A method and system for intelligent analysis of spectral sensor signals

By setting a reference spectral band and a local drift function in the spectral sensor, the problem of inconsistent light source drift was solved, and accurate compensation and analysis of the spectral signal were achieved.

CN123065987APending Publication Date: 2026-11-10HANGZHOU LUHENG BIOLOGICAL TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611283410.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-24
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Existing spectral sensors exhibit inconsistent spectral drift when the light source ages or temperature changes, leading to a mismatch between the target spectral band compensation and the actual local drift, making it difficult to reflect the change in light source drift with wavelength.

Method used

Reference spectral bands are set on both sides of the target spectral band. A local drift function is established through effective reference sampling points to compensate for the target spectral band. The recovery confidence is judged based on the fitting quality and continuity of the local drift function, and the analytical results are output.

Benefits of technology

It enables continuous processing of spectral signals, reduces the impact of local light source drift and low-confidence compensation data on the system's analytical results, and improves the accuracy and reliability of the spectral sensor.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123065987A_ABST
    Figure CN123065987A_ABST
Patent Text Reader

Abstract

This invention relates to the field of spectral detection and signal processing technology, and particularly to an intelligent analysis method and system for spectral sensor signals. The method includes: acquiring the current measured spectrum and a reference response spectrum; generating a local response difference sequence covering the target spectral band and a reference spectral band; establishing a local drift function spanning the target spectral band; obtaining an equivalent target spectral band; determining the recovery confidence level; and outputting the analysis result based on the equivalent target spectral band when the recovery confidence level meets a preset condition. This invention employs a technical solution that sets reference spectral bands on both sides of the target spectral band, establishes a local drift function spanning the target spectral band based on effective reference sampling points, and compensates for the target spectral band based on the local drift function. This solves the problems of difficulty in reflecting the change of light source drift with wavelength and the mismatch between target spectral band compensation and actual local drift.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of spectral detection and signal processing technology, and in particular to an intelligent method and system for analyzing spectral sensor signals. Background Technology

[0002] Spectral sensors acquire the spectral response of a measured object at different wavelengths, extracting peak position, peak height, peak area, absorbance, or spectral shape characteristics within the target spectral band to determine target component parameters or identify the category of the measured object. Spectral sensors typically include a light source, optical transmission components, a spectral detector, and signal processing devices; their acquisition results are affected by changes in the output state of the light source.

[0003] During long-term operation, light source aging, temperature changes, or fluctuations in drive state can cause output spectrum drift in spectral sensors, and this drift is usually inconsistent across different wavelengths. Existing technologies often employ global normalization or single reference wavelength compensation. Global normalization typically applies a uniform scaling to the entire spectrum, while single reference wavelength compensation determines a uniform compensation amount based on a reference position. Neither of these methods can adequately characterize the local drift trend with wavelength near the target spectral band. When the degree of drift varies across different wavelengths within the target spectral band, it can easily lead to a mismatch between the compensation amount for the target spectral band and the actual local drift. Summary of the Invention

[0004] To overcome the above shortcomings, this invention provides an intelligent analysis method and system for spectral sensor signals, aiming to improve the problems of difficulty in reflecting the changes of light source drift with wavelength and the mismatch between target spectral band compensation and actual local drift.

[0005] In a first aspect, the present invention provides the following technical solution: an intelligent analysis method and system for spectral sensor signals, comprising the following steps:

[0006] S1, acquire the current measurement spectrum and the reference response spectrum, and determine the target spectral band and the reference spectral band located on both sides of the target spectral band that does not contain significant characteristic responses of the target component;

[0007] S2, Based on the current measured spectrum and the reference response spectrum, generate a local response difference sequence covering the target spectral band and the reference spectral band;

[0008] S3, based on the local response difference sequence, select the effective reference sampling points within the reference spectrum, mask the sampling points corresponding to the target spectrum from the drift fitting data, and establish a local drift function across the target spectrum based on the effective reference sampling points;

[0009] S4, compensate the target spectral segment in the current measured spectrum according to the local drift function to obtain an equivalent target spectral segment;

[0010] S5. Determine the recovery confidence level based on the fitting quality of the local drift function and the continuity between the equivalent target spectral segment and the reference spectral segment, and output the analytical result based on the equivalent target spectral segment when the recovery confidence level meets the preset conditions.

[0011] By adopting the above technical solution, reference spectral bands are set on both sides of the target spectral band, and a local drift function spanning the target spectral band is established based on the effective reference sampling points. At the same time, the target spectral band is shielded from the drift fitting data. Therefore, the target spectral band can be compensated based on the wavelength-related drift near the target spectral band. By judging the recovery reliability of the compensation results, the influence of abnormal reference data and compensation results that do not meet the recovery conditions on the analysis results can be reduced.

[0012] Secondly, the present invention provides the following technical solution: an intelligent analysis system for spectral sensor signals, comprising the following modules:

[0013] The spectral acquisition and spectral band determination module is used to acquire the current measurement spectrum and the reference response spectrum, and to determine the target spectral band and the reference spectral bands located on both sides of the target spectral band.

[0014] The difference sequence generation module is used to generate a local response difference sequence covering the target spectral band and the reference spectral band based on the current measured spectrum and the reference response spectrum;

[0015] The drift modeling module is used to filter valid reference sampling points within the reference spectral band based on the local response difference sequence, mask the sampling points corresponding to the target spectral band from the drift fitting data, and establish a local drift function across the target spectral band based on the valid reference sampling points.

[0016] The target spectral band recovery module is used to compensate for the target spectral band in the current measured spectrum according to the local drift function to obtain an equivalent target spectral band;

[0017] The reliable analysis module is used to determine the recovery reliability based on the fitting quality of the local drift function and the continuity between the equivalent target spectral segment and the reference spectral segment, and output the analysis result based on the equivalent target spectral segment when the recovery reliability meets the preset conditions.

[0018] By adopting the above technical solution, each module sequentially completes spectral acquisition, differential sequence generation, reference sampling point screening, local drift modeling, target spectral band compensation, and confidence judgment. Therefore, it can realize continuous processing of spectral signals and reduce the impact of local light source drift and low-confidence compensation data on the system's analytical results.

[0019] 1. In this invention, a technical solution is adopted to set reference spectral bands on both sides of the target spectral band, establish a local drift function across the target spectral band based on effective reference sampling points, and compensate the target spectral band according to the local drift function. This achieves the technical effect of characterizing wavelength-related drift near the target spectral band and restoring the equivalent target spectral band. Compared with the existing technical solution of using overall normalization or a single reference wavelength for unified compensation, this solves the problem that it is difficult to reflect the change of light source drift with wavelength and the mismatch between target spectral band compensation and actual local drift.

[0020] 2. In this invention, a technical solution is adopted to evaluate the reliability of sampling points within the reference spectral band, remove abnormal reference sampling points, and shield the sampling points corresponding to the target spectral band from the drift fitting data. This achieves the technical effect of establishing a local drift function using effective reference information on both sides of the target spectral band. Compared with the existing technical solution of directly using all spectral sampling points or fixed reference sampling points for drift fitting, this solves the problem that it is easily affected by the characteristic response of the target component, interference response, and abnormal sampling points.

[0021] 3. In this invention, the technical solution of determining the recovery reliability based on the fitting quality of the local drift function and the continuity between the equivalent target spectral segment and the reference spectral segment, and controlling the analysis process based on the recovery reliability, achieves the technical effect of judging the reliability and controlling the analysis of the target spectral segment compensation result. Compared with the existing technical solution of directly outputting the analysis result after completing the spectral compensation, this invention solves the problem that it cannot identify insufficient reference information, abnormal drift fitting, or abnormal spectral segment connection. Attached Figure Description

[0022] Figure 1 This is a flowchart of a method for intelligent analysis of spectral sensor signals proposed in this invention;

[0023] Figure 2 This is a schematic diagram showing the configuration of the target spectral band and the two reference spectral bands on both sides of the intelligent analysis method for spectral sensor signals proposed in this invention;

[0024] Figure 3 This is a schematic diagram of the effective reference sampling point selection process for an intelligent analysis method of spectral sensor signals proposed in this invention.

[0025] Figure 4 This is a schematic diagram illustrating the process of establishing the local drift function, compensating for the target spectral band, and determining the reliability of the recovery in an intelligent analysis method for spectral sensor signals proposed in this invention.

[0026] Figure 5 This is a schematic diagram of the module framework of an intelligent analysis system for spectral sensor signals proposed in this invention. Detailed Implementation

[0027] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] Example 1:

[0029] In a first embodiment of the present invention, the present invention provides an intelligent analysis method for spectral sensor signals, such as... Figure 1 As shown, it includes the following steps:

[0030] S1, acquire the current measurement spectrum and the reference response spectrum, and determine the target spectral band and the reference spectral bands located on both sides of the target spectral band that do not contain significant characteristic responses of the target component;

[0031] Furthermore, the current dark signal corresponding to the current measured spectrum and the reference dark signal corresponding to the reference response spectrum are obtained; dark current is subtracted from the current measured spectrum and the reference response spectrum based on the current dark signal and the reference dark signal, respectively; and wavelength alignment is performed on the current measured spectrum and the reference response spectrum after dark current subtraction.

[0032] Specifically, in the first In each acquisition cycle, the spectral sensor acquires the spectrum of the object being measured, obtaining the current measured spectral vector: ;in, Indicates the first The current measured spectral vector for each acquisition cycle; Indicates the first The first collection cycle, the first The current measurement signal corresponding to each wavelength sampling point.

[0033] When the light source is in a reliable reference state, obtain the reference response spectral vector: ;in, Represents the reference response spectral vector; Indicates the first under the trusted baseline state The reference response signal corresponding to each wavelength sampling point; the subscript 0 indicates the reliable reference state.

[0034] A reliable reference state can be the factory calibration state of the spectral sensor, the field calibration state, the calibration state after the light source is replaced, or the calibration state established through a standard reference medium.

[0035] Get the current dark signal vector: ;

[0036] And the reference dark signal vector: ;

[0037] in, Indicates the first The current dark signal vector for each acquisition cycle; Indicates the first The first collection cycle, the first The current dark signal corresponding to each wavelength sampling point; Represents the reference dark signal vector; Indicates the first under the trusted baseline state The reference dark signal corresponding to each wavelength sampling point.

[0038] Dark current subtraction was performed on both the current measured spectrum and the reference response spectrum:

[0039] ;

[0040] ;

[0041] in, This represents the current effective spectral signal after dark current subtraction; This represents the reference effective spectral signal after dark current subtraction; This means taking the value inside the parentheses and... The larger value in; This indicates a preset positive lower limit; dark current subtraction is a common spectral preprocessing method.

[0042] Set the lower limit for positive values To prevent the value after dark current subtraction from being less than or equal to 0, which would prevent subsequent ratio or logarithmic calculations from being performed, this solution... Preferably set as the full-scale signal of the spectral detector to Further optimized by setting it to a full-scale signal .

[0043] The current effective spectral vector after dark current subtraction is expressed as: ;

[0044] The reference effective spectral vector after dark current subtraction is expressed as: ;

[0045] Furthermore, based on preset wavelength markers, calibration peak positions, or the calibration relationship between detector pixels and wavelengths, the current effective spectrum and the reference effective spectrum are wavelength aligned.

[0046] Let the first The wavelength axis offset for each acquisition cycle is According to Interpolate and resample the current effective spectrum to make and Corresponding to the same wavelength .

[0047] in, Indicates the first Wavelength axis offset per acquisition cycle; Units and same.

[0048] In this scheme, when When the sampling interval is greater than one wavelength, it is preferable to mark the current acquisition period as a wavelength alignment abnormal period and stop the local drift compensation of the current period.

[0049] like Figure 2 As shown, after completing dark current subtraction and wavelength alignment, the target spectral band is determined based on the standard spectrum of the target component.

[0050] The set of wavelengths corresponding to the target spectral band is represented as follows: ;

[0051] The set of wavelengths corresponding to the low-wavelength reference spectrum is represented as follows: ;

[0052] The set of wavelengths corresponding to the high-wavelength side reference spectrum is represented as follows: ;

[0053] in, Indicates the target spectral band; Indicates the low-wavelength side reference spectrum; This indicates the reference spectrum on the higher wavelength side.

[0054] In this scheme, the reference spectral band preferably satisfies the following conditions:

[0055] The response intensity of the target component in the reference spectral band is less than 10% of the maximum response intensity in the target spectral band; the response intensity of the known interfering component in the reference spectral band is less than the preset interference threshold; the maximum wavelength of the low-wavelength reference spectral band is less than the minimum wavelength of the target spectral band; the minimum wavelength of the high-wavelength reference spectral band is greater than the maximum wavelength of the target spectral band; the distance between the reference spectral band and the target spectral band does not exceed three times the width of the target spectral band; the change in the light source output meets the preset continuity condition within a local wavelength range containing the three spectral bands.

[0056] For example, the target spectral band can be set to 258 nm to 262 nm, the low-wavelength side reference spectral band can be set to 248 nm to 252 nm, and the high-wavelength side reference spectral band can be set to 268 nm to 272 nm.

[0057] Step S1 yields the current effective spectrum after dark current subtraction and wavelength alignment. Reference effective spectrum The data includes the index set corresponding to the three spectral bands. This data serves as the input for step S2, generating the local response difference sequence.

[0058] By subtracting dark current and aligning wavelengths, the impact of dark current variations and wavelength axis offsets on the calculation of response differences at the same wavelength position can be reduced. By setting reference spectral bands on both sides of the target spectral band, reference data for different wavelength positions near the target spectral band can be obtained.

[0059] S2, Based on the current measured spectrum and the reference response spectrum, generate a local response difference sequence covering the target spectral band and the reference spectral band;

[0060] Furthermore, the local wavelength range is determined based on the low-wavelength side reference spectrum, target spectrum, and high-wavelength side reference spectrum determined in step S1; the spectral data of the current effective spectrum and the reference effective spectrum obtained in step S1 within the local wavelength range are extracted; and the response difference is determined based on the current effective spectral signal and the reference effective spectral signal at the same wavelength position.

[0061] Specifically, the local wavelength range is defined as the minimum wavelength of the low-wavelength side reference spectrum to the maximum wavelength of the high-wavelength side reference spectrum.

[0062] The set of sampling point indices corresponding to a local wavelength range is represented as follows: ;

[0063] in, Represents the set of indices for all wavelength sampling points within a local wavelength range; This indicates the minimum wavelength of the low-wavelength reference spectrum. This indicates the maximum wavelength of the reference spectrum on the higher wavelength side.

[0064] The number of sampling points within a local wavelength range is expressed as: ,in, This represents the total number of wavelength sampling points within a local wavelength range.

[0065] According to the results obtained in step S1 and Calculate the first Logarithmic response difference at each wavelength sampling point: ,in, Indicates the first The first collection cycle, the first Logarithmic response difference at each wavelength sampling point; To represent the natural logarithm operation, since and Having the same dimensions, therefore the ratio and logarithmic response of the two differ. All parameters are dimensionless. The logarithmic ratio is used to convert the multiplicative intensity change into an additive change, which is a common data transformation method.

[0066] In this scheme, the natural logarithm ratio is preferably used to determine the local response difference. When the processor's computing power is limited or the local light source variation is small, a common ratio can also be used to determine the response difference.

[0067] Let the indices of the sampling points arranged in ascending order of wavelength within the local wavelength range be: Arrange the response differences in the above order to obtain the local response difference vector: ,in, Indicates the first Local response difference vector for each acquisition cycle; Represents the first wavelength within a local wavelength range The index of each sampling point in the complete spectrum; .

[0068] Based on the three spectral band index sets determined in step S1, the low-wavelength side reference difference vector, the target spectral band difference vector, and the high-wavelength side reference difference vector are extracted from the local response difference vector, respectively:

[0069] ; ; ;

[0070] in, This represents the response difference vector of the low-wavelength reference spectral band; Represents the response difference vector of the target spectral band; This represents the response difference vector of the reference spectrum on the higher wavelength side.

[0071] The result obtained in step S2 and As the data input for evaluating the reliability of the reference sampling points in step S3 and establishing the local drift function, the data obtained in step S2 It is only used for data comparison before and after target spectral band compensation, and does not participate in the local drift function fitting in step S3. Therefore, step S2 takes the wavelength aligned spectrum and spectral band configuration data obtained in step S1, and provides reference response difference data arranged by wavelength for step S3. By constructing a local response difference sequence, the difference between the current measurement state and the reliable reference state can be mapped to the same local wavelength range.

[0072] S3, based on the local response difference sequence, select the effective reference sampling points within the reference spectrum, mask the sampling points corresponding to the target spectrum from the drift fitting data, and establish a local drift function across the target spectrum based on the effective reference sampling points;

[0073] Further, valid reference sampling points are selected, including:

[0074] The local dispersion of each reference sampling point is determined based on the signal difference between each reference sampling point and the adjacent wavelength sampling points;

[0075] The time stability of each reference sampling point is determined based on the signal changes at each reference sampling point within a preset sliding time window.

[0076] The degree of interference at each reference sampling point is determined based on the matching relationship between each reference sampling point and the pre-stored interference spectral template.

[0077] The corresponding reliability evaluation results are determined based on the local dispersion, time stability, and interference level of each reference sampling point.

[0078] Based on the reliability evaluation results, reference sampling points that meet the preset reliability conditions are determined as valid reference sampling points;

[0079] Furthermore, a local drift function is established, including:

[0080] Based on the reliability evaluation results of each valid reference sampling point, determine the fitting weight corresponding to each valid reference sampling point;

[0081] Wavelength sampling points within the target spectral band are excluded from the fitting data used to establish the local drift function;

[0082] Based on the fitting weights, a robust fitting is performed on the response differences corresponding to the effective reference sampling points located on both sides of the target spectral band.

[0083] By applying continuity and rate of change constraints to the robust fitting results, the local drift function across the target spectral band is obtained.

[0084] Specifically, the set of indices of all candidate reference sampling points on both sides of the target spectral band is represented as: ,in, Represents the set of indices for all candidate reference sampling points; Represents the union of sets.

[0085] Determine the degree of local discretization: for any candidate reference sampling point Determine the set of adjacent sampling points of the reference sampling point within the same reference spectral band. .

[0086] The degree of local dispersion is calculated based on the deviation between the median difference in response between the reference sampling point and its adjacent sampling points: ;

[0087] in, Indicates the first The first collection cycle, the first The degree of local dispersion of each reference sampling point; Indicates the index of adjacent sampling points; Indicates the first The set of adjacent sampling points of each reference sampling point; This represents the median; Indicates the absolute deviation of the median; This represents a positive constant to prevent the denominator from being zero. The median and absolute deviation of the median are common robust statistics. Since both the numerator and denominator are dimensionless differences in response, therefore... It is a dimensionless parameter.

[0088] In this plan, Preferred includes the first Each reference sampling point has 2 to 4 sampling points on the left and right sides; The preferred setting is 1% to 5% of the response difference scale within the corresponding reference spectral band.

[0089] Determine the degree of time instability: settings include The sliding time window of the acquisition cycle is calculated to determine the first acquisition cycle. Mean difference in response among the reference sampling points within the sliding time window: ;

[0090] in, Indicates the first The mean difference in response of each reference sampling point within the current sliding time window; Indicates the number of acquisition cycles included in the sliding time window; This represents the historical period offset relative to the current acquisition period; Indicates the first The first collection cycle, the first The response differences at each wavelength sampling point.

[0091] Further calculation of the temporal instability of the reference sampling point: ;

[0092] in, Indicates the first The first collection cycle, the first The degree of temporal instability of each reference sampling point; This represents the scale of response differences among all candidate reference sampling points within the current sliding time window; This represents a positive constant to prevent the denominator from being zero.

[0093] In one implementation, The median of the time standard deviations of all candidate reference sampling points was used. Standard deviation is a common statistical indicator.

[0094] In this scheme, the sliding time window length The preferred setting is 10 to 30 acquisition cycles, and the more preferred setting is 20 acquisition cycles.

[0095] Determine the level of interference: Based on the matching relationship between the local wavelength region where the candidate reference sampling point is located and the pre-stored interference spectral template, determine the level of interference. ,in, Indicates the first The first collection cycle, the first The degree of interference at each reference sampling point; when The closer the value is to 1, the higher the degree of matching between the corresponding local spectral shape and the pre-stored interference spectral template. The degree of interference can be determined based on the correlation coefficient, cosine similarity, feature distance, or a pre-trained interference recognition model.

[0096] like Figure 3 As shown, the reliability weights are determined as follows: when the candidate reference sampling point is not in detector saturation and is above the detector's low response threshold, the reliability weights are determined based on the degree of local dispersion, temporal instability, and interference. ;

[0097] in, Indicates the first The first collection cycle, the first Reliability weights for each reference sampling point; The range of values ​​is ; Represents the natural exponential function; This represents the weighting coefficient corresponding to the degree of local dispersion. The weighting coefficients representing the degree of temporal instability; The weighting coefficient represents the degree of interference.

[0098] When the reference sampling point is in detector saturation or below the detector's low response threshold, the corresponding reliability weight is set as follows: In this plan, , and Preferred satisfaction: ,in, The preferred setting is 0.35; The preferred setting is 0.35; The preferred setting is 0.30.

[0099] The reliability weight is greater than or equal to the reliability threshold. The reference sampling points are determined as valid reference sampling points: ,in, Indicates the first A set of valid reference sampling point indices for each acquisition cycle; This represents the reliability threshold, in this scheme, The preferred value is 0.5 to 0.7, and more preferably 0.6.

[0100] The number of valid reference sampling points is expressed as: ,in, Indicates the first The total number of valid reference sampling points participating in local drift fitting in each acquisition cycle.

[0101] Wavelength normalization: To ensure that the fitting parameters across different wavelength ranges have a consistent numerical scale, the wavelengths are centered and normalized. ;

[0102] in, Indicates the first The normalized wavelength corresponding to each wavelength sampling point; Indicates the center wavelength of the target spectral band; Represents local wavelength scale parameters; It is a dimensionless parameter.

[0103] The center wavelength of the target spectral band is represented as: ,in, Indicates the minimum wavelength of the target spectral band; This indicates the maximum wavelength of the target spectral band.

[0104] The local wavelength scale parameter is expressed as: ,in, Indicates the maximum wavelength of the reference spectrum on the higher wavelength side; This indicates the minimum wavelength of the low-wavelength reference spectrum.

[0105] Under the above normalization method, the normalized wavelength within the local wavelength range is usually located at... nearby.

[0106] Establishing a local drift function: For cases where the target spectral bandwidth is small and the light source drift changes gradually with wavelength, a first-order local drift function is used. ,in, Indicates the first The first collection cycle, the first Local drift corresponding to each wavelength sampling point; The constant term represents the local drift function; Represents the coefficients of the first-order term of the local drift function; Indicates the normalized wavelength; due to Therefore, in response to the difference in response to the number of cases, , and All of these are dimensionless parameters.

[0107] The valid reference sampling points are arranged in ascending order of wavelength, and their indices are as follows: ;

[0108] Construct an effective reference response vector: ;

[0109] in, This represents the response difference vector corresponding to the valid reference sampling point; Indicates the first The index of each valid reference sampling point in the complete spectrum; .

[0110] Constructing a design matrix: ;

[0111] Construct the local drift parameter vector: ;

[0112] Construct the weight matrix: ;

[0113] in, The weight matrix represents the effective reference sampling points; This indicates that a diagonal matrix is ​​constructed using the elements within the parentheses as the main diagonal elements.

[0114] Weighted robust regression was used to determine the local drift parameters: ;in, This represents the local drift parameter vector obtained from the fitting process; This represents the parameter that minimizes the objective function; Representing vectors The first in One element; Design matrix The OK; The parameter is Huber loss function;

[0115] Represents the constraint coefficients of the first-order term; .

[0116] The Huber loss function is a common robust regression loss function, and its expression is:

[0117] ;

[0118] in, This represents the fitting residual of a single valid reference sampling point; This represents the switching threshold between the quadratic loss component and the linear loss component.

[0119] In this scheme, the Huber loss parameter The preferred setting is 1.345 times the scale of the fitted residuals; after the response differences are standardized, The preferred value is set to 1.345. In this scheme, the first-order term constraint coefficient is... Preferred settings to Further preferred settings When establishing the local drift function, the target spectral band index set The sampling points in the vector do not enter the vector Design Matrix and weight matrix When the local wavelength range is wide or the first-order function does not meet the preset fitting conditions, a restricted second-order function, a piecewise spline function, or a local regression function with smoothing constraints can also be used. When using a second-order function, it is preferable to set an upper limit on the coefficients of the second-order term; when using a piecewise spline function, it is preferable to limit the number of spline nodes and the curvature change between adjacent nodes.

[0120] Step S3 yields the local drift function. Valid reference sampling point set and its reliability weight The above data serves as the input for step S4, determining the compensation initiation conditions and calculating the drift compensation amount.

[0121] By selecting effective reference sampling points, the influence of abnormal peaks, unstable sampling points, and interfering spectra on local drift fitting can be reduced; by setting the target spectral band as a shielding point, the response differences within the target spectral band can be excluded from the parameter estimation of the local drift function.

[0122] like Figure 4 As shown in Figure S4, the target spectral band in the current measured spectrum is compensated according to the local drift function to obtain the equivalent target spectral band;

[0123] Furthermore, compensation for the target spectral band is performed based on the local drift function, including:

[0124] Determine whether the number of valid reference sampling points on both sides of the target spectral band has reached the preset number;

[0125] Determine whether the fitting residual of the local drift function at the effective reference sampling points is lower than the preset fitting threshold;

[0126] Determine whether the rate of change of the local drift function between two reference spectral bands is within a preset range.

[0127] When the number of effective reference sampling points, the fitting residual and the rate of change all meet the corresponding preset conditions, the drift compensation amount of each wavelength sampling point in the target spectrum is determined according to the local drift function.

[0128] Based on the drift compensation amount corresponding to each wavelength sampling point, the measured signal of the corresponding wavelength sampling point in the target spectrum is compensated to obtain the equivalent target spectrum.

[0129] If any preset condition is not met, compensation for the target spectral band shall be stopped.

[0130] Specifically, the number of effective reference sampling points on the low-wavelength side is expressed as follows: ;

[0131] The number of effective reference sampling points on the high wavelength side is expressed as: ;

[0132] in, Indicates the first Number of effective reference sampling points on the low wavelength side per acquisition cycle; Indicates the first Number of effective reference sampling points on the high-wavelength side per acquisition cycle.

[0133] Calculate the weighted root mean square fitting residual based on the actual response differences and local drift function values ​​of the valid reference sampling points: ;in, Indicates the first Weighted root mean square fitting residuals for each acquisition period; It is a dimensionless parameter; root mean square error is a common evaluation index for model fit.

[0134] The rate of change of the local drift function relative to the actual wavelength is determined based on the first-order term coefficients obtained from the fitting: ;in, This represents the rate of change of the local drift function relative to the actual wavelength; This represents the coefficients of the first-order term obtained from the fitting; Represents local wavelength scale parameters; The unit is .

[0135] Target spectral band compensation is initiated when the following conditions are met simultaneously: ; ; ; ;in, This indicates the minimum number of effective reference sampling points on the low-wavelength side. This indicates the minimum number of effective reference sampling points on the high-wavelength side; This represents the maximum permissible fitting residual; This represents the maximum permissible rate of change of local drift.

[0136] In this plan, and The preferred value is 3; in this scheme, Preferably, the margin is determined by adding a preset margin to the 95th quantile of the fitted residuals under a reliable baseline condition; when using logarithmic response differences... The preferred value is set to 0.02 to 0.05; in this scheme, The optimal value is determined through a light source aging calibration experiment and set to 1.1 to 1.3 times the maximum local change rate of a known normal aging sample.

[0137] If any of the above conditions are not met, the target spectral band compensation for the current acquisition cycle will not be performed, and the corresponding condition failure flag will be recorded.

[0138] When the compensation activation condition is met, the drift factor corresponding to each wavelength sampling point within the local wavelength range is calculated based on the local drift function: ;in, Indicates the first The first collection cycle, the first The drift factor corresponding to each wavelength sampling point; It is a dimensionless parameter greater than 0.

[0139] Based on the drift magnification, calculate the equivalent local spectral signal under the reference light source condition: ;

[0140] Substituting the drift ratio expression, we get: ;in, Indicates the first The first collection cycle, the first The equivalent spectral signal of each wavelength sampling point under the reference light source condition; and They have the same dimensions.

[0141] The equivalent local spectral vector is represented as: , where superscript It indicates a local wavelength range, not a transpose.

[0142] Based on the target spectral band index set Extract the equivalent target spectral vector from the equivalent local spectrum: ;in, Indicates the first Equivalent target spectral segment vector for each acquisition cycle; superscript Indicates the target spectral band, not the transpose.

[0143] Step S4 generates an equivalent local spectrum across the entire local wavelength range and extracts the equivalent target spectral band from it. The equivalent local spectrum and the equivalent target spectral band serve as inputs for step S5 to determine boundary alignment consistency and recovery reliability.

[0144] By first forming an equivalent local spectrum and then comparing the target spectral band with the reference spectral band, the two can be placed in the same light source reference state.

[0145] S5. Determine the recovery confidence level based on the fitting quality of the local drift function and the continuity between the equivalent target spectral segment and the reference spectral segment. When the recovery confidence level meets the preset conditions, output the analytical results based on the equivalent target spectral segment.

[0146] Furthermore, the recovery confidence level is determined and the parsing results are output, including:

[0147] The fitting quality is determined based on the deviation between the fitting result of the local drift function at the effective reference sampling point and the corresponding response difference;

[0148] The continuity of the low-wavelength side boundary is determined based on the signal differences and variation trends between the equivalent target spectral band and the low-wavelength side reference spectral band at adjacent boundaries.

[0149] The continuity of the high-wavelength side boundary is determined based on the signal differences and variation trends between the equivalent target spectral band and the high-wavelength side reference spectral band at adjacent boundaries.

[0150] The recovery reliability is determined based on the fitting quality, the continuity of the low-wavelength side boundary, and the continuity of the high-wavelength side boundary.

[0151] When the recovery confidence meets the preset conditions, the equivalent target spectral band is input into the target parameter analysis model or the category recognition model to obtain the analysis result;

[0152] When the recovery confidence does not meet the preset conditions, the corresponding status information is output;

[0153] Specifically, determine the fit quality: based on the weighted root mean square fitting residuals obtained in step S4. Determine the fitting quality index: ;in, Indicates the first Fitting quality index for each acquisition cycle; The range of values ​​is ; This represents the scaling parameter of the fitted residuals; In this plan, The preferred setting is the maximum allowable fitting residual. Determine the consistency of boundary connections: Let the index of the sampling point corresponding to the minimum wavelength of the target spectral band be... The sampling point index corresponding to the maximum wavelength of the target spectral band is .

[0154] A local prediction function for the low-wavelength side is established based on several equivalent spectral sampling points of the low-wavelength side reference band close to the target spectral band, and then extrapolated to the low-wavelength boundary of the target spectral band to obtain the predicted value of the low-wavelength side boundary. .

[0155] A local prediction function for the high-wavelength side is established based on several equivalent spectral sampling points of the high-wavelength side reference band close to the target spectral band, and then extrapolated to the high-wavelength boundary of the target spectral band to obtain the predicted value of the high-wavelength side boundary. ;in, This represents the target spectral band low-wavelength boundary signal predicted based on the equivalent reference spectrum on the low-wavelength side. This indicates the high-wavelength boundary signal of the target spectral band predicted based on the equivalent reference spectrum on the high-wavelength side.

[0156] The normalized amplitude difference on the low-wavelength side is expressed as: ;

[0157] The normalized amplitude difference on the high wavelength side is expressed as: ;

[0158] in, This indicates the normalized amplitude difference on the low-wavelength side; This indicates the normalized amplitude difference on the higher wavelength side; This represents a positive constant in boundary calculations.

[0159] In this plan, Preferably set to the full-scale signal of the detector to .

[0160] Furthermore, the low-wavelength lateral slope difference and the high-wavelength lateral slope difference are determined based on the local slopes on both sides of the target spectral band boundary.

[0161] The low-wavelength side slope difference is denoted as The high-wavelength side slope difference is denoted as ,in, This represents the difference in normalized slope on both sides of the low-wavelength boundary of the target spectral band. This represents the difference in normalized slope on both sides of the high-wavelength boundary of the target spectral band.

[0162] The local slope is preferably obtained by first-order fitting based on at least three wavelength sampling points near the boundary.

[0163] Determine the consistency of boundary alignment based on the difference in amplitude and slope: ;in, Indicates the first Consistency index of boundary connection for each collection cycle; The range of values ​​is ; This indicates the weight corresponding to the magnitude difference; This represents the weight corresponding to the slope difference.

[0164] In this plan, and Preferred satisfaction: Furthermore: The preferred setting is 0.6; The preferred setting is 0.4.

[0165] Determine the degree of preservation of target feature structure: Determine the position and width of the target feature peaks from the target spectral band before compensation. The position of the target feature peak before compensation is denoted as... The peak width of the target feature before compensation is denoted as The compensated target characteristic peak positions and peak widths are determined from the equivalent target spectral band. The compensated target characteristic peak positions are denoted as... The peak width of the target feature after compensation is denoted as The target feature structure preservation index is determined based on the changes in peak position and peak width. ;in, Indicates the first The target feature structure retention index for each acquisition cycle; The range of values ​​is ; Indicates the position of the target characteristic peak before compensation; Indicates the position of the target characteristic peak after compensation; Indicates the peak width of the target feature before compensation; Indicates the peak width of the target feature after compensation; Indicates the allowable range of peak position variation; This indicates the allowable variation scale of peak width.

[0166] In this plan, The preferred setting is 1 to 2 wavelength sampling intervals; The preferred setting is 5% to 15% of the reference peak width of the target characteristic peak.

[0167] Determining the reliability of the recovery: The reliability of the recovery is determined based on the fitting quality index, the boundary connection consistency index, and the target feature structure preservation index. ;in, Indicates the first The reliability of recovery in each acquisition cycle; This represents the combined weights corresponding to the fit quality index; This represents the combined weights corresponding to the boundary alignment consistency index. This represents the combined weights corresponding to the target feature structure preservation index.

[0168] The combined weights satisfy: ;and: ; ; ;

[0169] In this plan: The preferred setting is 0.40; The preferred setting is 0.35; The preferred setting is 0.25.

[0170] Will restore credibility Compared with the preset confidence threshold Compare them.

[0171] In this plan, The value is preferably set to 0.75 to 0.90, and more preferably to 0.80.

[0172] When the following conditions are met: At that time, the equivalent target spectral vector obtained in step S4 is... Transmitted to the target parameter parsing model or category recognition model.

[0173] When the following conditions are met: When the time is right, stop transmitting data from the equivalent target spectral band to the target parameter analysis model or category recognition model, and record the corresponding status identifier.

[0174] The target parameter analysis model can determine the target parameters based on the peak height, peak area, absorbance, peak width, or spectral shape parameters of the equivalent target spectral band. The category recognition model can use support vector machines, decision trees, probabilistic classification models, or neural network models. Thus, step S5 takes the equivalent local spectrum and equivalent target spectral band formed in step S4 and controls the data transmission between the equivalent target spectral band and the subsequent analysis model based on the recovery confidence. By judging the recovery confidence, the local drift fitting results, the target spectral band compensation results, and the subsequent analysis process can be correlated and controlled.

[0175] Example 2:

[0176] During long-term operation of an online ultraviolet absorption spectroscopy sensor, the output of the ultraviolet light source may change with the time of use, and the change may have different amplitudes at different wavelength positions. When the characteristic absorption spectrum of the target component is located within the wavelength range where the local output of the light source changes, the baseline position and local change trend of the target spectrum may change with the state of the light source.

[0177] When using the overall normalization method, the average value or total energy of the entire spectral range is used for unified normalization, which cannot represent the wavelength-related changes near the target spectral band individually. When using a single reference wavelength for compensation, only a single compensation ratio can be obtained, which cannot represent the local variation differences between the low-wavelength side and the high-wavelength side of the target spectral band.

[0178] To address the aforementioned problems, this embodiment provides an intelligent analysis system for spectral sensor signals used in online water sample detection, such as... Figure 5 As shown, the specific implementation process of this system is as follows:

[0179] The system includes an ultraviolet light source, a flow-through measurement cell, an array-type spectrometer, a dark signal acquisition circuit, a processor, a memory, and a data output interface;

[0180] The ultraviolet light source is used to emit detection light into the flow-through measurement cell, which is used to hold the water sample to be tested. After the detection light passes through the water sample, it enters the array-type spectrometer. The array-type spectrometer is used to convert the light signals at different wavelength positions into spectral sampling data. The dark signal acquisition circuit is used to acquire the dark signal of the detector when the ultraviolet light source is turned off or the detection light path is blocked. The memory is used to store the reference response spectrum, the reference dark signal, the target spectral band configuration data, the reference spectral band configuration data, the interference spectral template, the reliability evaluation parameters, the local drift function parameter range, and the confidence threshold.

[0181] The processor is used to execute steps S1 to S5 in Embodiment 1. In this embodiment, 258nm to 262nm is set as the target spectral band, 248nm to 252nm is set as the low-wavelength side reference spectral band, and 268nm to 272nm is set as the high-wavelength side reference spectral band. The above wavelength range is only an example. In other detection scenarios, the target spectral band and reference spectral band can be reconfigured according to the actual spectral response of the target component.

[0182] First, with the ultraviolet light source in a reliable calibration state, a standard reference medium is introduced into the flow-through measurement cell to collect the reference response spectrum.

[0183] The ultraviolet light source is turned off or the detection optical path is blocked. The reference dark signal is collected, and the reference response spectrum, the reference dark signal and the corresponding wavelength calibration data are stored in the memory. During the online detection process, the ultraviolet light source emits detection light to the flow-through measurement cell according to the preset driving parameters, and the array spectrometer collects the current measurement spectrum of the current water sample according to the preset acquisition cycle.

[0184] The dark signal acquisition circuit acquires the current dark signal according to the preset dark signal acquisition cycle. The processor performs dark current subtraction on the current measured spectrum based on the current dark signal and performs dark current subtraction on the reference response spectrum based on the reference dark signal. The processor performs wavelength alignment on the current effective spectrum and the reference effective spectrum based on the pre-stored calibration relationship between detector pixels and wavelength.

[0185] The processor extracts the current effective spectrum and the reference effective spectrum in the range of 248nm to 272nm, and calculates the logarithmic response difference corresponding to each wavelength sampling point to form a local response difference vector. The processor marks the wavelength sampling points in the range of 258nm to 262nm as the target spectral band shielding points so that the response difference in the target spectral band does not enter the local drift function fitting process. For the candidate reference sampling points in the range of 248nm to 252nm and 268nm to 272nm, the processor calculates the local discreteness, time instability and interference degree respectively.

[0186] When a reference sampling point exhibits an isolated spike relative to adjacent wavelength sampling points, the local dispersion of the reference sampling point increases. When a reference sampling point exhibits irregular fluctuations within multiple consecutive acquisition cycles, the temporal instability of the reference sampling point increases. When the matching degree between the local spectral shape of a reference sampling point and the pre-stored interference spectral template increases, the interference level of the reference sampling point increases. The processor determines the reliability weight of each candidate reference sampling point based on the above indicators, and determines the candidate reference sampling points with reliability weights greater than or equal to a preset reliability threshold as valid reference sampling points. In this scheme, the reliability threshold is preferably set to 0.6.

[0187] The processor counts the number of valid reference sampling points in the low-wavelength side reference spectrum and the high-wavelength side reference spectrum respectively. In this scheme, it is required that there are at least 3 valid reference sampling points in the low-wavelength side reference spectrum and the high-wavelength side reference spectrum respectively. The processor establishes a first-order local drift function based on the normalized wavelength, response difference and reliability weight corresponding to the valid reference sampling points. In this embodiment, the center wavelength of the target spectrum is 260nm and the local wavelength range is 248nm to 272nm.

[0188] The processor uses the Huber loss function and first-order term constraints to perform weighted robust fitting on the parameters of the local drift function. The processor calculates the weighted root mean square fitting residual of the local drift function at the effective reference sampling points and compares the weighted root mean square fitting residual with the maximum allowable fitting residual. The processor calculates the local drift change rate based on the coefficients of the first-order term of the local drift function and compares the local drift change rate with the maximum allowable change rate.

[0189] When the number of effective reference sampling points on both the low-wavelength and high-wavelength sides reaches the preset number, the weighted root mean square fitting residual does not exceed the maximum allowable fitting residual, and the local drift change rate does not exceed the maximum allowable change rate, the processor performs target spectral band compensation. The processor calculates the drift factor corresponding to each wavelength sampling point in the range of 248nm to 272nm based on the local drift function, compensates the current effective spectrum based on the drift factor, and forms an equivalent local spectrum in the range of 248nm to 272nm. Spectral data in the range of 258nm to 262nm are extracted from the equivalent local spectrum to form an equivalent target spectral band.

[0190] The processor determines the fitting quality index based on the weighted root mean square fitting residual. The processor determines the low-wavelength side boundary connection consistency based on the amplitude difference and slope difference between the equivalent target spectral segment and the equivalent reference spectral segment on the low wavelength side at the low wavelength boundary. The processor determines the high-wavelength side boundary connection consistency based on the amplitude difference and slope difference between the equivalent target spectral segment and the equivalent reference spectral segment on the high wavelength side at the high wavelength boundary. The processor determines the target feature structure preservation index based on the target feature peak position and peak width before and after compensation.

[0191] The processor determines the recovery confidence level based on the fitting quality index, the boundary connection consistency index, and the target feature structure preservation index. In this scheme, the preset confidence threshold for the recovery confidence level is preferably set to 0.80. When the recovery confidence level is greater than or equal to the preset confidence threshold, the processor transmits the equivalent target spectral band to the target parameter analysis model. The target parameter analysis model extracts the absorbance, peak height, or peak area in the equivalent target spectral band and determines the target parameters based on the pre-established calibration relationship.

[0192] When the recovery confidence is less than the preset confidence threshold, the processor blocks the data transmission between the equivalent target spectrum and the target parameter analytical model and records the corresponding status identifier. The status identifier may include insufficient reference sampling points, abnormal local drift fitting, abnormal local drift change rate, or abnormal boundary connection. During continuous detection, the processor repeats the above process according to the preset acquisition cycle and stores the local drift function parameters, fitting residuals, number of effective reference sampling points, and recovery confidence corresponding to each acquisition cycle in the memory. The processor can also establish a time series based on the local drift function parameters of multiple consecutive acquisition cycles.

[0193] When the change in local drift function parameters between adjacent acquisition cycles exceeds a preset time change threshold, the processor marks the current acquisition cycle as a cycle to be reviewed and stops the data transmission of the corresponding equivalent target spectral band to the target parameter analytical model. This embodiment takes online ultraviolet absorption spectroscopy detection as an example for illustration. The present invention can also be applied to visible light spectroscopy, near-infrared spectroscopy, fluorescence spectroscopy, Raman spectroscopy, reflection spectroscopy or transmission spectroscopy sensors.

[0194] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for intelligent analysis of spectral sensor signals, characterized in that, Includes the following steps: S1, acquire the current measurement spectrum and the reference response spectrum, and determine the target spectral segment and the reference spectral segments located on both sides of the target spectral segment that do not contain significant characteristic responses of the target component; S2, Based on the current measured spectrum and the reference response spectrum, generate a local response difference sequence covering the target spectral band and the reference spectral band; S3, based on the local response difference sequence, select the effective reference sampling points within the reference spectrum, mask the sampling points corresponding to the target spectrum from the drift fitting data, and establish a local drift function across the target spectrum based on the effective reference sampling points; S4, compensate for the target spectral segment in the current measured spectrum according to the local drift function to obtain an equivalent target spectral segment; S5. Determine the recovery confidence level based on the fitting quality of the local drift function and the continuity between the equivalent target spectral segment and the reference spectral segment, and output the analytical result based on the equivalent target spectral segment when the recovery confidence level meets the preset conditions.

2. The intelligent analysis method for spectral sensor signals according to claim 1, characterized in that, In step S1, acquiring the current measurement spectrum and the reference response spectrum includes: Obtain the current dark signal corresponding to the current measured spectrum, and the reference dark signal corresponding to the reference response spectrum; Dark current is subtracted from the current measured spectrum based on the current dark signal to obtain the current effective spectrum; Dark current subtraction is performed on the reference response spectrum based on the reference dark signal to obtain the reference effective spectrum; The current effective spectrum and the reference effective spectrum are aligned according to the preset wavelength marking.

3. The intelligent analysis method for spectral sensor signals according to claim 1, characterized in that, In step S1, determining the target spectral band and the reference spectral band includes: The target spectral band is determined based on the characteristic spectral response of the target component; A low-wavelength reference spectral band is determined on the low-wavelength side of the target spectral band, and a high-wavelength reference spectral band is determined on the high-wavelength side of the target spectral band. The low-wavelength side reference spectrum and the high-wavelength side reference spectrum are determined from the candidate reference spectrum based on the response intensity of the target component in the candidate reference spectrum, the response intensity of the interfering component in the candidate reference spectrum, and the wavelength distance between the candidate reference spectrum and the target spectrum.

4. The intelligent analysis method for spectral sensor signals according to claim 1, characterized in that, In step S2, generating a local response difference sequence includes: Determine the local wavelength range based on the target spectral band and the reference spectral band; Extract the spectral data of the current measured spectrum and the reference response spectrum within the local wavelength range, respectively; Based on the signal ratio or logarithmic ratio of the current measured spectrum and the reference response spectrum at the same wavelength position, determine the response difference corresponding to each wavelength sampling point; The local response difference sequence is obtained by arranging the response differences corresponding to each wavelength sampling point according to the wavelength position.

5. The intelligent analysis method for spectral sensor signals according to claim 1, characterized in that, In step S3, filtering valid reference sampling points includes: The local dispersion of each reference sampling point is determined based on the signal difference between each reference sampling point and the adjacent wavelength sampling points; The time stability of each reference sampling point is determined based on the signal changes at each reference sampling point within a preset sliding time window. The degree of interference at each reference sampling point is determined based on the matching relationship between each reference sampling point and the pre-stored interference spectral template. The corresponding reliability evaluation results are determined based on the local dispersion, time stability, and interference level of each reference sampling point. Based on the reliability evaluation results, the reference sampling points that meet the preset reliability conditions are determined as the valid reference sampling points.

6. The intelligent analysis method for spectral sensor signals according to claim 1, characterized in that, In S3, the local drift function is established, including: Based on the reliability evaluation results of each valid reference sampling point, determine the fitting weight corresponding to each valid reference sampling point; The wavelength sampling points within the target spectral band are excluded from the fitting data used to establish the local drift function; Based on the fitting weights, robust fitting is performed on the response differences corresponding to the effective reference sampling points located on both sides of the target spectral band; By applying continuity and rate of change constraints to the robust fitting results, a local drift function spanning the target spectral band is obtained.

7. The intelligent analysis method for spectral sensor signals according to claim 1, characterized in that, In step S4, compensating for the target spectral band based on the local drift function includes: Determine whether the number of valid reference sampling points located on both sides of the target spectral band has reached a preset number; Determine whether the fitting residual of the local drift function at the effective reference sampling point is lower than a preset fitting threshold; Determine whether the rate of change of the local drift function between the two reference spectral bands is within a preset rate of change range; When the number of effective reference sampling points, the fitting residual, and the rate of change all meet the corresponding preset conditions, the drift compensation amount of each wavelength sampling point in the target spectrum is determined according to the local drift function. Based on the drift compensation amount corresponding to each wavelength sampling point, the measured signal of the corresponding wavelength sampling point in the target spectrum is compensated to obtain the equivalent target spectrum. If any preset condition is not met, compensation for the target spectral band shall be stopped.

8. The intelligent analysis method for spectral sensor signals according to claim 1, characterized in that, In step S5, determining the recovery confidence level and outputting the parsing result includes: The fitting quality is determined based on the deviation between the fitting result of the local drift function at the effective reference sampling point and the corresponding response difference; The continuity of the low-wavelength side boundary is determined based on the signal differences and variation trends between the equivalent target spectral band and the low-wavelength side reference spectral band at adjacent boundaries. The continuity of the high-wavelength side boundary is determined based on the signal differences and variation trends between the equivalent target spectral band and the high-wavelength side reference spectral band at adjacent boundaries. The recovery reliability is determined based on the fitting quality, low-wavelength side boundary continuity, and high-wavelength side boundary continuity. When the recovery confidence meets the preset conditions, the equivalent target spectral band is input into the target parameter analysis model or the category recognition model to obtain the analysis result; When the recovery reliability does not meet the preset conditions, the corresponding status information is output.

9. An intelligent analysis system for spectral sensor signals, characterized in that, The intelligent analysis method for a spectral sensor signal according to any one of claims 1-8 includes the following modules: The spectral acquisition and spectral band determination module is used to acquire the current measurement spectrum and the reference response spectrum, and to determine the target spectral band and the reference spectral bands located on both sides of the target spectral band. The difference sequence generation module is used to generate a local response difference sequence covering the target spectral band and the reference spectral band based on the current measured spectrum and the reference response spectrum; The drift modeling module is used to filter valid reference sampling points within the reference spectral band based on the local response difference sequence, mask the sampling points corresponding to the target spectral band from the drift fitting data, and establish a local drift function across the target spectral band based on the valid reference sampling points. The target spectral band recovery module is used to compensate for the target spectral band in the current measured spectrum according to the local drift function to obtain an equivalent target spectral band; The reliable analysis module is used to determine the recovery reliability based on the fitting quality of the local drift function and the continuity between the equivalent target spectral segment and the reference spectral segment, and output the analysis result based on the equivalent target spectral segment when the recovery reliability meets the preset conditions.

10. The intelligent analysis system for spectral sensor signals according to claim 9, characterized in that, The trusted resolution module includes: The credibility determination unit is used to determine the fitting quality based on the deviation between the fitting result of the local drift function at the effective reference sampling point and the corresponding response difference, to determine the boundary continuity based on the signal difference and change trend difference between the equivalent target spectrum and the reference spectrum on both sides of the target spectrum, and to determine the recovery credibility based on the fitting quality and the boundary continuity. The analysis and control unit is used to compare the recovery confidence with preset confidence conditions, and control the data transmission between the equivalent target spectral band and the target parameter analysis model or category recognition model according to the comparison result.