from peak attenuation, valley passivation, slowed edge transitions, and weak feature submersion during denoising. This invention, however, does not uniformly smooth or threshold-shrink the entire signal. Instead, it performs differentiated processing on different regions within each modal component after decomposition. Therefore, it can effectively suppress noise while better preserving the peak shape, valley shape, local abrupt changes, edge variations, and weak feature information in the original ultraviolet light signal.

[0067] 3. Ultraviolet (UV) light signals, especially UV-correlated spectral signals, typically exhibit significant nonlinear, non-stationary, and strongly aperiodic characteristics. Traditional frequency domain methods and general time-frequency methods are insufficiently adapted to this type of signal, while existing thresholding methods are prone to noise and feature confusion. This invention, based on adaptive decomposition results and combined with local morphology recognition for noise extraction, is more suitable for processing UV light signals in complex environments.

[0068] 4. This invention is applicable not only to random high-frequency noise, but also to localized abnormal noise and irregular noise caused by detector switching, environmental disturbances, etc. Under different noise intensity conditions, this invention maintains good processing results, demonstrating its good stability and engineering applicability.

[0069] 5. The signal quality output by this invention is higher, which can provide a more accurate data foundation for feature extraction, target recognition, state assessment, parameter inversion and fault diagnosis in subsequent ultraviolet detection, and therefore has clear engineering application value.

[0070] 6. Compared with existing processing methods that require adjustment of multiple parameters, the technical route of this invention is clearer, the signal processing logic is more stable, the consistency of repeated processing is better, and it is easier to deploy and promote its application. Attached Figure Description

[0071] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0072] Figure 1 This is a flowchart of the method of the present invention;

[0073] Figure 2 This is a system architecture diagram of the present invention. Detailed Implementation

[0074] To enable those skilled in the art to better understand the technical solutions of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0075] Example 1:

[0076] Taking the corona discharge ultraviolet spectral signal collected by a certain type of ultraviolet spectrometer in the 200 nm~400 nm band as an example, the number of sampling wavelength points M = 2048.

[0077] Step S1: Adaptive Mode Decomposition

[0078] The raw spectral signal to be processed Adaptive Complete Empirical Mode Decomposition (CEEMDAN) is performed. The auxiliary noise standard deviation is set to 0.2 times the signal standard deviation, the ensemble averaging is set to 100, and the maximum number of iterations is set to 500. After decomposition, K intrinsic mode functions (IMFs) are obtained, denoted as imf1(λ), ​​imf2(λ), ..., imf_K(λ), along with a margin. In this embodiment, K is set to 11.

[0079] Step S2: Identify the noise-dominated IMF component

[0080] First, determine the number K of the dominant noise modes as follows: d :

[0081] Calculate the average value μ of each order IMF across the entire wavelength range. k and standard deviation And estimate the noise standard deviation. Its estimation formula can be derived from the median absolute deviation of the first-order IMF:

[0082]

[0083] For k>1, the reduction can be performed according to the noise energy attenuation relationship. Based on statistical criteria, the largest k value satisfying the following relationship is taken as K. d :

[0084] imf k mean μ k It approaches zero, and its autocorrelation function exhibits significant noise characteristics, while satisfying... and The ratio is within a preset range. In this embodiment, K is obtained through step-by-step determination. d =6, meaning the first 6 orders of IMF are considered to be the dominant noise components.

[0085] Additionally, determine the maximum order. : This represents the highest order among all IMFs containing at least one distinctly flat segment with low-amplitude noise. The IMFs are observed by iterating through them order by order. k The local morphology revealed that flat noise regions no longer exist in IMFs of order 7 and above, therefore... =7. Therefore, subsequent conditions only apply to conditions satisfying 1≤k≤ The IMF will process it.

[0086] Step S3: Noise component extraction

[0087] For each target IMF (k = 1, 2, ..., 7):

[0088] 3.1 Constructing the envelope and median envelope

[0089] Cubic spline interpolation was used to perform IMF on the following methods respectively. k Interpolation is performed on the local maxima and local minima, and the local maxima envelope is obtained. and local minimum envelope Calculate the median envelope: .

[0090] 3.2 Flat Region Recognition

[0091] Set the discrimination threshold In this embodiment, ε is taken as 0.05×std(imf k ).

[0092] According to the discriminant Identify low-amplitude flat noise regions. Mark continuous wavelength intervals that meet the conditions as flat regions, and denote the boundary wavelength between adjacent flat regions and undulating signal regions as... .

[0093] 3.3 Median envelope smoothing and convergence / divergence determination

[0094] Robust locally weighted regression smoothing (rlowess) is used for... Smoothing is performed, with the smoothing window width set to 10% of the total number of wavelength points, to obtain the smoothed median envelope. .

[0095] Compare and The differences are analyzed to detect the convergence and divergence regions of the two regions. The boundary wavelengths between the convergence and divergence regions are then identified. The judgment rule is: when and When the difference between two consecutive points changes from being less than the set threshold to being greater than the threshold, it is recorded as a boundary.

[0096] 3.4 Constructing a continuous reference curve and noise extraction

[0097] use Constructing a continuous reference curve Within the convergence region, let = Within the divergence region, let = .

[0098] Then extract the noise component of the k-th order IMF. :

[0099] If λ lies in a flat region, then ;

[0100] If λ is located in a region of non-flat undulating signals, then This means preserving the original signal characteristics and not extracting noise from this region.

[0101] Step S4: Total noise reconstruction and denoising signal acquisition

[0102] The noise components extracted from the first 7 IMFs are summed to obtain the total noise:

[0103] .

[0104] From the raw spectral signal After subtracting the total noise, the denoised target spectrum is obtained. .

[0105] After processing using the method described in this embodiment, the spectral baseline is smoother than the original spectrum, the characteristic peak shape is well preserved, and parameters such as peak position, peak height, and full width at half maximum (FWHM) are highly consistent with the pure signal. In tests with various types of simulated noise added, the signal-to-noise ratio of the denoised signal is improved by more than 12 dB, the root mean square error is reduced by more than 40% compared to the traditional CEEMDAN-soft thresholding method, and it also shows a reliable suppression effect on local anomalous jump noise.

[0106] Example 2:

[0107] This embodiment provides an implementation method for a system corresponding to the above method. The system includes:

[0108] The raw spectrum input module is used to acquire raw spectral digital signals collected by devices such as ultraviolet spectrometers and photodetectors.

[0109] The CEEMDAN adaptive decomposition module, using the parameter settings in Example 1, decomposes the signal into multi-order IMF components.

[0110] The noise-dominated IMF discrimination module calculates IMF statistics of various orders and provides K. d and .

[0111] The median envelope construction module calls the cubic spline interpolation program to generate the upper and lower envelopes and the median envelope.

[0112] The smoothed median envelope generation module performs rlowess smoothing to generate a smoothed median envelope.

[0113] The noise component extraction module generates noise components of various orders based on flat region identification and convergence / divergence analysis.

[0114] The total noise reconstruction module superimposes the effective order noise components according to wavelength. In particular, it can be based on... Replace K d Multi-order noise superposition and reconstruction are performed to improve adaptability to complex non-stationary signals.

[0115] The noise reduction output module performs the operation of subtracting the total noise from the original spectrum, outputs the final noise-reduced spectrum, and sends it to the display or storage unit.

[0116] The modules have a clear logical sequence and data transfer relationship. The raw input spectrum first enters the CEEMDAN decomposition module, where it is decomposed into multi-order intrinsic mode functions (IMFs). Subsequently, the noise-dominant IMF discrimination module determines the noise-dominant IMF based on statistical and waveform characteristics. or Next, the median envelope construction module generates upper, lower, and median envelopes for the target intrinsic mode functions; the smoothed median envelope generation module smooths the median envelope to obtain a reference trend; the noise extraction module separates noise from each order of intrinsic mode functions based on the flat region identification results and the convergence / divergence region boundaries; finally, the total noise reconstruction module superimposes the noise of each order, and the output module obtains the final denoised spectrum.

[0117] The mechanism of this invention lies in the fact that real spectral signals typically contain gradually varying trends, peak-valley structures, and local details, while noise is more often manifested as random high-frequency fluctuations superimposed on locally flat regions. Traditional thresholding methods mainly process based on amplitude, which can easily miss weak signal features. This invention, however, transforms the noise identification problem into a discrimination problem based on the geometric relationship of local envelopes and the smoothing trend by constructing the upper and lower envelopes and the median envelope of the intrinsic mode functions. This allows for a more effective distinction between random noise and real signal details. Especially in non-stationary, multi-peak, and multi-locally perturbed spectral signals, this method can better preserve the structural and characteristic peak information of the original spectrum, reducing feature distortion caused by direct smoothing or direct thresholding.

[0118] This invention first adaptively decomposes ultraviolet light signals collected under complex environments to obtain multiple intrinsic mode components (IMCs). Then, it identifies target mode components containing apparent noise features from the decomposition results. Further, by constructing upper, lower, and median envelopes for the target mode components and combining the smoothing results of the median envelope, it distinguishes between flat noise regions and undulating feature regions within the mode components. Noise components are extracted only within the identified noise regions, and multi-order noise components are reconstructed to finally obtain the denoised ultraviolet light signal. This technical approach differs from existing unified threshold denoising and overall smoothing denoising methods, achieving a balance between noise suppression and feature preservation.

[0119] In summary, this invention addresses the problem of noise and features intermingling in ultraviolet (UV) signals under complex environments. It achieves high-precision noise suppression and feature enhancement of UV signals through a technical approach of "adaptive decomposition + local envelope construction + noise region identification + regional noise extraction + signal reconstruction." This technical effect is directly derived from the invention's technical features and has clear, stable, and verifiable beneficial effects.

[0120] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A high-precision noise suppression and feature enhancement method suitable for strong aperiodic ultraviolet light signals, characterized in that, Includes the following steps: S1: Perform adaptive complete ensemble empirical mode decomposition on the original spectral signal to be processed to obtain several eigenmode functions; S2: Identify the noise-dominated component from the first few eigenmode functions based on statistical characteristics; S3: Construct upper and lower envelopes for the identified noise-dominated intrinsic mode function components of each order, and generate a median envelope based on the upper and lower envelopes; extract the noise components in each order intrinsic mode function by smoothing the median envelope and combining the convergence and divergence relationship between the median envelopes before and after smoothing. S4: The noise components extracted from each order of intrinsic mode function are superimposed to obtain the total noise, and the total noise is subtracted from the original spectral signal to obtain the denoised target spectral signal.

2. The high-precision noise suppression and feature enhancement method for strong aperiodic ultraviolet light signals according to claim 1, characterized in that: The specific steps in S2 for identifying noise-dominated intrinsic mode function components are as follows: Define the number of noise-dominant intrinsic mode functions. Its determination method satisfies the following formula: in, In the formula, Denotes the k-th order eigenmode function; This represents the average value of the k-th eigenmode function over the entire wavelength range; This represents the estimated standard deviation of noise in the k-th eigenmode function; This represents the standard deviation of the k-th eigenmode function; Indicates the total number of sampling wavelength points; This represents the total number of intrinsic mode functions; This represents the number of noise-dominant intrinsic mode functions determined according to statistical criteria.

3. The high-precision noise suppression and feature enhancement method for strong aperiodic ultraviolet light signals according to claim 1, characterized in that: In step S3, the noise components in each order intrinsic mode function are extracted, and calculations are performed based on the relationship that the noise energy attenuates with increasing order. Specifically: Energy of noise components in the first-order eigenmode function Represented as: Energy of the noise component in the k-th order eigenmode function The attenuation is approximated by the following formula: in, This represents the energy of the noise component in the first-order eigenmode function. This represents the energy of the noise component in the k-th eigenmode function; this relationship reflects that as the order of the eigenmode function increases, the contribution of the noise component in each eigenmode function to the total noise gradually decreases.

4. The high-precision noise suppression and feature enhancement method for strong aperiodic ultraviolet light signals according to claim 1, characterized in that: The method for determining the range of processing for each order of intrinsic mode function components in S3 is as follows: Determine a maximum order ,Should This represents the highest order among all intrinsic mode functions that contain at least one distinctly flat segment with low-amplitude noise; only for all satisfying The intrinsic mode functions perform the operations of constructing the envelope and extracting noise.

5. The high-precision noise suppression and feature enhancement method for strong aperiodic ultraviolet light signals according to claim 4, characterized in that: For all satisfying The eigenmode functions are used to construct their local maximum envelopes. and local minimum envelope And further, the median envelope was obtained. : in, The upper envelope of the k-th eigenmode function is represented. The lower envelope of the k-th eigenmode function is represented. Let represent the median envelope of the k-th eigenmode function; Subsequently, based on the relationship between the upper and lower envelopes and the median envelope of the intrinsic mode functions, flat regions in the k-th order intrinsic mode functions are identified; the criteria for identifying flat regions are: in, A threshold is used to distinguish between flat noise regions and fluctuating signal regions; segments satisfying this inequality are considered low-amplitude flat noise regions; the boundary wavelength between adjacent regions is denoted as... ;Right now This indicates the boundary between the flat noise region and the fluctuating signal region.

6. The high-precision noise suppression and feature enhancement method for strong aperiodic ultraviolet light signals according to claim 5, characterized in that: The specific steps in S3 for extracting noise components by combining the convergence and divergence relationship of the median envelope before and after smoothing are as follows: A robust locally weighted regression smoothing method is used to smooth the median envelope. Smoothing is performed to obtain the smoothed median envelope. ; Determine the boundary wavelengths of the convergence and divergence regions between the median envelope and the smooth median envelope. It satisfies: in, This indicates the boundary positions of the convergence and divergence regions between the median envelope and the smoothed median envelope; Construct a continuous reference curve based on the boundary wavelength. : Take the region where the median envelope converges to the smooth median envelope Take within the region where the two diverge .

7. The high-precision noise suppression and feature enhancement method for strong aperiodic ultraviolet light signals according to claim 6, characterized in that: After obtaining the continuous reference curve, the noise component corresponding to the k-th eigenmode function can be extracted: For the identified flat region, we have: For non-flat undulating signal regions, let: in, This represents the noise component extracted from the k-th eigenmode function; This represents the continuous reference curve obtained from the reconstruction.

8. The high-precision noise suppression and feature enhancement method for strong aperiodic ultraviolet light signals according to claim 1, characterized in that: In step S4, the noise components extracted from each eigenmode function are superimposed to obtain the total noise. Then, by subtracting the total noise from the original spectrum, the denoised target signal can be obtained, and its expression is: In the formula, Indicates total noise; This represents the target spectrum after denoising. This represents the original input spectrum; when reconstructing the total noise using SMEDM, the following can be used: Alternative That is, the maximum order of the intrinsic mode function in the actual region containing obvious flat noise is used for reconstruction, so as to improve the adaptability to complex non-stationary signals.

9. A high-precision noise suppression and feature enhancement system suitable for strong aperiodic ultraviolet light signals, characterized in that, include: The raw spectrum input module is used to acquire the spectral signal to be processed; The CEEMDAN adaptive decomposition module is used to decompose the original signal into eigenmode function components at different scales. The noise-dominated IMF discrimination module is used to determine the order range of the intrinsic mode functions involved in noise extraction based on statistical characteristics. The median envelope construction module is used to generate the upper envelope, lower envelope, and median envelope of the target intrinsic mode function. A smoothed median envelope generation module is used to smooth the median envelope to obtain a reference trend; The noise component extraction module is used to separate and extract noise components from the intrinsic mode functions of each order based on the flat region identification results and the convergence / divergence relationship between the median envelope and the smooth median envelope. The total noise reconstruction module is used to superimpose the noise components extracted from each order to generate the total noise; The denoising result output module is used to subtract the total noise from the original spectral signal and output the final denoised spectral result.

10. The system according to claim 9, characterized in that: The noise-dominant IMF discrimination module determines the highest intrinsic mode function order containing a distinctly flat noise region based on statistical and waveform characteristics. And the total noise reconstruction module is based on this Multi-order noise superposition and reconstruction are performed to improve adaptability to complex non-stationary signals.