A method for enhancing spectral sensitivity based on the correlation characteristics of nuclear magnetic resonance signals

Abstract

The present invention discloses a spectrum sensitivity enhancement method based on the correlation characteristics of nuclear magnetic resonance signals, comprising preprocessing multiple original NMR spectra to obtain an initial two-dimensional matrix; matching the initial two-dimensional matrix with correlation data blocks and obtaining a basic estimated spectrum using three-dimensional transformation, frequency domain hard threshold filtering, and first fusion weighting; matching the correlation data blocks with the basic estimated spectrum and guiding the initial two-dimensional matrix to re-match the correlation data blocks; obtaining a spatial three-dimensional correlation data group after Wiener collaborative filtering through three-dimensional transformation and Wiener collaborative filtering; and performing a second fusion weighting based on the spatial three-dimensional correlation data group after Wiener collaborative filtering to obtain a final NMR spectrum with enhanced sensitivity. The present invention can reduce pseudo-signal oscillations around the edges and avoid the classical ringing effect, while also effectively extracting and retaining the characteristic line shape of the nuclear magnetic resonance spectrum, thereby significantly improving the sensitivity of the NMR spectrum.

Description

Technical Field

[0001] The present invention relates to the field of nuclear magnetic resonance spectrum analysis, and in particular to a spectrum sensitivity enhancement method based on nuclear magnetic resonance signal correlation characteristics. Background Art

[0002] Nuclear magnetic resonance (NMR) spectroscopy plays an indispensable role in chemistry, biology, and life sciences, with applications encompassing structural analysis, quantitative analysis, reaction monitoring, and many other areas. However, compared to spectroscopic techniques, NMR spectroscopy often suffers from low sensitivity, particularly for nuclei with low natural abundance and low gyromagnetic ratio. Noise interference is unavoidable during the acquisition and transmission of NMR signals, with thermal noise generated by the detection circuit being the primary noise source. The signal-to-noise ratio is proportional to the square root of the average number of scans, meaning that obtaining a spectrum with sufficient sensitivity typically requires a long acquisition time to average multiple signals. However, in practical applications, multiple acquisitions are often difficult to implement due to time and cost constraints. Therefore, effectively improving sensitivity within a limited number of scans is particularly important. For solid samples, the presence of large anisotropic interactions, such as chemical shift anisotropy, dipole interactions, and nuclear quadrupole interactions, results in significant broadening of the corresponding NMR resonance peaks, posing additional challenges to signal detection and attribution. To overcome the limitations of low NMR spectroscopy sensitivity, researchers are constantly exploring and developing various methods, including increasing magnetic field strength, improving probe design, optimizing pulse sequences, developing novel spectral analysis methods (such as sparse sampling and compressed sensing), and hyperpolarization enhancement techniques (such as dynamic NMR polarization) to improve the sensitivity of NMR signals per unit time. In addition to the aforementioned methods, the sensitivity of NMR spectroscopy can also be improved by using different algorithms, such as singular value decomposition (SDD) using the exponential characteristics of NMR spectroscopy signals, Cadzow filtering, wavelet transforms, principal component analysis, multivariate curve decomposition, and sparse convex wavelet clustering. However, wavelet transforms are less effective in improving NMR spectroscopy sensitivity, and Cadzow filtering can produce artifacts in signal-free regions. Sparse convex wavelet clustering methods can obtain paramagnetic NMR spectra without phase distortion, but they lose quantitative information about peak intensities. Each of these methods has its advantages and disadvantages, and in practical applications, the choice is usually based on the specific situation. Summary of the Invention

[0003] The purpose of the present invention is to overcome the shortcomings and deficiencies of the prior art, such as improper retention of characteristic line shapes of nuclear magnetic resonance (NMR) spectra and oscillation of false signals around the edges, and to propose a spectrum sensitivity enhancement method based on the correlation characteristics of nuclear magnetic resonance signals.

[0004] The above-mentioned purpose of the present invention is achieved by the following technical means:

[0005] A method for enhancing spectral sensitivity based on nuclear magnetic resonance signal correlation characteristics comprises the following steps:

[0006] Step 1: Perform multiple independent samplings on the same sample to obtain the corresponding original NMR spectra;

[0007] Step 2: construct an initial two-dimensional matrix using the original NMR spectrum;

[0008] Step 3: Set a matching window. The matching window traverses the initial two-dimensional matrix row by row, searches for data corresponding to the matching window at each position in the initial two-dimensional matrix, and obtains the corresponding correlation data block S.

[0009] Step 4: Sort and stack the associated data blocks S into a three-dimensional array M(P);

[0010] Step 5: Perform a three-dimensional transformation and a frequency-domain hard threshold filter on the three-dimensional array M(P) in sequence to obtain a frequency-domain three-dimensional correlation data set after the frequency-domain hard threshold filter. Perform a three-dimensional inverse transformation on the frequency-domain three-dimensional correlation data set after the frequency-domain hard threshold filter to obtain a spatial-domain three-dimensional correlation data set after the frequency-domain hard threshold filter. The three-dimensional transformation is used to achieve a sparse expression effect.

[0011] Step 6: Use the spatial domain three-dimensional correlation data group after frequency domain hard threshold filtering to perform the first fusion weighting to obtain a basic estimated spectrum;

[0012] Step 7: The matching window traverses the basic estimation map row by row, searches for the data corresponding to the matching window at each position in the basic estimation map, and obtains the corresponding correlation data block S′;

[0013] Step 8: Sort and stack the associated data blocks S′ to obtain a three-dimensional array M basic (P′);

[0014] Step 9: Re-stack the associated data blocks S to obtain a three-dimensional array M origin (P), the order in which the associative data blocks S are re-stacked and the three-dimensional array M basic The order in which the associative data blocks S′ are stacked in (P′) is the same;

[0015] Step 10: For the three-dimensional array M origin (P) and the three-dimensional array M basic (P′) performs three-dimensional transformation respectively, and obtains the corresponding three-dimensional frequency domain array and a three-dimensional frequency domain array Then use the three-dimensional frequency domain array As a guide to the 3D frequency domain array Performing Wiener collaborative filtering to obtain a frequency domain three-dimensional correlation data group after Wiener collaborative filtering, and performing a three-dimensional inverse transformation on the frequency domain three-dimensional correlation data group after Wiener collaborative filtering to obtain a spatial domain three-dimensional correlation data group after Wiener collaborative filtering;

[0016] Step 11: Perform a second fusion weighting on the spatial three-dimensional correlation data set after Wiener collaborative filtering to obtain a second fused spectrum, and extract the final NMR spectrum with enhanced sensitivity based on the second fused spectrum.

[0017] The above-mentioned step 2 specifically includes the following steps:

[0018] Step 2.1: Determine whether the original NMR spectrum type is one-dimensional or two-dimensional. If the original NMR spectrum type is one-dimensional, proceed to step 2.2; if the original NMR spectrum type is two-dimensional, proceed to step 2.3.

[0019] Step 2.2, concatenate the original one-dimensional NMR spectra by column to construct an initial two-dimensional matrix;

[0020] Step 2.3, sequentially reading the real part data and the imaginary part data of each two-dimensional original NMR spectrum, and rearranging the read real part data and the imaginary part data according to the original time domain sampling point number to obtain a rearranged NMR spectrum, wherein the rearranged NMR spectrum includes a real part data matrix and an imaginary part data matrix;

[0021] The real data matrix and the imaginary data matrix in each rearranged NMR spectrum are added together to obtain the corresponding two-dimensional complex matrix;

[0022] Sum all two-dimensional complex matrices to obtain an initial two-dimensional matrix.

[0023] As described above, both step 3 and step 7 include the following steps:

[0024] Set the matching window size n hard ×n hard , n hard Indicates the side length of the matching window, and the minimum value of the length and width of the initial two-dimensional matrix obtained in step 2 is recorded as min, n hard ≤min; the matching window is traversed row by row in the object matrix at a set step;

[0025] For each matching window, select data of a part of the continuous area in the object matrix corresponding to the matching window as a reference data block, and obtain the associated data blocks of each reference data block from the data of the object matrix corresponding to the matching window;

[0026] In step 3, the object matrix takes the initial two-dimensional matrix, the corresponding reference data block is recorded as P, and the correlation data block is recorded as S;

[0027] In step 7, the object matrix takes the basic estimation map, the corresponding reference data block is recorded as P′, and the correlation data block is recorded as S′.

[0028] Obtaining the associated data blocks of each of the above-mentioned reference data blocks specifically includes the following steps:

[0029] For each matching window, the center of the reference data block is set as the center of the corresponding matching window, and the size of the reference data block is k hard ×k hard , k hard is the side length of the benchmark data block, k hard <n hard ;

[0030] Set the scanning window, the size of the scanning window is k hard ×k hard The scanning window scans row by row in the range of the object matrix corresponding to each matching window, and the data in the object matrix corresponding to each scanning window is the data block to be matched;

[0031] For each matching window, calculate the normalized Euclidean distance between the reference data block and each to-be-matched data block in the corresponding matching window;

[0032] For each reference data block, the data blocks to be matched whose normalized Euclidean distance is less than or equal to the distance threshold are used as the associated data blocks of the reference data block, and the set of associated data blocks is the associated data set;

[0033] In step 3, the data block to be matched is recorded as Q, and the distance threshold is recorded as τ hard ,The set of related data is denoted as R(P);

[0034] In step 7, the data block to be matched is recorded as Q′, and the distance threshold is recorded as τ wien , the associated data set is recorded as R(P′).

[0035] As described above, both step 4 and step 8 include the following steps:

[0036] For each reference data block, the associated data blocks of the reference data block are sorted in ascending order according to the normalized Euclidean distance from the associated data block to the corresponding reference data block and stacked to obtain a corresponding three-dimensional array;

[0037] In step 4, the reference data block is P, the associated data block is S, and the three-dimensional array is denoted as M(P);

[0038] In step 8, the reference data block is P', the associated data block is S', and the three-dimensional array is M basic (P′).

[0039] As mentioned above, the normalized Euclidean distance is calculated by the following formula:

[0040]

[0041] d(A,B) represents the normalized Euclidean distance between data block A and data block B, ‖.‖2 is the L2 distance operator; γ′(·) is the data block hard threshold operator; in step 3, data block A takes the reference data block P, and data block B takes the to-be-matched data block Q; in step 7, data block A takes the reference data block P′, and data block B takes the to-be-matched data block Q′;

[0042] The data block hard threshold operator γ′(·) is calculated according to the following formula:

[0043]

[0044]

[0045] y represents the data block, x y represents the pixels in the data block y, γ y ′(.) represents the pixel hard threshold operator for the pixels in the data block y, V xy Represents pixel x in data block y y For values ​​in the object matrix, σ y represents the noise intensity of the data block, is the hard threshold, which is a fixed value. represents the hard threshold coefficient; σ T is the noise intensity threshold, the noise intensity threshold σ T Determined by the noise intensity of the corresponding entire object matrix;

[0046] In step 4, the object matrix takes the initial two-dimensional matrix;

[0047] In step 8, the object matrix takes the basic estimated map.

[0048] The first fusion weighting performed in step 6 and the second fusion weighting performed in step 11 as described above both include calculating a basic estimate for each pixel x, and arranging the basic estimate for each pixel x according to the pixel coordinates of the corresponding pixel x in the initial two-dimensional matrix;

[0049] The basic estimate is given by:

[0050]

[0051] in,

[0052]

[0053] u(x) represents the basic estimate, u Q,P (x) is the value corresponding to pixel x in the filtered data block; w p is the fusion weight; h Q (x) is the pixel x membership coefficient, which is used to indicate whether the pixel x belongs to the filtered data block; N p is the number of non-zero frequency domain coefficients retained in the filtered frequency domain three-dimensional correlation data set;

[0054] In step 5, The filtered data block is the correlation data block S after the frequency domain hard threshold filtering h ; is the correlation data block S of pixel x after hard threshold filtering in the frequency domain h The corresponding value in ; is the number of non-zero frequency domain coefficients retained in the frequency domain three-dimensional correlation data group after frequency domain hard threshold filtering; the data of each layer in the spatial domain three-dimensional correlation data group after frequency domain hard threshold filtering is recorded as the correlation data block S after frequency domain hard threshold filtering h , the correlation data block S after frequency domain hard threshold filtering h One-to-one correspondence with the association data block S; the basic estimate of the calculation is recorded as u basic (x), the basic estimate u basic (x) Arrange the pixel coordinates of the corresponding pixel x in the initial two-dimensional matrix to obtain a basic estimated map;

[0055] In step 11, The filtered data block is the correlation data block S after Wiener collaborative filtering w ; is the correlation data block S of pixel x after Wiener collaborative filtering w The corresponding value in ; is the number of non-zero frequency domain coefficients retained in the frequency domain three-dimensional correlation data group after Wiener collaborative filtering; the data of each layer in the spatial domain three-dimensional correlation data group after Wiener collaborative filtering is recorded as the correlation data block S after Wiener collaborative filtering w , the correlation data block S after Wiener collaborative filtering w One-to-one correspondence with the association data block S; the basic estimate of the calculation is recorded as u origin (x), the basic estimate u of each pixel x origin (x) Arrange according to the pixel coordinates of the corresponding pixel x in the initial two-dimensional matrix to obtain the second fusion map.

[0056] Depending on the type of the original NMR spectrum collected in step 1, step 11 described above, extracting the final NMR spectrum with enhanced sensitivity based on the spectrum after the second fusion, includes the following steps:

[0057] For the original two-dimensional NMR spectrum, the spectrum after the second fusion is the final NMR spectrum with enhanced sensitivity;

[0058] For a one-dimensional original NMR spectrum, the final NMR spectrum after sensitivity enhancement is: any column in the spectrum after the second fusion, or the sum of the columns of data in the spectrum after the second fusion divided by the number of columns.

[0059] The three-dimensional transformation described above includes one or more of the following transformations in sequence:

[0060] Discrete cosine transform, Walsh-Hadamard transform, Fourier transform and wavelet transform.

[0061] In step 4 as described above, the noise intensity of the initial two-dimensional matrix is ​​equal to the noise intensity of a single original NMR spectrum.

[0062] The present invention has the following advantages and positive effects:

[0063] 1. It can greatly improve the sensitivity of nuclear magnetic resonance NMR spectrum;

[0064] 2. By using the basic estimated spectrum as a guide, Wiener collaborative filtering is performed on the correlation data blocks obtained from the original NMR spectrum, which can effectively extract and preserve the characteristic line shape of the original NMR spectrum;

[0065] 3. Through a special fusion weighting method, the oscillation of false signals around the edges is reduced, avoiding the classic ringing effect observed with the transformation threshold method. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 This is a flow chart of a method for enhancing spectral sensitivity based on the correlation characteristics of nuclear magnetic resonance signals;

[0067] Figure 2 It is the original one-dimensional NMR spectrum without sensitivity enhancement;

[0068] Figure 3 It is a one-dimensional NMR spectrum with enhanced sensitivity after being processed by a spectrum sensitivity enhancement method based on the correlation characteristics of nuclear magnetic resonance signals proposed in the present invention;

[0069] Figure 4 It is the original two-dimensional NMR spectrum without sensitivity enhancement;

[0070] Figure 5It is a two-dimensional NMR spectrum with enhanced sensitivity after being processed using a spectrum sensitivity enhancement method based on the correlation characteristics of nuclear magnetic resonance signals proposed in the present invention.

[0071] Wherein, δ represents chemical shift, i.e. the abscissa of the nuclear magnetic resonance spectrum; ppm stands for part per million, which is the unit of chemical shift. DETAILED DESCRIPTION

[0072] In order to facilitate those skilled in the art to understand and implement the present invention, the present invention is further described in detail below with reference to examples. The implementation examples described herein are only used to illustrate and explain the present invention and are not used to limit the present invention.

[0073] Example 1:

[0074] like Figure 1 As shown, a method for enhancing spectrum sensitivity based on the correlation characteristics of nuclear magnetic resonance signals comprises the following steps:

[0075] Step 1: The same sample is sampled multiple times independently to obtain the corresponding original NMR spectra (original nuclear magnetic resonance spectra). The types of the original NMR spectra corresponding to the same sample are consistent.

[0076] In this embodiment, metal organic framework material MIL-53 (Al) and aluminum isopropoxide are selected as samples. 27 Al QCPMG (Quadrupolar Carr–Purcell–Meiboom–Gill) nuclear magnetic resonance experiment, the original NMR spectrum obtained is 27 Al QCPMG spectrum, 27 The Al QCPMG spectrum is one-dimensional; the Al isopropoxide is independently 1 H- 13 CHETCOR NMR experiment, the original NMR spectrum obtained is 1 H- 13 C heteronuclear chemical shift correlation (HETCOR) spectrum, 1 H- 13 The type of C HETCOR map is two-dimensional.

[0077] Step 2: Preprocessing: construct an initial two-dimensional matrix using all the original NMR spectra corresponding to the same sample obtained in step 1. This includes the following steps:

[0078] Step 2.1: Determine whether the type of the original NMR spectrum obtained is one-dimensional or two-dimensional. If the type of the original NMR spectrum obtained is one-dimensional, use step 2.2 to preprocess the original NMR spectrum; if the type of the original NMR spectrum obtained is two-dimensional, use step 2.3 to preprocess the original NMR spectrum.

[0079] Step 2.2: Preprocess the original one-dimensional NMR spectrum:

[0080] Each one-dimensional original NMR spectrum is spliced ​​column by column to construct an initial two-dimensional matrix.

[0081] Step 2.3: Preprocess the original two-dimensional NMR spectrum:

[0082] Step 2.3.1: The two-dimensional original NMR spectrum includes real data and imaginary data, and the real data and the imaginary data are read separately, so the two-dimensional original NMR spectrum needs to be rearranged into the matrix format during the original time domain sampling. Specifically, the following steps are included: the real data and the imaginary data of each two-dimensional original NMR spectrum are read in sequence, and the read real data and imaginary data are rearranged according to the original time domain sampling point number to obtain a rearranged NMR spectrum, the rearranged NMR spectrum is the matrix format during the initial acquisition, and the rearranged NMR spectrum includes a real data matrix and an imaginary data matrix, the number of rows of the real data matrix is ​​equal to the number of rows of the imaginary data matrix, and the number of columns of the real data matrix is ​​equal to the number of columns of the imaginary data matrix. In this embodiment, there are a total of 16 two-dimensional original NMR spectra. The real data and imaginary data of each two-dimensional original NMR spectrum are read and rearranged respectively, and the corresponding rearranged NMR spectra include a real data matrix of size 1024*128 and an imaginary data matrix of size 1024*128.

[0083] Step 2.3.2: Add the real data matrices and the imaginary data matrices of each rearranged NMR spectrum obtained in step 2.3.1 to obtain corresponding two-dimensional complex matrices. In this embodiment, the real data matrices and the corresponding imaginary data matrices of each rearranged NMR spectrum obtained in step 2.3.1 are added to obtain a corresponding two-dimensional complex matrix of type complex double with a size of 1024*128, and the number of two-dimensional complex matrices is 16.

[0084] Step 2.3.3: Sum all the obtained two-dimensional complex matrices to obtain an initial two-dimensional matrix. In this embodiment, 16 two-dimensional complex matrices are summed to obtain an initial two-dimensional matrix of size 1024*128.

[0085] Step 3: Matching relevance data blocks: Set a matching window, traverse the initial two-dimensional matrix row by row, search for data corresponding to the matching window at each position in the initial two-dimensional matrix, and obtain the corresponding relevance data blocks S (where the matching relevance data blocks S are in the form of a two-dimensional matrix). Specifically, the following steps are included:

[0086] Step 3.1, set the matching window size n hard ×n hard , n hard Indicates the side length of the matching window (in pixels), and the minimum value of the length and width of the initial two-dimensional matrix obtained in step 2 is recorded as min, n hard ≤min; the matching window is traversed row by row in the initial two-dimensional matrix obtained in step 2 with the set step;

[0087] Step 3.2: For each matching window, select a portion of continuous data in the initial two-dimensional matrix corresponding to the matching window as a reference data block P. Then, obtain the associated data blocks S of the reference data block P from the data in the initial two-dimensional matrix corresponding to the matching window. There are multiple associated data blocks S corresponding to each reference data block P. The specific steps are as follows:

[0088] Step 3.2.1: For each matching window, set the corresponding reference data block P. The center of the reference data block P is the center of the corresponding matching window. The size of the reference data block P is k. hard ×k hard , k hard is the side length of the reference data block P (k hard <n hard );

[0089] Step 3.2.2, set the scanning window, the size of the scanning window is k hard ×k hard The scanning window scans row by row in the initial two-dimensional matrix corresponding to each matching window. The data in the initial two-dimensional matrix corresponding to each scanning window is recorded as a to-be-matched data block Q. Each matching window corresponds to multiple to-be-matched data blocks Q, and some data between the to-be-matched data blocks Q may overlap.

[0090] Step 3.2.3: For each matching window, calculate the normalized Euclidean distance d(P, Q) between the reference data block P and each to-be-matched data block Q in the corresponding matching window.

[0091] To improve computational efficiency, the normalized Euclidean distance d(P,Q) between the reference data block P and the data block to be matched Q is calculated using the following formula:

[0092]

[0093] d(A,B) represents the normalized Euclidean distance between data blocks A and B, ‖.‖2 is the L2 distance operator, and γ′(·) is the hard threshold operator. In step 3, data block A takes the reference data block P, and data block B takes the to-be-matched data block Q. Then the normalized Euclidean distance d(P,Q) is:

[0094]

[0095] The data block hard threshold operator γ′(·) is calculated according to the following formula:

[0096]

[0097]

[0098] y represents the data block, x y represents the pixels in the data block y, γ y ′(.) represents the pixel hard threshold operator for the pixels in the data block y, Represents pixel x in data block y y Corresponding to the value in the initial two-dimensional matrix, σ y represents the noise intensity of the data block, σ y The larger the value is, the greater the position noise intensity in the original NMR spectrum corresponding to the data block is. T is the noise intensity threshold, is the hard threshold, which is a fixed value set; Represents the hard threshold coefficient that is set.

[0099] In step 3, the data block noise intensity σ y The noise intensity is calculated from the data blocks (i.e., the reference data block P and the data block to be matched Q) divided by the initial two-dimensional matrix obtained in step 2.3.3; the noise intensity of the initial two-dimensional matrix is ​​determined by a series of original NMR spectra used to construct the initial two-dimensional matrix. The noise intensity of a series of original NMR spectra obtained by sampling is similar, so the present invention selects the noise intensity of each data block in an original NMR spectrum as the data block noise intensity σ of the data blocks divided by the initial two-dimensional matrix. y , noise intensity threshold σ T Determined by the noise intensity of the corresponding entire original NMR spectrum. Get the data block noise intensity σ of each data block y The noise intensity of the entire original NMR spectrum is based on existing technology.

[0100] The hard threshold operator γ′(y) operates on the pixels in the data block y, assuming that y >σ T When (in this embodiment, σ T=40), the same relative pixel coordinates of the reference data block P and the data block to be matched Q are both smaller than the hard threshold The pixels of σ are highly correlated, so when σ y >σ T When the hard threshold operator γ y ′(x y ) The data block y is below the hard threshold Pixels x y The corresponding values ​​are reset to zero, and the corresponding values ​​of other pixels remain unchanged; in the process of calculating the normalized Euclidean distance d(P,Q), when the corresponding pixels in the two data blocks are set to 0, the normalized Euclidean distance d(P,Q) will decrease; if the corresponding pixels in the two data blocks remain unchanged, it will not affect the normalized Euclidean distance d(P,Q); if the pixels on one data block are set to 0 and the corresponding pixels on the other data block remain unchanged, it is considered that the data correlation is poor, which will cause d(P,Q) to increase.

[0101] Step 3.2.4: For each reference data block P, based on the set distance threshold τ hard And the normalized Euclidean distance d(P,Q), each corresponding data block Q to be matched is screened to obtain the associated data set R(P), which can be expressed by the following formula:

[0102] R(P)={Q:d(P,Q)≤τ hard}

[0103] Where {Q:d(P,Q)≤τ hard} means that for the reference data block P, the normalized Euclidean distance d(P,Q) is less than or equal to the distance threshold τ hard The to-be-matched data block Q is used as the associated data block S of the reference data block P, and the set of associated data blocks S is the associated data set R(P).

[0104] Step 4: Sort and stack the associated data blocks S of the reference data block P to obtain a three-dimensional array M(P):

[0105] Step 4.1: For each reference data block P, sort the associated data blocks S of the reference data block P in ascending order according to the normalized Euclidean distance from the associated data block S to the corresponding reference data block P; in order to reduce the amount of calculation, this embodiment retains the first n hard There are S associative data blocks, for which the sequence number is greater than n hard The associated data block S is discarded; if the total number of domain block data is less than n hard , retain all domain block data.

[0106] Step 4.2: For each reference data block P, stack the sorted associated data blocks S according to the arrangement order in step 4.1 to obtain a three-dimensional array M(P).

[0107] Step 5: Perform a three-dimensional transformation, a frequency domain hard threshold filter, and a three-dimensional inverse transformation on the three-dimensional array M(P) in sequence to obtain a spatial domain three-dimensional correlation data group after frequency domain hard threshold filter. The three-dimensional transformation is used to achieve a sparse expression effect. The three-dimensional transformation includes a combination of one or more of the following transformations: discrete cosine transform, Walsh-Hadamard transform, Fourier transform, and wavelet transform. In this embodiment, the three-dimensional transformation includes discrete cosine transform and Walsh-Hadamard transform in sequence, and the three-dimensional inverse transformation includes Walsh-Hadamard inverse transform and discrete cosine inverse transform in sequence.

[0108] Step 5.1: Perform discrete cosine transform and Walsh-Hadamard transform on the three-dimensional array M(P) obtained in step 3 to obtain the frequency domain three-dimensional correlation data set M 3D (P);

[0109] Step 5.2: frequency domain three-dimensional correlation data set M 3D (P) performing frequency domain hard threshold filtering to obtain a frequency domain three-dimensional correlation data set after frequency domain hard threshold filtering;

[0110] Step 5.3, three-dimensional inverse transformation, the frequency domain three-dimensional correlation data group obtained after the frequency domain hard threshold filtering in step 5.2 is subjected to Walsh-Hadamard inverse transformation and discrete cosine inverse transformation in sequence to obtain the spatial domain three-dimensional correlation data group after the frequency domain hard threshold filtering.

[0111] Step 5.1, transform the three-dimensional array M(P) from the spatial domain to the frequency domain, step 5.2 improve the sensitivity by frequency domain hard threshold filtering, and then transform it back to the spatial domain by step 5.3 to obtain the spatial domain three-dimensional correlation data group after frequency domain hard threshold filtering. The data of each layer in the spatial domain three-dimensional correlation data group after frequency domain hard threshold filtering is recorded as the correlation data block S after frequency domain hard threshold filtering. h , the correlation data block S after frequency domain hard threshold filtering h One-to-one correspondence with the associative data block S.

[0112] Step 6: Use the number of non-zero frequency domain coefficients retained in the frequency domain three-dimensional correlation data group after the frequency domain hard threshold filtering in step 5.2 as the fusion weight, and perform the fusion on the correlation data block S after the frequency domain hard threshold filtering. h The first fusion weighting of each pixel in the image is performed to obtain the basic estimated map. The specific steps include:

[0113] After the hard threshold filtering in the frequency domain in step 5.2 is completed, for each matching window (equivalent to each reference data block P), the fusion weight corresponding to each matching window is obtained by using the number of non-zero frequency domain coefficients retained in the frequency domain three-dimensional correlation data group after the hard threshold filtering in the frequency domain in step 5.2. (i.e., the weight of the reference data block P); correspond the pixel x to the correlation data block S after hard threshold filtering in each frequency domain h The values ​​in are added and based on the fusion weight Perform weighted averaging to obtain the basic estimate u for each pixel x in the initial two-dimensional matrix basic (x), the basic estimate u of each pixel x basic (x) is arranged according to the pixel coordinates of the corresponding pixel x in the initial two-dimensional matrix to obtain a basic estimated map.

[0114] Basic estimate u basic (x) is given by:

[0115]

[0116] in,

[0117]

[0118] is the correlation data block S of pixel x after hard threshold filtering in the frequency domain h The corresponding value in ; is the fusion weight; α Q (x) is the pixel x membership coefficient, which is used to indicate whether the pixel x belongs to the associated data block S after frequency domain hard threshold filtering. h In the example, a pixel x is included in multiple reference data blocks P, and each reference data block P has an associated data set R(P). The associated data set R(P) contains a series of associated data blocks S. Each associated data block S corresponds to a frequency domain hard threshold filtered associated data block S. h ;

[0119] is the number of non-zero frequency domain coefficients retained in the frequency domain three-dimensional correlation data group after frequency domain hard threshold filtering. The more non-zero frequency domain coefficients there are in the frequency domain, the more high-frequency signals and noise there are, so its weight should be reduced. Therefore, it is limited to When greater than 1, the fusion weight equal The reciprocal of , in other cases the fusion weight Take 1, a three-dimensional array M(P) corresponds to a fusion weight It is a correlation data block S of the pixel x in a correlation data set R(P) after hard threshold filtering in each frequency domain. h Add the values ​​in and multiply them by the weight of the corresponding benchmark data block P.

[0120] Since the correlation data block S after frequency domain hard threshold filtering contains edges h The frequency domain has more non-zero coefficients, so they are given a lower weight. This weighting method prioritizes blocks that do not contain edges. The advantage is that it reduces artifacts around edges and avoids the classic ringing effect observed with the transform threshold method.

[0121] Step 7: Rematch the associated data. Similar to step 3, the basic estimated graph obtained in step 6 is matched with the associated data blocks. That is, the matching window traverses the basic estimated graph row by row, and the data corresponding to the matching window at each position in the basic estimated graph is searched to obtain the corresponding associated data blocks S′. The specific process is as follows:

[0122] Step 7.1, set the matching window size n hard ×n hard , n hard Indicates the side length of the matching window (in pixels). The matching window is traversed row by row in the basic estimation map with a set step.

[0123] Step 7.2: For each matching window, select a portion of continuous data in the basic estimated map corresponding to the matching window as the reference data block P′, and obtain the associated data blocks S′ of the reference data block P′ from the data in the basic estimated map corresponding to the matching window. There are multiple associated data blocks S′ corresponding to each reference data block P′. The specific steps are as follows:

[0124] Step 7.2.1. For each matching window, set the corresponding reference data block P′. The center of the reference data block P′ is the center of the corresponding matching window. The size of the reference data block P′ is k. hard ×k hard , k hard is the side length of the reference data block P′ (k hard <n hard );

[0125] Step 7.2.2, set the scanning window, the size of the scanning window is k hard ×k hard The scanning window scans row by row in the basic estimation map corresponding to each matching window. The data in the basic estimation map corresponding to each scanning window is recorded as a data block Q' to be matched. Each matching window corresponds to multiple data blocks Q' to be matched. Some data between the data blocks Q' to be matched can overlap.

[0126] Step 7.2.3: For each matching window, calculate the normalized Euclidean distance d(P′, Q′) between the reference data block P′ and each to-be-matched data block Q′ in the corresponding matching window.

[0127] To improve computational efficiency, the normalized Euclidean distance d(P′, Q′) between the reference data block P′ and the data block to be matched Q′ is calculated using the following formula:

[0128]

[0129] d(A,B) represents the normalized Euclidean distance between data blocks A and B, ‖.‖2 is the L2 distance operator, and γ′(·) is the hard threshold operator. In step 7, data block A takes the reference data block P′, and data block B takes the to-be-matched data block Q′. Then the normalized Euclidean distance d(P′,Q′) is:

[0130]

[0131] The hard threshold operator γ′(·) for the data block is calculated according to the following formula:

[0132]

[0133]

[0134] y represents the data block, x y represents the pixel in the data block y, γ′ y (.) represents the pixel hard threshold operator for the pixels in the data block y, Represents pixel x in data block y y Corresponding to the value in the basic estimation spectrum, σ y represents the noise intensity of the data block, σ y The larger the value, the greater the position noise intensity in the basic estimated map corresponding to the data block. T is the noise intensity threshold. In step 7.2.3, the noise intensity threshold σ T Determined by the noise intensity of the corresponding entire basic estimated spectrum, is the hard threshold, which is a fixed value set; Represents the hard threshold coefficient that is set.

[0135] Step 7.2.4: For each reference data block P′, based on the set distance threshold τ wien And the normalized Euclidean distance d(P′,Q′), each corresponding data block to be matched Q′ is screened to obtain the associated data set R(P′). The associated data set R(P′) can be expressed by the following formula:

[0136] R(P′)={Q′:d(P′,Q′)≤τ wien}

[0137] Where {Q′:d(P′,Q′)≤τ wien} means that for the reference data block P′, the normalized Euclidean distance d(P′,Q′) is less than or equal to the distance threshold τ wien The data block to be matched Q' is taken as the associated data block S' of the reference data block P', and the set of associated data blocks S' is the associated data set R(P'). wien =τ hard In this embodiment, since the number of associated data blocks S of the reference data block P in step 3 is less than the initially divided data blocks Q to be matched, and the number of associated data blocks S′ of the reference data block P′ in the subsequent step 8 is less than the data blocks Q′ to be matched, τ may not be used. wien and τ hard The size relationship is limited.

[0138] Step 8: Sort and stack the associated data blocks S′ of the reference data block P′ to obtain a three-dimensional array M basic (P′):

[0139] Step 8.1: For each reference data block P′, sort the associated data blocks S′ of the reference data block P′ in ascending order according to the normalized Euclidean distance from the associated data block S′ to the corresponding reference data block P′. To reduce the amount of calculation, this embodiment retains the first n hard There are associative data blocks S', for which the sequence number is greater than n hard The associated data block S' is discarded; if the total number of domain block data is less than n hard , retain all domain block data.

[0140] Step 8.2: For each benchmark data block P′, stack the sorted associated data blocks S′ in the order of step 8.1 to obtain a three-dimensional array M basic (P′).

[0141] Step 9: Re-stack the associated data blocks S in the associated data set R(P) obtained in step 3.2 to obtain a three-dimensional array M origin (P), where the associative data block S is re-stacked in order and the three-dimensional array M basic The order in which the associative data blocks S′ are stacked in (P′) is the same; that is, the three-dimensional array M origin The center pixel of the correlation data block S in (P) corresponds to the pixel coordinates in the initial two-dimensional matrix and the three-dimensional array M basic The center pixel of the correlation data block S′ at the same layer in (P′) has the same coordinates as the pixel point in the basic estimation map.

[0142] Step 10: For the three-dimensional array M obtained in step 9 origin (P) and the three-dimensional array M obtained in step 8 basic (P′) are respectively subjected to three-dimensional transformation (i.e., discrete cosine transform and Walsh-Hadamard transform are performed in sequence), and then Wiener collaborative filtering is performed to obtain a frequency domain three-dimensional correlation data group after Wiener collaborative filtering. The frequency domain three-dimensional correlation data group after Wiener collaborative filtering is subjected to a three-dimensional inverse transformation to obtain a spatial domain three-dimensional correlation data group after Wiener collaborative filtering, wherein the three-dimensional transformation is used to achieve the effect of sparse expression, and the three-dimensional transformation includes a combination of one or more of the following transformations in sequence: discrete cosine transform, Walsh-Hadamard transform, Fourier transform and wavelet transform. In this embodiment, the three-dimensional transformation includes discrete cosine transform and Walsh-Hadamard transform in sequence, and the three-dimensional inverse transformation includes Walsh-Hadamard inverse transform and discrete cosine inverse transform in sequence. Obtaining the spatial domain three-dimensional correlation data group after Wiener collaborative filtering specifically includes the following steps:

[0143] Step 10.1: For the three-dimensional array M obtained in step 8 basic (P′) is transformed into a three-dimensional frequency domain array For the three-dimensional array M obtained in step 9 origin (P) Perform three-dimensional transformation to obtain three-dimensional frequency domain array

[0144] Step 10.2: Use a three-dimensional frequency domain array As a guide to the 3D frequency domain array Perform Wiener collaborative filtering to obtain frequency domain three-dimensional correlation data after Wiener collaborative filtering. Wiener collaborative filtering is an existing technology. Specifically, it calculates the three-dimensional frequency domain array The empirical Wiener coefficient ω corresponding to each frequency domain coefficient ξ P (ξ), through a three-dimensional frequency domain array and the empirical Wiener coefficient ω P (ξ) Multiply the frequency domain coefficients ξ by one by one to obtain the frequency domain three-dimensional correlation data group after Wiener collaborative filtering.

[0145] Step 10.3: Inverse three-dimensional transformation. Perform Walsh-Hadamard inverse transformation and discrete cosine inverse transformation on the frequency domain three-dimensional correlation data set after Wiener collaborative filtering obtained in step 10.2 to obtain the spatial domain three-dimensional correlation data set after Wiener collaborative filtering. The data of each layer in the spatial domain three-dimensional correlation data set after Wiener collaborative filtering is recorded as the correlation data block S after Wiener collaborative filtering. w , the correlation data block S after Wiener collaborative filtering w They correspond one to one with the associated data blocks S respectively.

[0146] Step 11, the second fusion weighting, similar to step 6, takes the number of non-zero frequency domain coefficients in the frequency domain coefficient ξ of the frequency domain three-dimensional correlation data group after Wiener collaborative filtering in step 10.2 as the fusion weight, and performs the fusion weight on the correlation data block S after Wiener collaborative filtering. w The pixels in the image are weighted for the second fusion to obtain the second fusion spectrum, and the final NMR spectrum with enhanced sensitivity is obtained based on the second fusion spectrum.

[0147] According to the different types of original NMR spectra collected in step 1, the steps for obtaining the final NMR spectra with enhanced sensitivity based on the second fusion spectra are as follows:

[0148] For the original two-dimensional NMR spectrum, the spectrum after the second fusion is the final NMR spectrum with enhanced sensitivity;

[0149] For the original one-dimensional NMR spectrum, the final NMR spectrum after sensitivity enhancement is:

[0150] Any column in the atlas after the second fusion, or the sum of all columns in the atlas after the second fusion divided by the number of columns.

[0151] The second fusion weighting process is as follows:

[0152] After the Wiener collaborative filtering is completed, for each matching window (equivalent to each reference data block P), the weight corresponding to each matching window is obtained by using the number of non-zero frequency domain coefficients retained in the frequency domain three-dimensional correlation data group after the Wiener collaborative filtering in step 10.2 (i.e., the weight of the reference data block P); correspond the pixel x to the associated data block S after each Wiener collaborative filtering w The values ​​in are added and based on the fusion weight Perform weighted averaging to obtain the basic estimate u for each pixel x origin (x), the basic estimate u of each pixel x origin (x) Arrange according to the pixel coordinates of the corresponding pixel x in the initial two-dimensional matrix to obtain the second fusion map.

[0153] Basic estimate u origin (x) is given by:

[0154]

[0155] in,

[0156]

[0157] is the correlation data block S of pixel x after Wiener collaborative filtering w The corresponding value in ; is the fusion weight; α Q (x) is the pixel x membership coefficient, which is used to indicate whether the pixel x belongs to the associated data block S after Wiener collaborative filtering. w middle; is the number of non-zero frequency domain coefficients retained in the frequency domain three-dimensional correlation data group after Wiener collaborative filtering.

[0158] Figure 2 This is the original one-dimensional NMR spectrum without sensitivity enhancement; Figure 3 The one-dimensional NMR spectrum with enhanced sensitivity is processed using a spectrum sensitivity enhancement method based on the correlation characteristics of nuclear magnetic resonance signals proposed in the present invention.

[0159] Figure 4 This is the original two-dimensional NMR spectrum without sensitivity enhancement; Figure 5 It is a two-dimensional NMR spectrum with enhanced sensitivity after being processed using a spectrum sensitivity enhancement method based on the correlation characteristics of nuclear magnetic resonance signals proposed in the present invention.

[0160] The specific embodiments described herein are merely illustrative of the spirit of the present invention. Persons skilled in the art may make various modifications, additions, or substitutions to the described specific embodiments without departing from the spirit of the present invention or exceeding the scope of the appended claims.

Claims

1. A method for enhancing spectral sensitivity based on nuclear magnetic resonance signal correlation characteristics, characterized in that: The steps include: Step 1: Perform multiple independent samplings on the same sample to obtain the corresponding original NMR spectra; Step 2: construct an initial two-dimensional matrix using the original NMR spectrum; Step 3: Set a matching window. The matching window traverses the initial two-dimensional matrix row by row, searches for data corresponding to the matching window at each position in the initial two-dimensional matrix, and obtains the corresponding correlation data block S. Step 4: Sort and stack the associated data blocks S into a three-dimensional array M(P); Step 5: Perform a three-dimensional transformation and a frequency-domain hard threshold filter on the three-dimensional array M(P) in sequence to obtain a frequency-domain three-dimensional correlation data set after the frequency-domain hard threshold filter. Perform a three-dimensional inverse transformation on the frequency-domain three-dimensional correlation data set after the frequency-domain hard threshold filter to obtain a spatial-domain three-dimensional correlation data set after the frequency-domain hard threshold filter. The three-dimensional transformation is used to achieve a sparse expression effect. Step 6: Use the spatial domain three-dimensional correlation data group after frequency domain hard threshold filtering to perform the first fusion weighting to obtain a basic estimated spectrum; Step 7: The matching window traverses the basic estimation map row by row, searches for the data corresponding to the matching window at each position in the basic estimation map, and obtains the corresponding correlation data block S′; Step 8: Sort and stack the associated data blocks S′ to obtain a three-dimensional array M basic (P′); Step 9: Re-stack the associated data blocks S to obtain a three-dimensional array M origin (P), the order in which the associative data blocks S are re-stacked and the three-dimensional array M basic The order in which the associative data blocks S′ are stacked in (P′) is the same; Step 10: For the three-dimensional array M origin (P) and the three-dimensional array M basic (P′) performs three-dimensional transformation respectively, and obtains the corresponding three-dimensional frequency domain array and a three-dimensional frequency domain array Then use the three-dimensional frequency domain array As a guide to the 3D frequency domain array Performing Wiener collaborative filtering to obtain a frequency domain three-dimensional correlation data group after Wiener collaborative filtering, and performing a three-dimensional inverse transformation on the frequency domain three-dimensional correlation data group after Wiener collaborative filtering to obtain a spatial domain three-dimensional correlation data group after Wiener collaborative filtering; Step 11: Perform a second fusion weighting on the spatial three-dimensional correlation data set after Wiener collaborative filtering to obtain a second fused spectrum, and extract the final NMR spectrum with enhanced sensitivity based on the second fused spectrum.

2. The method for enhancing spectral sensitivity based on nuclear magnetic resonance signal correlation characteristics according to claim 1, characterized in that: The step 2 specifically includes the following steps: Step 2.1: Determine whether the original NMR spectrum type is one-dimensional or two-dimensional. If the original NMR spectrum type is one-dimensional, proceed to step 2.2; if the original NMR spectrum type is two-dimensional, proceed to step 2.

3. Step 2.2, concatenate the original one-dimensional NMR spectra by column to construct an initial two-dimensional matrix; Step 2.3, sequentially reading the real part data and the imaginary part data of each two-dimensional original NMR spectrum, and rearranging the read real part data and the imaginary part data according to the original time domain sampling point number to obtain a rearranged NMR spectrum, wherein the rearranged NMR spectrum includes a real part data matrix and an imaginary part data matrix; The real data matrix and the imaginary data matrix in each rearranged NMR spectrum are added together to obtain the corresponding two-dimensional complex matrix; Sum all two-dimensional complex matrices to obtain an initial two-dimensional matrix.

3. The method for enhancing spectral sensitivity based on nuclear magnetic resonance signal correlation characteristics according to claim 1, characterized in that: Both step 3 and step 7 include the following steps: Set the matching window size n hard ×n hard , n hard Indicates the side length of the matching window, and the minimum value of the length and width of the initial two-dimensional matrix obtained in step 2 is recorded as min, n hard ≤min; the matching window is traversed row by row in the object matrix at a set step; For each matching window, select data of a part of the continuous area in the object matrix corresponding to the matching window as a reference data block, and obtain the associated data blocks of each reference data block from the data of the object matrix corresponding to the matching window; In step 3, the object matrix takes the initial two-dimensional matrix, the corresponding reference data block is recorded as P, and the correlation data block is recorded as S; In step 7, the object matrix takes the basic estimation map, the corresponding reference data block is recorded as P′, and the correlation data block is recorded as S′.

4. The method for enhancing spectrum sensitivity based on nuclear magnetic resonance signal correlation characteristics according to claim 3, characterized in that: Obtaining the associated data blocks of each of the reference data blocks specifically includes the following steps: For each matching window, the center of the reference data block is set as the center of the corresponding matching window, and the size of the reference data block is k hard ×k hard , k hard is the side length of the benchmark data block, k hard <n hard ; Set the scanning window, the size of the scanning window is k hard ×k hard The scanning window scans row by row in the object matrix corresponding to each matching window, and the data in the object matrix corresponding to each scanning window is the data block to be matched; For each matching window, calculate the normalized Euclidean distance between the reference data block and each to-be-matched data block in the corresponding matching window; For each reference data block, the data blocks to be matched whose normalized Euclidean distance is less than or equal to the distance threshold are used as the associated data blocks of the reference data block, and the set of associated data blocks is the associated data set; In step 3, the data block to be matched is recorded as Q, and the distance threshold is recorded as τ hard ,The set of related data is denoted as R(P); In step 7, the data block to be matched is recorded as Q′, and the distance threshold is recorded as τ wien , the associated data set is recorded as R(P′).

5. The method for enhancing spectrum sensitivity based on nuclear magnetic resonance signal correlation characteristics according to claim 4, characterized in that: Both step 4 and step 8 include the following steps: For each reference data block, the associated data blocks of the reference data block are sorted in ascending order according to the normalized Euclidean distance from the associated data block to the corresponding reference data block and stacked to obtain a corresponding three-dimensional array; In step 4, the reference data block is P, the associated data block is S, and the three-dimensional array is denoted as M(P); In step 8, the reference data block is P', the associated data block is S', and the three-dimensional array is M basic (P′).

6. The method for enhancing spectrum sensitivity based on nuclear magnetic resonance signal correlation characteristics according to claim 5, characterized in that: The normalized Euclidean distance is calculated using the following formula: d(A,B) represents the normalized Euclidean distance between data block A and data block B, ‖.‖2 is the L2 distance operator; γ′(·) is the data block hard threshold operator; in step 3, data block A takes the reference data block P, and data block B takes the to-be-matched data block Q; in step 7, data block A takes the reference data block P′, and data block B takes the to-be-matched data block Q′; The data block hard threshold operator γ′(·) is calculated according to the following formula: y represents the data block, x y represents the pixels in the data block y, γ y ′(.) represents the pixel hard threshold operator for the pixels in the data block y, Represents pixel x in data block y y For values ​​in the object matrix, σ y represents the noise intensity of the data block, is the hard threshold, which is a fixed value. represents the hard threshold coefficient; σ T is the noise intensity threshold, the noise intensity threshold σ T Determined by the noise intensity of the corresponding entire object matrix; In step 4, the object matrix takes the initial two-dimensional matrix; In step 8, the object matrix takes the basic estimated map.

7. The method for enhancing spectrum sensitivity based on nuclear magnetic resonance signal correlation characteristics according to claim 5, characterized in that: The first fusion weighting performed in step 6 and the second fusion weighting performed in step 11 both include calculating a basic estimate for each pixel x, and arranging the basic estimate for each pixel x according to the pixel coordinates of the corresponding pixel x in the initial two-dimensional matrix; The basic estimate is given by: in, u(x) represents the basic estimate, u Q,P (x) is the value corresponding to pixel x in the filtered data block; w p is the fusion weight; α Q (x) is the pixel x membership coefficient, which is used to indicate whether the pixel x belongs to the filtered data block; N p is the number of non-zero frequency domain coefficients retained in the filtered frequency domain three-dimensional correlation data set; In step 5, The filtered data block is the correlation data block S after the frequency domain hard threshold filtering h ; is the correlation data block S of pixel x after hard threshold filtering in the frequency domain h The corresponding value in ; is the number of non-zero frequency domain coefficients retained in the frequency domain three-dimensional correlation data group after frequency domain hard threshold filtering; the data of each layer in the spatial domain three-dimensional correlation data group after frequency domain hard threshold filtering is recorded as the correlation data block S after frequency domain hard threshold filtering h , the correlation data block S after frequency domain hard threshold filtering h One-to-one correspondence with the association data block S; the basic estimate of the calculation is recorded as u basic (x), the basic estimate u basic (x) Arrange the pixel coordinates of the corresponding pixel x in the initial two-dimensional matrix to obtain a basic estimated map; In step 11, The filtered data block is the correlation data block S after Wiener collaborative filtering w ; is the correlation data block S of pixel x after Wiener collaborative filtering w The corresponding value in ; is the number of non-zero frequency domain coefficients retained in the frequency domain three-dimensional correlation data group after Wiener collaborative filtering; the data of each layer in the spatial domain three-dimensional correlation data group after Wiener collaborative filtering is recorded as the correlation data block S after Wiener collaborative filtering w , the correlation data block S after Wiener collaborative filtering w One-to-one correspondence with the association data block S; the basic estimate of the calculation is recorded as u origin (x), the basic estimate u of each pixel x origin (x) Arrange according to the pixel coordinates of the corresponding pixel x in the initial two-dimensional matrix to obtain the second fusion map.

8. The method for enhancing spectrum sensitivity based on nuclear magnetic resonance signal correlation characteristics according to claim 2, characterized in that: Depending on the type of the original NMR spectrum collected in step 1, step 11 extracts the final NMR spectrum with enhanced sensitivity based on the spectrum after the second fusion, including the following steps: For the original two-dimensional NMR spectrum, the spectrum after the second fusion is the final NMR spectrum with enhanced sensitivity; For a one-dimensional original NMR spectrum, the final NMR spectrum after sensitivity enhancement is: any column in the spectrum after the second fusion, or the sum of the columns of data in the spectrum after the second fusion divided by the number of columns.

9. The method for enhancing spectrum sensitivity based on nuclear magnetic resonance signal correlation characteristics according to claim 1, characterized in that: The three-dimensional transformation includes one or more of the following transformations in sequence: Discrete cosine transform, Walsh-Hadamard transform, Fourier transform and wavelet transform.

10. The method for enhancing spectrum sensitivity based on nuclear magnetic resonance signal correlation characteristics according to claim 6, characterized in that: In step 4, the noise intensity of the initial two-dimensional matrix is ​​equal to the noise intensity of a single original NMR spectrum.

Images