Spectral baseline correction method based on DE algorithm and non-uniform B-spline curve

Abstract

The present invention provides a spectral baseline correction method based on DE algorithm and non-uniform B-spline curve. Aiming at the baseline drift phenomenon existing in the ICP-AES detection system, the standard deviation of the minimum value sequence in the spectral baseline correction process is used as an evaluation index, the non-uniform B-spline curve is used as the baseline model, the control point sequence C and the inner node sequence T of the curve are used as the characteristic parameters of the evaluation function, and the spectral baseline correction evaluation function is established. The differential evolution algorithm is used to solve the optimal solution of the evaluation function, and the NURBS curve is fitted as the spectral baseline, so as to correct the spectral curve. The problem of large limitations and high computational complexity of the existing common baseline correction methods is solved. The spectral baseline is fitted by non-uniform B-spline, and the control point sequence C and the inner node sequence T of the non-uniform B-spline are used as the characteristic parameters of the evaluation function and the initial conditions of differential evolution. A set of optimal parameters C and T are adaptively selected, and the empirical selection of parameters C and T is avoided.

Description

Technical Field

[0001] The invention relates to the technical field of spectroscopy and information processing, and in particular to a spectral baseline correction method based on a DE algorithm and a non-uniform B-spline curve. Background Art

[0002] Inductively coupled plasma atomic emission spectrometry (ICP-AES) is an atomic emission spectrometry analysis technique that uses inductively coupled plasma as the excitation light source. Under thermal or electrical excitation, each atom or ion in the excited state radiates composite light of different characteristic wavelengths when returning to the ground state. By detecting the light intensity at the characteristic wavelength of the element, qualitative and quantitative element analysis is performed. ICP-AES has the characteristics of simultaneous multi-element detection, fast analysis speed and high accuracy, and is widely used in many fields such as rare earth, metallurgy, chemical industry, inorganic materials and medicine.

[0003] However, during the ICP-AES measurement process, the measured spectrum usually has a certain degree of baseline drift due to the influence of the instrument dark current and the stray light of the light source, resulting in errors in the quantitative analysis of the element content. Therefore, it is often necessary to perform baseline correction on the spectrum during the measurement process. Common baseline correction methods include iterative polynomial fitting, derivative method, moving window smoothing, wavelet transform, etc. However, the above methods all have certain limitations in practical applications. For example, when using the iterative polynomial fitting method for correction, for different spectral signals, the polynomial fitting order often needs to be demonstrated and determined, and it is difficult to establish a spectral baseline fitting model with strong universality; when using the derivative method and the moving window smoothing method for correction, both will reduce the resolution of the spectrum in the time domain, affect the peak shape of the spectrum, and are not conducive to subsequent quantitative analysis; when using the wavelet transform method for correction, it is necessary to establish corresponding wavelet basis functions for different spectral signals, so the algorithm has a relatively high amount of calculation and the application efficiency is relatively low.

[0004] Therefore, how to provide a spectral baseline correction method that breaks through the limitations of common baseline correction methods, has low computational complexity, and is efficient has become an urgent problem to be solved. Summary of the invention

[0005] The present invention provides a spectral baseline correction method based on DE algorithm and non-uniform B-spline curve, which is used to solve the problems of large limitations and high computational complexity of existing common baseline correction methods.

[0006] In order to achieve the above-mentioned purpose, the technical solution of the present invention provides a spectral baseline correction method based on differential evolution and non-uniform B-spline curve, including: using Gaussian filtering to eliminate the noise signal in the original spectral signal to obtain a filtered spectral curve. A spectral baseline model is obtained by dividing the trough points of the filtered spectral curve. A spectral baseline correction evaluation function F(C, T) is established based on the spectral intensity value of the filtered spectral curve and the spectral intensity value of the filtered spectral curve after baseline correction:

[0007]

[0008] Wherein, p is the set of minimum value sequences in the original spectral signal, and STD represents the standard deviation of the calculated set. The optimal solution of the spectral baseline evaluation function is obtained, and the optimal solution is decomposed into an optimal control point vector and an optimal internal node sequence, and the optimal control point vector and the optimal internal node sequence are substituted into the spectral baseline mathematical model to obtain a corrected spectral curve.

[0009] As a preferred embodiment of the above scheme, preferably, a Gaussian filter is used to eliminate the noise signal in the spectral signal, including: performing a discrete convolution operation on the original spectral signal and a Gaussian convolution kernel:

[0010]

[0011] Among them, f is the original spectral signal, g is the Gaussian convolution kernel, n is the nth point of the convolved signal, τ is the offset, and f is the original spectral curve.

[0012] As a preferred embodiment of the above scheme, a non-uniform B-spline curve is obtained by dividing the trough points of the filtered spectral curve, including: recording the trough points of the filtered spectral curve to obtain a sequence set X, and dividing the data point sequence X with a node symbol t to obtain an inner node sequence T to obtain a non-uniform B-spline curve, wherein the expression of the non-uniform B-spline curve is the spectral baseline model B(C, T), which is expressed as follows:

[0013] Where C = [c0, c1, ..., c n ] is the B-spline curve control point sequence, c is the B-spline curve control point, c i is the i-th control point in the B-spline control point sequence, y is the B-spline curve coordinate point sequence, N i k (x) is the B-spline basis function.

[0014] As a preferred embodiment of the above scheme, preferably, the trough points of the filtered spectral curve are recorded to obtain a sequence set X, and the data point sequence X is divided by a node symbol t to obtain an inner node sequence T, including: X = {(x 1|u , x 1|v ), (x 2|u , x 2|v ), (x 3|u , x 3|v ),…,(x n|u , x n|v )}, u is the horizontal coordinate of the trough point, and v is the vertical coordinate of the trough point. After the node sequence X is divided into the internal node sequence T, T = [t -k+1 , t -k+2 ,…,t -1 , t0, t1, …, t n , t n+1 , t n+2 ,…,t n+k ], each node in the node sequence set T satisfies the following relationship:

[0015] t -k+1 ≤t -k+2 ≤…≤t -1 ≤t0<t1<…<t n <t n+1 ≤t n+2 <…<t n+k ;

[0016] Among them, t1 to t n are called inner nodes, and the rest are called outer nodes. n is the number of inner nodes, and k is the order of the B-spline curve.

[0017] As a preferred embodiment of the above scheme, preferably, a spectral baseline correction evaluation function F(C, T) is established according to the spectral intensity value of the filtered spectral curve and the spectral intensity value of the filtered spectral curve after baseline correction, including: assuming that the spectral intensity value of the filtered spectral curve obeys the model I'(x), the spectral intensity value of the filtered spectral curve after baseline correction obeys the model I(x), the baseline model is B(C, T), the superposition of the spectral intensity values ​​is a linear superposition, and the following formula is obtained: I(x)=I'(x)-B(C, T), and the correction intensity value set I is obtained according to the non-uniform B-spline curve Y , with the correction intensity value set I Y The standard deviation of is used as the measurement criterion, and the spectral baseline evaluation function F(C, T) is established.

[0018] As a preferred embodiment of the above scheme, preferably, the correction intensity value set I is obtained according to the non-uniform B-spline curve Y, including: suppose the minimum value sequence set in the filtered spectrum curve is P: P = {(p0, I'(p0)), (p1, I'(p1)), ..., (p n , I'(p n ))}, and record P X ={p0, p1, ..., p n}, P Y ={I'(p0), I'(p1),...,I'(p n )}, where P X is the abscissa set of the minimum value sequence set P, P Y is the ordinate set of the minimum value sequence set P; the baseline correction is performed on the filtered spectral curve, specifically, the difference between the non-uniform B-spline curve and the filtered spectral curve is obtained to obtain the set P X The corresponding correction intensity value set I Y , I Y = {I(p0), I(p1), ..., I(p n )}.

[0019] As a preferred and better embodiment of the above scheme, it includes: a. taking the control point sequence C and the internal node sequence T in the non-uniform B-spline as the population individual vectors of the differential evolution algorithm, and initializing the population; b. bringing the several individual vectors obtained after the population initialization into the spectral baseline correction evaluation function respectively to obtain the evaluation value of each individual vector; c. in each round of evolutionary iteration, performing mutation operation on all individual vectors in the current iteration to obtain mutation vectors corresponding to each individual vector; d. cross-exchanging the information components of each individual vector with the corresponding mutation vectors to obtain several cross vectors; e. substituting all individual vectors and cross vectors into the fitness function for comparison, and bringing the vectors with small fitness into the next round of differential evolution algorithm.

[0020] As a preferred embodiment of the above scheme, preferably, step e also includes that after the last iteration, the individual vector with the smallest fitness in the population is the optimal vector of the spectral baseline evaluation function.

[0021] As a preferred embodiment of the above scheme, preferably, after decomposing the optimal vector into the optimal control point vector C and the optimal internal node sequence T, the spectral baseline model B (C, T) is introduced to obtain a new non-uniform B-spline curve; the new non-uniform B-spline curve is used as the current spectral baseline, combined with the formula:

[0022] I(x)=I′(x)-B(C, T), and the corrected spectral intensity value obeys the model I(x), thereby obtaining a corrected spectral curve.

[0023] The technical solution of the present invention provides a spectral baseline correction method based on DE algorithm and non-uniform B-spline curve. Aiming at the baseline drift phenomenon existing in the ICP-AES detection system, the standard deviation of the minimum value sequence in the spectral baseline correction process is used as the evaluation index, the non-uniform B-spline curve is used as the baseline model, the control point sequence C and the in-contact point sequence T of the curve are used as the characteristic parameters of the evaluation function, and the ICP-AES spectral baseline correction evaluation function is established. Then, the differential evolution algorithm is used to solve the optimal solution of the evaluation function, and the NURBS curve is fitted as the spectral baseline, thereby obtaining the corrected spectral curve. The problem of large limitations and high computational complexity of the existing common baseline correction methods is solved.

[0024] The advantages are: the spectral baseline is fitted using non-uniform B-splines, which have low-order smoothing characteristics. The fitted curve is closer to the real spectral baseline (the spectral baseline of ICP-AES is usually a low-order linear curve), and it has an advantage in solving the problem of ICP-AES spectral baseline drift; by taking the standard deviation of the minimum value sequence in the spectral baseline correction process as the evaluation index, the control point sequence C and the internal node sequence T of the non-uniform B-spline as the characteristic parameters of the evaluation function, an ICP-AES spectral baseline correction evaluation function is constructed, and the characteristic parameters C and T are used as the second initial condition of differential evolution, and then the differential evolution algorithm can be used to adaptively select a set of optimal parameters C and T, which effectively avoids the problem of artificial experience in selecting parameters C and T. BRIEF DESCRIPTION OF THE DRAWINGS

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

[0026] Figure 1 A flow chart of a spectral baseline correction method based on a DE algorithm and a non-uniform B-spline curve provided by the present invention. ;

[0027] Figure 2 It is the original sample spectral signal adopted in the spectral baseline correction method based on DE algorithm and non-uniform B-spline curve provided by the present invention.

[0028] Figure 3 for Figure 2 The test sample spectral signal (filtered spectral signal) is obtained after the spectral signal is filtered.

[0029] Figure 4 These are the spectral signal waveforms before and after the filtered spectral baseline correction.

[0030] Figure 5 It is the minimum value diagram of the spectral signal before and after the baseline correction of the filtered spectrum.

[0031] Figure 6 This is a flow chart of the differential evolution algorithm adopted by the present invention.

[0032] Figure 7 is the distribution diagram of the control point sequence C.

[0033] Figure 8 is the distribution graph of the internal node sequence T.

[0034] Fig. 9 It is the waveform diagram of the test sample spectrum (original spectrum), the fitted baseline and the spectrum curve after baseline correction.

[0035] in, Figure 2 , Figure 3 , Figure 4 , Figure 5 , Figure 7 , Fig. 9 The horizontal axis is the wavelength band, Figure 2 , Figure 3 , Figure 4 , Figure 5 , Figure 7 , Fig. 9 The ordinate is the normalized spectral intensity value; Figure 8 The horizontal axis is the i-th T node, and the vertical axis is the spectral intensity value. DETAILED DESCRIPTION

[0036] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0037] The spectral baseline correction method provided by the present invention is now described in conjunction with a specific embodiment. The flow chart is as follows: Figure 1 As shown:

[0038] Step 101: Eliminate the noise signal in the spectrum signal to obtain a filtered spectrum curve.

[0039] During the ICP-AES measurement process, affected by stray light, the spectral signal contains a large amount of noise signals, which seriously affects the selection of trough points during baseline fitting. Therefore, preprocessing is required to eliminate noise before baseline fitting. The technical solution of the present invention first uses Gaussian filtering to preprocess the ICP-AES spectrum.

[0040] Gaussian filtering uses discrete convolution operation of the original spectral signal and the Gaussian convolution kernel, as shown in formula (1):

[0041]

[0042] Among them, f is the original spectral signal, and g is the Gaussian convolution kernel. Since the ICP-AES spectral signal basically obeys the Gaussian distribution, convolution through the Gaussian convolution kernel can retain the spectral line information as much as possible and avoid reducing the spectral resolution.

[0043] The spectrum of 294.547nm-297.400nm is selected as the original sample and normalized, such as Figure 2 As shown; the original sample is Gaussian filtered to eliminate high-frequency noise to obtain the filtered spectral signal, which is used as the final baseline correction test sample, as shown Figure 3 shown.

[0044] Step 102: Establish a spectrum baseline mathematical model.

[0045] Record the trough points of the filtered spectrum curve obtained in step 101, and record these trough points to obtain a sequence set data point sequence set X, X = {(x 1|u , x 1|v ), (x 2|u , x 2|v ), (x 3|u , x 3|v ),…,(x n|u , x n|v )}, the data point sequence X is divided by node symbols (the node symbols are between the maximum horizontal coordinate value +10 and the minimum horizontal coordinate value -10 of the original spectrum), and the node sequence set is recorded as the inner node sequence T (its length is determined by the length of the filtered spectrum curve), T = [t -k+1 , t -k+2 ,…,t -1 , t0, t1, …, t n , t n+1 , t n+2 ,…,t n+k ], after the division, each node in the node sequence T in the node sequence set satisfies the following relationship:

[0046] t k+1 ≤t -k+2≤…≤t -1 ≤t0<t1<…<t n <t n+1 ≤t n+2 <…<t n+k (2)

[0047] Specifically, u is the horizontal coordinate, v is the vertical coordinate, t1 to t n are called inner nodes, and the rest are called outer nodes. n is the number of inner nodes, and k is the order of the B-spline curve.

[0048] Furthermore, for a k-order B-spline curve, its expression is shown in formula (3).

[0049]

[0050] Where C = [c0, c1, ..., c n ] is the B-spline curve control point sequence (its length is determined by the length of the filtered spectrum curve), c is the B-spline curve control point coordinate, y is the B-spline curve coordinate point sequence, N i k (x) is the B-spline basis function, which is defined as shown in formula (4):

[0051]

[0052] In the subsequent calculation process, the control point sequence C and the curve coordinate point sequence y are both calculated in the form of coordinates, so the formula (3) of the spectral baseline mathematical model can also be written as:

[0053]

[0054] Similarly, let the control point sequence be C = {(c 0|u , c 0|v ), (c 1|u , c 1|v ),…,(c n|u , c n|v )}.

[0055] The non-uniform B-spline curve used in this step is often used in the construction of free curves and surfaces, and has good low-order smoothness, so it is also suitable for fitting spectral baselines.

[0056] Step 103: Establish a spectral baseline correction evaluation function.

[0057] Assume that the spectral intensity value of the filtered spectral curve obeys the model I'(x), the spectral intensity value after baseline correction obeys the model I(x) (the spectral intensity of the corrected spectral model), and the spectral baseline model is B(C, T); since the superposition of spectral intensity values ​​can be regarded as linear superposition, the following relationship is satisfied:

[0058] I(x)=I′(x)-B(C,T) (6)

[0059] Assume that the minimum value sequence set in the filtered spectral curve is P:

[0060] P={(p0,I'(p0)), (p1,I'(p1)),...,(p n , I'(p n ))}, and record P X ={p0, p1, ..., p n}, P Y ={I'(p0), I'(p1),...,I'(p n )}. X is the abscissa set of the minimum value sequence set P, P Y is the ordinate set of the minimum sequence set P. Subtract the non-uniform B-spline curve from the filtered spectral curve (baseline correction) to obtain a set of corrected intensity values. Specifically, the minimum sequence set P X The corresponding correction intensity value set is I Y , I Y = {I(p0), I(p1), ..., I(p n )}. The difference in the distribution of the spectral signal and its minimum value points before and after the baseline correction of the filtered spectrum is shown in Figure 4 and Figure 5 shown.

[0061] With the correction intensity value set I Y The standard deviation of is used as the measurement criterion, and the evaluation function is established by calculating the standard deviation of the set, as shown in formula (7).

[0062]

[0063] Among them, y u and v The definition is the same as that in Formula 5, both represent the horizontal and vertical coordinates of the NURBS curve, and STD represents the standard deviation of the calculation set. Therefore, the spectral baseline correction problem is transformed into determining a set of appropriate C and T values ​​so that the F(C, T) value is minimized, that is, the problem of global optimization for F(C, T).

[0064] Step 104: Find the optimal solution of the evaluation function by using a differential evolution algorithm.

[0065] The differential evolution algorithm includes four steps: population initialization, mutation, crossover and selection. The flowchart is as follows: Figure 6As shown. After the population is initialized, the mutation, crossover and selection operations are performed cyclically until the maximum evolutionary generation is reached; in each round of evolution, the mutation operation is performed on all individual vectors of the population to obtain the mutation vector; then the mutation vector is cross-operated with the individual vector to obtain the crossover vector; finally, the fitness of the crossover vector and the individual vector is compared, and the vector with the smaller fitness is selected as the individual vector of the next generation population. The differential evolution algorithm process is as follows Figure 5 shown.

[0066] Step 1041: Initialize the population.

[0067] The control point sequence C and the internal node sequence T in the non-uniform B-spline are used as the population individual vectors of the differential evolution algorithm. The initial population should cover the entire search space as much as possible, so the maximum individual vector is set before the population is initialized. and the minimum individual vector Where D is the dimension of the individual vector; NP vectors are generated by initialization to represent all individuals in the population;

[0068] The i-th individual vector in the G-th generation population is expressed as The generation rule of the jth parameter in the individual vector is shown in formula (8).

[0069]

[0070] Step 1402: Perform individual fitness evaluation.

[0071] All individual vectors X in the current evolutionary generation i，G Substitute the fitness function F(·) into the output result F(X i，G ) is the fitness value of the individual. The smaller the fitness value, the better the corresponding individual. In the present invention, the spectrum baseline correction evaluation function F(C, T) is regarded as the fitness function of the differential evolution algorithm. Step 1402 is also executed to record the number of evolutions and determine whether the number of evolutions reaches a preset maximum value. If so, the optimal individual screened from the total number of evolutions is output, otherwise the subsequent steps are continued.

[0072] Step 1403: Perform mutation operation on all current individual vectors.

[0073] In each round of evolution, for all individual vectors X under the current evolutionary generation G i，G Perform mutation operation to obtain mutation vector V i，G ,

[0074] Commonly used mutation operators are DE / rand / 1 / bin and DE / best / 1 / bin, the former has stronger global search capability, while the latter has stronger local optimization capability. The present invention adopts a mutation rule combining the DE / rand / 1 / bin operator and the DE / best / 1 / bin operator. The DE / rand / 1 / bin operator is used in the early stage of evolution to quickly screen high-quality individuals by utilizing the global search characteristics, and the DE / best / 1 / bin operator is used in the later stage to quickly converge by utilizing the local optimization capability to obtain the optimal solution. The DE / rand / 1 / bin operator expression is shown in formula (9), and the DE / best / 1 / bin operator expression is shown in formula (10).

[0075]

[0076]

[0077] in, and are three random individual vectors in the population in the current evolutionary generation, and the subscript X best，G is the optimal individual vector in the G-th generation population, and F is a constant in [0, 1], representing the scaling factor.

[0078] Step 1404: cross-exchange the individual vectors and their corresponding mutation vectors.

[0079] All individual vectors X i，G And the corresponding mutation vector V i，G The information components of cross exchange, get the cross vector U i，G , The crossover rule is shown in formula (11):

[0080]

[0081] Among them, CR is the crossover probability, j rand is a random integer in [1, D]. When CR takes a larger value, the algorithm has a stronger local search capability, but it is also prone to premature convergence; when CR takes a smaller value, the evolutionary stability is higher and the algorithm has a stronger global optimization capability. In the technical solution of the present invention, CR will be assigned using the gradient increasing method, that is, a smaller CR value is used in the early stage to obtain a stronger global optimization capability, and a larger CR value is used in the later stage to increase the local search capability of the algorithm. The specific CR value rule is shown in formula (12).

[0082]

[0083] Among them, G * is the current evolutionary generation in the evolutionary process, CRmax and CR min They respectively represent the value range of CR, and usually the value range of CR is (o, 1).

[0084] Step 1045: Select the vector with the smaller fitness from the individual vector and the cross vector as the new individual vector.

[0085] All individual vectors X i，G and the cross vector U i，G Substitute into the fitness function for comparison and select the vector with smaller fitness as the new individual vector for the next generation of evolution, as shown in formula (13).

[0086]

[0087] Repeat steps 1042 to 1045, and then evolve multiple individual vectors in the population until the maximum evolutionary generation is reached, and the fitness value of each individual in the population is obtained. At this time, the individual vector with the smallest fitness value among the individuals in the population is the best feature vector, and the best feature vector contains the characteristic parameters of the spectral baseline mathematical model. The vector with the best fitness is output as the best feature vector, as the optimal solution of the spectral baseline mathematical model.

[0088] Step 105: Obtain a corrected spectrum curve according to the optimal solution.

[0089] The optimal solution is decomposed into a control point sequence C and an internal node sequence T in a non-uniform B-spline.

[0090] Step 105 interprets the optimal solution as a vector of length 1*(s+h+q), where the lengths of s and h represent the values ​​of the horizontal and vertical coordinates of the control point C, respectively, and the length of q represents the internal node sequence T.

[0091] Specifically, the optimal solution sequence is recorded as U = {C x , C y , T}, where C x is the set of horizontal coordinates of each point in the control point sequence C, C x ={c x1 , c x2 ...c xn}, C y is the ordinate set C of each point in the control point sequence C y ={c y1 , c y2 ...c yn}, T is the internal node sequence.

[0092] Specific as Figure 7 and Figure 8As shown, according to the result obtained in step 105, the spectral baseline model formula (3) in step 102 is used to fit the corresponding non-uniform B-spline curve as the spectral baseline, and then the corrected spectral curve is obtained using formula (6). The original spectral baseline is corrected as shown in Fig. 9 shown.

[0093] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein by equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A spectral baseline correction method based on DE algorithm and non-uniform B-spline curve, characterized in that: The method comprises: Gaussian filtering is used to eliminate the noise signal in the original spectral signal to obtain the filtered spectral curve; The spectrum baseline model is obtained by dividing the trough points of the filtered spectrum curve; A spectrum baseline correction evaluation function F(C, T) is established according to the spectrum intensity value of the filtered spectrum curve and the spectrum intensity value of the filtered spectrum curve after baseline correction: Where p is the minimum value sequence set in the original spectral signal, and STD represents the standard deviation of the calculated set; The optimal solution of the spectral baseline evaluation function is obtained, the optimal solution is decomposed into an optimal control point vector and an optimal internal node sequence, the optimal control point vector and the optimal internal node sequence are substituted into the spectral baseline mathematical model, and a corrected spectral curve is obtained.

2. The method according to claim 1, characterized in that The method of eliminating noise signals in the spectral signal by using Gaussian filtering includes: Perform discrete convolution operation on the original spectral signal and Gaussian convolution kernel: Among them, f is the original spectral signal, g is the Gaussian convolution kernel, n is the nth point of the convolved signal, τ is the offset, and f is the original spectral curve.

3. The method according to claim 1, characterized in that The method of obtaining a non-uniform B-spline curve by dividing the trough points of the filtered spectrum curve includes: The trough points of the filtered spectral curve are recorded to obtain a sequence set X, and the data point sequence X is divided by the node symbol t to obtain an inner node sequence T to obtain a non-uniform B-spline curve. The expression of the non-uniform B-spline curve is the spectral baseline model B(C, T), which is expressed as follows: Where C = [c0, c1, ..., c n ] is the B-spline curve control point sequence, c is the B-spline curve control point, c i is the i-th control point in the B-spline control point sequence, y is the B-spline curve coordinate point sequence, is the B-spline basis function.

4. The method according to claim 3, characterized in that The trough points of the filtered spectrum curve are recorded to obtain a sequence set X, and the data point sequence X is divided by node symbols to obtain an inner node sequence T, including: X={(x1| u ,x1| v ),(x2| u ,x2| v ),(x3| u ,x3| v ),…,(x n | u ,x n | v )}, u is the horizontal coordinate of the trough point, v is the vertical coordinate of the trough point; After the node sequence X is divided into the inner node sequence T, T = [t -k+1 ,t -k+2 ,…,t -1 ,t0,t1,…,t n ,t n+1 ,t n+2 ,…,t n+k ], each node in the node sequence set T satisfies the following relationship: t -k+1 ≤t -k+2 ≤…≤t -1 ≤t0<t1<…<t n <t n+1 ≤t n+2 <…<t n+k ; Among them, t1 to t n are called inner nodes, and the rest are called outer nodes. n is the number of inner nodes, and k is the order of the B-spline curve.

5. The method according to claim 1, characterized in that The method of establishing a spectral baseline correction evaluation function F(C, T) according to the spectral intensity value of the filtered spectral curve and the spectral intensity value of the filtered spectral curve after baseline correction comprises: Assume that the spectral intensity value of the filtered spectral curve obeys the model I'(x), the spectral intensity value of the filtered spectral curve after baseline correction obeys the model I(x), the baseline model is B(C, T), and the superposition of the spectral intensity values ​​is a linear superposition, then the following formula is obtained: I(x) = I'(x) - B(C, T), According to the non-uniform B-spline curve, a correction intensity value set I is obtained. Y , with the correction intensity value set I Y The standard deviation of is used as the measurement criterion, and the spectral baseline evaluation function F(C,T) is established.

6. The method according to claim 5, characterized in that The correction intensity value set I is obtained according to the non-uniform B-spline curve Y ,include: Assume that the minimum value sequence set in the filtered spectral curve is P: P={(p0,I'(p0)),(p1,I'(p1)),…,(p n ,I'(p n ))}, and record P X ={p0,p1,…,p n }, P Y = {I'(p0),I'(p1),…,I'(p n )}, where P X is the horizontal coordinate value of the minimum value sequence of the spectrum after filtering, P Y is the ordinate value of the spectral minimum sequence after filtering; After baseline correction is performed on the filtered spectral curve, the set P is obtained. X The corresponding correction intensity value set I Y , I Y = {I(p0),I(p1),…,I(p n )}.

7. The method according to claim 3, characterized in that Obtaining the optimal solution of the spectral baseline evaluation function includes: a. The control point sequence C and the inner node sequence T in the non-uniform B-spline are used as population individual vectors of the differential evolution algorithm, and the population is initialized; b. Substituting several individual vectors obtained after the population is initialized into the spectral baseline correction evaluation function to obtain the evaluation value of each individual vector; c. In each round of evolution iteration, a mutation operation is performed on all individual vectors in the current iteration to obtain mutation vectors corresponding to each individual vector; d. Cross-exchange the information components of each individual vector and the corresponding mutation vector to obtain a plurality of cross-vectors; e Substitute all individual vectors and crossover vectors into the fitness function for comparison, and the vectors with small fitness are brought into the next round of differential evolution algorithm.

8. The method according to claim 7, characterized in that The step e also includes that after the last iteration, the individual vector with the smallest fitness in the population is the optimal vector of the spectral baseline evaluation function.

9. The method according to claim 8, characterized in that After decomposing the optimal vector into the optimal control point vector C and the optimal inner node sequence T, the spectral baseline model B(C, T) is introduced to obtain a new non-uniform B-spline curve; The new non-uniform B-spline curve is used as the current spectral baseline, combined with the formula: I(x)=I′(x)-B(C, T), and the corrected spectral intensity value obeys the model I(x), thereby obtaining a corrected spectral curve.

Images