CN110208248A - A method of identification Raman spectrum exception measuring signal - Google Patents

Abstract

The invention provides a method for identifying abnormal Raman spectrum measurement signals, which can identify and eliminate abnormal spectra in single Raman spectrum measurement and parallel measurement. After eliminating the baseline drift for the single measurement spectral signal, estimate the signal-to-noise ratio; subtract the baseline for the parallel measurement spectral signal, calculate the standard deviation of each wave point of each measurement spectrum relative to the average spectrum, and add them to calculate the total average standard deviation. Design a suitable threshold to eliminate abnormal signals in single measurement and parallel measurement to ensure the accuracy and adaptability of subsequent qualitative and quantitative analysis of Raman spectroscopy.

Description

A Method for Identifying Raman Spectrum Abnormal Measurement Signals

technical field

The invention belongs to the technical field of Raman spectrum analysis, and in particular relates to a method for identifying abnormal measurement signals in the process of collecting Raman spectra.

Background technique

Raman spectroscopy has the advantages of no need for sample pretreatment, short operation time, simplicity, and high sensitivity, and can accurately perform qualitative and quantitative analysis of substances. It has been widely used in petrochemical, biomedical, geological archaeology, criminal justice and other fields. . Raman spectrum peaks are usually sharper, making it easier to identify mixtures. However, in Raman spectroscopy tests, fluorescence interference is often encountered, and because the Raman scattered light itself is extremely weak, the background will drown out the Raman signal. At the same time, Raman spectroscopy is an absolute intensity measurement without background, which is easily interfered by external factors. Therefore, in the practical application of Raman spectroscopy analysis technology, in order to obtain high-quality Raman spectroscopy signals, the average spectrum of multiple measurement spectra of the same sample is often used as the representative spectrum for analysis and processing. If there is an abnormal spectral signal and a large signal-to-noise ratio due to improper operation, instrument disturbance or other unknown reasons during the collection of Raman spectrum, the measurement is invalid, and the average spectrum can only be calculated after removing the abnormal signal. However, in the actual measurement process, due to subjective or objective factors, it is difficult for the measurement personnel to find this invalid measurement, which leads to the wrong preservation of the measurement results.

In the process of drug and food testing, the commonly used spectroscopic detectors use the above-mentioned Raman spectral analysis technology as a theoretical basis to perform non-edible chemical substances, food and drug additives, pesticide residues, veterinary drug residues, heavy metals, and bacterial colonies on drugs and foods. Qualitative and quantitative detection of the total number and illegal additions in health food. In the rapid inspection of drugs using Raman spectroscopy, the use of abnormal signals for analysis will cause misjudgment of drug categories. At present, the identification of abnormal spectra in most studies is based on the modeling sample set combined with pattern recognition methods such as Mahalanobis distance and leverage value. In the abnormal signal based on pattern recognition detection, there may be anomalies caused by the signal itself being invalid or confusion caused by the effective signal being a new type of anomaly, which increases the difficulty of identifying the type of abnormal signal, resulting in food or drug testing results. Therefore, as a preventive mechanism, Before using the sample spectrum for formal analysis, the abnormal signal in the single and repeated measurement of the sample is automatically identified, and the interference caused by the abnormal signal itself to the subsequent pattern recognition judgment is eliminated, so as to provide accurate and reliable qualitative and quantitative analysis for subsequent substances. source of information.

Contents of the invention

The purpose of the present invention is to provide a method that can identify and eliminate abnormal spectra in single measurement and parallel measurement of Raman spectrum, and can eliminate abnormal signals in single measurement and parallel measurement to ensure the qualitative and quantitative quality of Raman spectrum Analytical accuracy and adaptability.

The invention provides a method for identifying abnormal Raman spectrum measurement signals, comprising the following steps:

1) The spectrometer collects the spectral data of the sample, and uses the spectrum identification method combining the amplitude method and the first order differential to detect the peak "valley point". After the peak "valley point" is identified, the rubber band method is used to determine the baseline, and then the baseline is subtracted.

2) In order to improve the calculation speed, on the basis of ensuring the spectral waveform, the spectral data is sampled at equal intervals according to the spectral length to form an initial spectral sequence for subsequent calculations. According to the sampling theorem, the number of data points contained in each interval is more than three times the number of data points contained between the smallest adjacent peaks and valleys. If the calculation speed of the original spectral sequence meets the application requirements, the sampling process can be omitted.

3) For each measured spectrum, calculate the spectral measurement signal, including the mean value and standard deviation after sampling or not, and then calculate the difference between each absorbance value and the mean value in the spectral sequence to form a new mean-centered signal sequence Sc.

4) Establish the corresponding threshold (according to the actual application, one standard deviation, two standard deviations or three standard deviations), if the median value of the Sc sequence in step 3 is greater than this threshold, then this point is regarded as a useful signal, and put into the signal sequence S .

5) Remove the data points placed in the signal area from the original spectral sequence, and repeat the calculations of steps 3), 4) and 5) with the remaining spectral sequence until there is no signal greater than the threshold, and the iteration ends, then the remaining spectral sequence is Noise sequence R.

6) According to the obtained signal sequence S and noise sequence R, use the signal-to-noise ratio formula Calculate the signal-to-noise ratio of a single measurement spectrum.

7) If SNR<10 (signal-to-noise ratio threshold value obtained from a large number of data processing experiments), the measurement is considered to be an invalid spectrum and should be eliminated.

8) For multiple measurement spectra in the same sample whose signal-to-noise ratio meets the requirements, at least 3 measurements, calculate the average spectrum after deducting the baseline, and then calculate the standard deviation sequence corresponding to each wavenumber point of each measurement spectrum relative to the average spectrum, and calculate the standard The average value of the difference series can be obtained for each measured spectrum with respect to the total standard deviation STD of the average spectrum.

9) If STD>1.5 (standard deviation threshold obtained from a large number of data processing experiments), this spectrum is regarded as an "outlier" abnormal spectrum in parallel measurement and should be eliminated.

10) Eliminate the abnormal spectrum determined in step 9 from the parallel measurement spectrum, and repeat steps 8 and 9 with the remaining parallel measurement spectrum until the parallel spectrum <3 or STD<1.5, the iteration ends, and the identification of the abnormal measurement signal of the Raman spectrum is realized with culling.

The beneficial effect of the present invention is that: before using the sample spectrum, the validity of a single measurement signal (based on signal-to-noise ratio estimation) and the similarity of multiple parallel measurements ("outlier" judgment based on standard deviation) can be comprehensively considered, Automatically identify and eliminate abnormal spectral signals in sample measurement, so as to ensure the validity, accuracy and reliability of the signal sources required for qualitative and quantitative analysis of Raman spectroscopy.

Description of drawings

Fig. 1 is the Raman spectrogram of 6 parallel measurements of the same sample;

Figure 2 is the Raman spectrum after baseline subtraction;

Figure 3 is the remaining parallel spectrum after removing the non-information spectrum according to the signal-to-noise ratio;

Fig. 4 is the parallel measurement spectrum and the final spectrum after eliminating the abnormal spectrum.

specific implementation plan

The present invention will be further described more clearly and completely below, obviously, the described examples are only some examples of the present invention, rather than all the embodiments.

A method for identifying abnormal Raman spectrum measurement signals provided by the present invention comprises the following steps:

In step 1, the spectral data of the sample is collected, and the spectral data are sampled at equal intervals according to the spectral length to form an initial spectral sequence.

Spectral data of the samples collected above are specifically:

The spectrum recognition method combining the amplitude method and the first order differential is used to detect the peak and valley points; after the peak and valley points are identified, the tape method is used to determine the baseline, and then the baseline is subtracted to obtain a signal-to-noise ratio that eliminates baseline interference Raman spectrum.

On the basis of ensuring the spectral waveform, the spectral data is sampled at equal intervals according to the spectral length to form the initial spectral sequence for subsequent calculations; according to the sampling theorem, the number of data points contained in each interval is more than that contained between the smallest adjacent peaks and valleys 3 times the number of data points.

Step 2, for each measured spectrum, calculate the mean value and standard deviation of the spectral measurement signal, and then calculate the difference between each absorbance value and the mean value in the spectrum sequence to form a new mean-centered signal sequence Sc;

Step 3, setting a threshold, which is determined according to the actual application and is one standard deviation, two standard deviations or three standard deviations. In the second step, if the median value of the Sc sequence is greater than the threshold value, this point is regarded as a useful signal, and put into the useful signal sequence S;

Step 4: After removing the data points put into the useful signal sequence from the original spectral sequence, use the remaining spectral sequence as the new signal sequence Sc, and return to step 3 for iterative calculation until there is no signal larger than the set threshold, and record the remaining The spectral sequence is the noise sequence R, go to step 5;

Step 5, according to the obtained useful signal sequence S and noise sequence R, use the signal-to-noise ratio formula to calculate the signal-to-noise ratio of a single measurement spectrum; the signal-to-noise ratio formula is

Step 6, based on the signal-to-noise ratio to eliminate invalid spectra, when the signal-to-noise ratio SNR<10, it is considered that the measurement is an invalid spectrum and should be eliminated.

Step 7: For the parallel measurement spectra whose signal-to-noise ratio meets the requirements in the same sample, the parallel measurement spectra whose signal-to-noise ratio meets the requirements in the same sample are not less than 3 times, and calculate the standard deviation STD of each measurement spectrum relative to the average spectrum.

Step 8, when the standard deviation STD>1.5, eliminate the "outlier" abnormal spectrum in the parallel measurement, repeat step 7 with the remaining parallel measurement spectrum, until the parallel spectrum <3 or STD<1.5, the iteration ends, and the realization Identification and elimination of abnormal measurement signals of Raman spectroscopy.

The content of the technical scheme of the present invention is described in detail below in conjunction with the example that the medicine of certain pharmaceutical factory is detected, and Raman spectrum data is taken from the medicine Raman spectrum of certain pharmaceutical factory, selects amoxicillin clavulanate potassium chewable tablet and metformin hydrochloride slow The 6 measured spectra of the release tablet are shown in Figure 1.

Referring to Fig. 1, the method for identifying the abnormal measurement signal of Raman spectrum includes the following steps:

1. Collect the spectral data of the drug through the spectrometer, and use the spectrum recognition method combining the amplitude method and the first order differential to detect the peak "valley point". After the peak "valley point" is identified, the "tape method" is used to determine the baseline, and then the baseline is subtracted. The result is shown in Figure 2.

2. For the convenience of calculation, on the basis of ensuring the spectral waveform, determine the appropriate sampling interval according to the spectral length, calculation speed requirements and peak width (valley-valley interval) to generate a new spectral sequence for subsequent calculations. For the Raman spectrum of the embodiment, the full spectrum signal is selected as the original spectrum sequence.

3. For each measured spectrum, calculate the mean value of the spectrum measurement signal, and obtain the result of subtracting the mean value from each point in the spectrum sequence.

4. The embodiment uses three times the standard deviation as the threshold value. If the result in step 3 is greater than the threshold value, then this point is regarded as a useful signal and put into the signal sequence S.

5. Remove the wave points placed in the signal area from the original spectral sequence, and repeat the calculations of steps 3 and 4 with the remaining spectral sequence until the signal with an incomparable threshold value is reached. After the iteration is over, the remaining spectral sequence is the noise sequence R.

6. Use the signal-to-noise ratio formula Calculate the signal-to-noise ratio of a single measurement spectrum. The signal-to-noise ratio results of a single measurement spectrum in the examples are shown in Table 1.

Table 1 Analysis results of single measurement spectrum anomaly

7. If SNR<10 (signal-to-noise ratio threshold value obtained from a large number of data processing experiments), the measurement is considered to be an invalid spectrum and should be discarded. As shown in Table 1, the signal-to-noise ratios of the 6 measurements of the Raman spectra of metformin hydrochloride sustained-release tablets are all greater than 10, which are all effective spectra; The SNR < 10 in the third measurement is an invalid abnormal signal and should be eliminated. The remaining effective spectrum obtained is shown in Fig. 3 .

8. For the measurement spectrum whose signal-to-noise ratio meets the requirements in the same sample, calculate the average spectrum of multiple measurements after deducting the baseline, and then calculate the standard deviation sequence corresponding to each wavenumber point of each measurement spectrum relative to the average spectrum and calculate the average, thus Obtain the standard deviation STD of each measured spectrum relative to the total average spectrum.

9. If STD>1.5 (standard deviation threshold obtained from a large number of data processing experiments), the spectrum is regarded as an abnormal spectrum and should be eliminated.

10. Remove the abnormal spectra identified in step 9 from the measured multiple measurement spectra, and repeat steps 8 and 9 with the remaining parallel measurement spectra until the repeated measurement spectra are <3 or STD<1.5, and the iteration ends to realize the Raman spectrum Identification and elimination of abnormal measurement signals. The iterative process of the total standard deviation of each measured spectrum relative to the average spectrum in the embodiment is shown in Table 2. The obtained remaining similar parallel measurement spectra are shown with reference to FIG. 4 .

Table 2 The total standard deviation of each measured spectrum relative to the average spectrum

The process of using Raman spectroscopy for rapid drug detection is susceptible to interference from various factors in the external environment, which will lead to uneven quality of the collected Raman spectroscopy data. Using the method described in this patent can effectively eliminate the two types of signal abnormal spectra collected in the on-site quick inspection, so as to ensure the reliability of the quality of drug Raman spectrum information sources, and help improve the accuracy of rapid drug identification results based on Raman spectroscopy .

The above is only a preferred embodiment of the present invention, and the scope of protection of the present invention is not limited thereto. Anyone skilled in the art can easily think of changes or substitutions within the technical scope disclosed in the present invention. All should be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention should be determined by the protection scope of the claims.

Claims (9)

1. a kind of method for recognizing Raman spectrum exception measuring signal, which is characterized in that the described method comprises the following steps:

Step 1 acquires the spectroscopic data of sample, carries out equal interval sampling to spectroscopic data according to spectra length, forms initial Spectral sequence；

Step 2 calculates spectroscopy signal mean value and standard deviation for the spectrum measured every time, then calculates every in spectral sequence The difference of a absorbance value and mean value constitutes new mean value centralization signal sequence Sc；

Step 3, given threshold think that this puts as useful signal for what Sc sequence intermediate value in the step 2 was greater than the threshold value, It is put into useful signal sequence S；

Step 4, from original spectrum sequence removal be put into the data point of useful signal sequence after, use remaining spectral sequence as New signal sequence Sc, return step three are iterated calculating, until note is remaining without occurring comparing given threshold big signal Spectral sequence is noise sequence R, goes to step five；

Step 5 calculates single measurement light using signal-to-noise ratio formula according to obtained useful signal sequence S and noise sequence R The signal-to-noise ratio of spectrum；

Step 6 rejects invalid spectrum based on signal-to-noise ratio；

It is relatively average to calculate every measure spectrum to the satisfactory parallel determination spectrum of signal-to-noise ratio in same sample for step 7 The standard deviation STD of spectrum；

Step 8, the exceptional spectrum " to peel off " in being measured in parallel is rejected according to the standard deviation STD, is measured in parallel spectrum with residue Repeat step 7, until parallel spectrum item number or STD are less than setting value, iteration terminates, and realizes that Raman spectrum measures extremely The identification and rejecting of signal.

2. a kind of method for recognizing Raman spectrum exception measuring signal according to claim 1, which is characterized in that the letter It makes an uproar and is than formula。

3. a kind of method for recognizing Raman spectrum exception measuring signal according to claim 1, which is characterized in that the step In rapid eight, parallel spectrum < 3 or STD < 1.5, iteration terminate.

4. a kind of method for recognizing Raman spectrum exception measuring signal according to claim 1, which is characterized in that step 1 In, acquire the spectroscopic data of sample specifically:

The mass spectrum database method combined using amplitude method and first differential carries out the detection of peak valley point；After the identification of peak valley point, adopt Baseline is determined with tape method, then deducts baseline, the Raman spectrum that can be used for calculating signal-to-noise ratio for the baseline interference that is eliminated.

5. a kind of method for recognizing Raman spectrum exception measuring signal according to claim 1, which is characterized in that the step In rapid one, under the basis for guaranteeing spectral waveform, equal interval sampling is carried out to spectroscopic data according to spectra length, is formed initial Spectral sequence carries out subsequent calculating；According to sampling thheorem, in each interval comprising data points more than minimum adjacent peak valley it Between comprising data points 3 times.

6. a kind of method for recognizing Raman spectrum exception measuring signal according to claim 1, which is characterized in that the step Threshold value in rapid three is determined according to practical application, is one times of standard deviation, twice of standard deviation or three times standard deviation.

7. a kind of method for recognizing Raman spectrum exception measuring signal according to claim 1, which is characterized in that the step Rapid six are specially when Signal to Noise Ratio (SNR) < 10, it is believed that this time is measured as invalid spectrum, it should be rejected.

8. a kind of method for recognizing Raman spectrum exception measuring signal according to claim 1, which is characterized in that the step In rapid eight as the standard deviation STD > 1.5, the exceptional spectrum " to peel off " in being measured in parallel is rejected.

9. a kind of method for recognizing Raman spectrum exception measuring signal according to claim 1, which is characterized in that the step In rapid seven, the satisfactory parallel determination spectrum of signal-to-noise ratio is no less than 3 times in same sample.