阅读说明：本技术 一种基于信号分段分类的导数谱平滑处理方法 (Derivative spectrum smoothing processing method based on signal segmentation classification ) 是由 周慧敏 李远禄 赵伟静 孙双龙 李腾 于 2019-10-14 设计创作，主要内容包括：本发明提出一种基于信号分段分类的导数谱平滑处理方法,能够在去噪的同时,进一步保护峰的特征。本发明通过对含噪导数谱信号进行分段处理、分类匹配、硬阈值去噪、小波变换、维纳滤波得到得到最终去噪结果。在对导数谱信号处理前,增加了对信号分段分类匹配的步骤,利用信号的相似性,更有利于提高信噪比。该平滑方法对导数谱具有更好的保峰去噪效果；该方法应用简便,节约大量的计算过程,峰形保护效果更佳,具有更好的应用前景。(The invention provides a derivative spectrum smoothing processing method based on signal segmentation classification, which can further protect the characteristics of peaks while denoising. The method obtains a final denoising result by performing segmentation processing, classification matching, hard threshold denoising, wavelet transformation and wiener filtering on the noisy derivative spectrum signal. Before the derivative spectrum signal is processed, a step of classifying and matching the signal in a segmented manner is added, and the signal-to-noise ratio is improved by utilizing the similarity of the signal. The smoothing method has better peak-preserving and denoising effects on the derivative spectrum; the method is simple and convenient to apply, saves a large number of calculation processes, has a better peak shape protection effect, and has a better application prospect.)

1. A derivative spectrum smoothing processing method based on signal segmentation classification is characterized by comprising the following specific steps:

(1) inputting a derivative spectrum signal s (X) containing noise to be processed, wherein X is a data point in the signal s (X), y (X) represents a clean signal, n (X) represents noise, and X represents a set of data points X;

(2) carrying out segmentation processing on the spectrum signal containing the noise derivative, and dividing the spectrum signal into a plurality of signal segments with equal length; the segmentation length range takes 5-15 data points, and the segmentation interval is more than 1 data point and less than the length of the signal segment;

(3) classifying and matching a plurality of signal segments obtained by segmentation, arranging the signal segments in various classes into a two-dimensional array according to the size of similarity, and extracting a similar data matrix corresponding to the signal segments in each class:

(4) carrying out hard threshold denoising on the similar data matrix corresponding to each type of signal segment through shrinkage of the transformation coefficient to obtain a denoised similar data matrix coefficient:

(5) carrying out two-dimensional inverse discrete cosine transform on the denoised similar data matrix coefficient to obtain similar data matrixes corresponding to the denoised signal sections

wherein the weight value of each type of signal segment is

in the formula (I), the compound is shown in the specification,

x_m,x_R∈x,is a characteristic function of the signal segments in each class,

(6) the result of preliminary estimation of the derivative spectrum signal obtained abovePerforming wavelet transform to obtain transformed wavelet coefficients, which are used for wiener filter coefficients W^wieThe formula (3):

in the formula (I), the compound is shown in the specification,

and filtering the derivative spectrum signal by using a wiener filter designed by the coefficient to obtain a final denoising result.

2. The method for smoothing derivative spectrum based on signal segment classification as claimed in claim 1, wherein the specific process of step (3) is as follows:

firstly, one signal segment is randomly selected as a reference segment, a signal segment similar to the reference segment is searched in a search window with the fixed size of n, and if the Euclidean distance between the two segments is less than a specified parameter T_matchThe same type is considered; wherein n is an integer which is larger than the length of the signal segment and smaller than 200; the parameter T_matchThe range is 0.01-0.05;

secondly, randomly selecting one signal segment from the remaining dissimilar signal segments as a reference segment, and repeating the process until all the signal segments participate in the classification process;

thirdly, stacking the signal segments classified into one type into a two-dimensional array to obtain a similar data matrix corresponding to each type of signal segments; the matrix is represented as

Wherein, the Euclidean distance is shown as formula (1):

in the formula (I), the compound is shown in the specification,

3. the method as claimed in claim 2, wherein in the first step of step (3), the threshold parameter is 0.05.

4. The method for smoothing derivative spectrum based on signal segmentation classification as claimed in any one of claims 1-3, wherein the specific process of step (4) is as follows:

firstly, performing two-dimensional discrete cosine transform on the similar data matrix, and then performing threshold shrinkage denoising on a coefficient lambda of the transformed similar data matrix to obtain a coefficient gamma (lambda ) of the denoised similar data matrix_thγ) The formula is as follows:

in the formula, λ_thrThe value range for the threshold parameter may be between 0.10 and 0.15.

The technical field is as follows:

the invention relates to a derivative spectrum smoothing and peak-protecting method, in particular to a peak-protecting smoothing processing method based on signal segmentation classification matching.

Background art:

derivative methods are a more common method used in signal analysis applications, where initially, derivative spectra are used to enhance the signal. With the progress of research, the derivative method is widely applied to spectral peak identification and spectral resolution improvement. In recent years, people have more and more researches on derivative spectrums, and in analysis applications such as biology, medicine and the like, the derivative spectrums have the characteristic that detection components of complex samples can be directly measured without separation, and can quickly and accurately perform qualitative and quantitative analysis on the samples. In chemical sample analysis, the peak position is a characteristic of a compound and the peak shift can indicate the interaction of two compounds or the chemical nature of the absorbent. However, the derivative spectrum is extremely sensitive to noise, and noisy spectral data cannot be used at all. Therefore, it is necessary to study the peak-hold smoothing process of the derivative spectrum.

With the development of the research and the deepening of the problem complexity, the traditional derivative spectrum smoothing method cannot meet the requirement of high precision. The spectral peak signal has large characteristic change of each section, a flat area is smooth and the peak is seriously weakened by adopting a uniform processing method, and if the peak characteristic is well protected, a large amount of noise still exists in the flat area. A typical sliding mean filtering method, although capable of removing noise to some extent, is easy to destroy the characteristics of the spectral peaks of the derivative; the Gaussian filtering is an improved method for smoothing a window by sliding mean filtering, a Gaussian function is used as the smoothing window, similarly, Kaiser proposes Kaiser filtering, and the Kaiser window is used as the smoothing window, and the two methods have high efficiency, but have weak peak protection effect. Spectral analysis is another method in signal smoothing, which requires the signal to be converted to the frequency domain before being spectrally analyzed. The wavelet method is the most commonly used spectrum analysis method at present, and can better protect the signal characteristics, however, the results of signal processing can be greatly different by selecting different wavelet bases and scales.

The unified processing method has certain defects on the smoothing method of the derivative spectrum. Therefore, how to further improve the smoothing effect is an important issue in the denoising of the derivative spectrum.

The invention content is as follows:

aiming at the defects of the prior art, the invention provides a derivative spectrum smoothing processing method based on signal segmentation classification, which can further protect the characteristics of peaks while denoising.

The specific technical scheme of the invention is as follows:

in order to realize the purpose of the invention, the invention adopts the following specific technical scheme:

a derivative spectrum smoothing processing method based on signal segmentation classification specifically comprises the following steps:

(1) inputting a derivative spectrum signal s (X) containing noise to be processed, wherein X is a data point in the signal s (X), y (X) represents a clean signal, n (X) represents noise, and X represents a set of data points X;

(2) carrying out segmentation processing on the spectrum signal containing the noise derivative, and dividing the spectrum signal into a plurality of signal segments with equal length; the segmentation length range takes 5-15 data points, and the segmentation interval is more than 1 data point and less than the length of the signal segment;

(3) classifying and matching a plurality of signal segments obtained by segmentation, arranging the signal segments in various classes into a two-dimensional array according to the size of similarity, and extracting a similar data matrix corresponding to the signal segments in each class:

(4) carrying out hard threshold denoising on the similar data matrix corresponding to each type of signal segment through shrinkage of the transformation coefficient to obtain a denoised similar data matrix coefficient:

(5) carrying out two-dimensional inverse discrete cosine transform on the denoised similar data matrix coefficient to obtain similar data matrixes corresponding to various signal segments

Returning the signal segments in each type of similar data matrix to the original positions, and obtaining a primary estimation value of the derivative spectrum signal through weighted average summation

The formula is as follows:

wherein the weight value of each type of signal segment is

Is defined as:

in the formula (I), the compound is shown in the specification,

representing the number of (non-zero) coefficients retained after the signal segments in each class are subjected to hard threshold filtering;

x_m,x_R∈x,

is a characteristic function of the signal segments in each class,

represents a set of similar segments, X represents a set of data points X;

(6) the result of preliminary estimation of the derivative spectrum signal obtained above

Performing wavelet transform to obtain transformed wavelet coefficients, which are used for wiener filter coefficients W^wieThe formula (3):

in the formula (I), the compound is shown in the specification,

is a preliminary estimate

Transformed wavelet coefficient, σ²Representing noise n (x)Variance;

and filtering the derivative spectrum signal by using a wiener filter designed by the coefficient to obtain a final denoising result.

Further, the specific process of step (3) is as follows:

firstly, one signal segment is randomly selected as a reference segment, a signal segment similar to the reference segment is searched in a search window with the size fixed to n, and if the Euclidean distance between the two segments is less than a specified parameter T_matchThe same type is considered; wherein n is an integer which is larger than the length of the signal segment and smaller than 200; parameter T_matchThe range is 0.01-0.05;

secondly, randomly selecting one signal segment from the remaining dissimilar signal segments as a reference segment, and repeating the process until all the signal segments participate in the classification process;

thirdly, stacking the signal segments classified into one type into a two-dimensional array to obtain a similar data matrix corresponding to each type of signal segments; the matrix is represented as

Wherein

Representing a set of similar segments; partial classification results are given in figure 2.

Wherein, the Euclidean distance is shown as formula (1):

in the formula (I), the compound is shown in the specification,

denotes a reference segment, S_xRepresenting the segment to be matched, N₁Is the length of the signal segment;

further, in the first step of the step (3), the parameter is 0.05.

Further, the specific process of step (4) is as follows:

first, to the phasesPerforming two-dimensional discrete cosine transform on the similar data matrix, and then performing threshold shrinkage denoising on the coefficient lambda of the transformed similar data matrix to obtain the coefficient gamma (lambda ) of the denoised similar data matrix_thr) The formula is as follows:

in the formula, λ_thrThe value range for the threshold parameter may be between 0.10 and 0.15.

Compared with the prior art, the invention has the following beneficial technical effects:

1. the method obtains a final denoising result by performing segmentation processing, classification matching, hard threshold denoising, wavelet transformation and wiener filtering on the noisy derivative spectrum signal.

2. Before the derivative spectrum signal is processed, the invention adds the step of classifying and matching the signal in segments, and is more beneficial to improving the signal-to-noise ratio by utilizing the similarity of the signal.

3. The invention designs a new smoothing method, which has better peak-preserving and denoising effects on a derivative spectrum;

4. the method disclosed by the invention is simple and convenient to apply, saves a large number of calculation processes, and has a better peak shape protection effect and a better application prospect.

Drawings

FIG. 1 is an overall flow chart of the present invention (also referred to as abstract figure);

FIG. 2 is a diagram illustrating the result of signal segmentation;

FIG. 3 is a diagram illustrating the signal classification result;

FIG. 4 is a graph illustrating the effect of spectral smoothing on noisy derivatives;

fig. 5 is a schematic diagram of the design of a wiener filter.

Detailed Description

The technical solution of the present invention is further described below with reference to the accompanying drawings and examples.