JPH0769216B2 - Method and device for identifying substances - Google Patents

Description

TECHNICAL FIELD The present invention relates to a method and an apparatus for identifying a substance using spectral information.

Background Identification and qualitative analysis of substances that analyze gases, liquids, and solids and determine what the substances are and what properties they have are important in a wide range of industrial fields. There are various analytical methods such as emission or absorption analysis, X-ray analysis, activation analysis, nuclear magnetic resonance, mass spectrometry, polarography, and gas chromatography. Among these methods, a method of analyzing spectral information showing properties peculiar to a substance is often used. In particular, in the qualitative analysis of organic compounds, analysis of infrared absorption spectrum is effective.

Hereinafter, a method for identifying a substance using an infrared absorption spectrum according to the present invention will be described.

2. Description of the Related Art Conventional techniques and their problems Conventionally, there are roughly two known methods for qualitative analysis using infrared absorption spectra. one,
This is a method of relying on empirical knowledge with the characteristic absorption band as a clue. That is, referring to the characteristic absorption band table, etc., select a specific atomic group, skeleton structure, etc., and further, while referring to the detailed characteristic absorption data table, estimate the structure so that there is no contradiction as a whole. This is a method of making a determination by comparing with the finally estimated standard spectrum of the substance. the other one is,
It is a mechanical method that relies on the collection of spectral data of known substances accumulated in the past. That is, it is a method of sequentially comparing the sample spectrum of the unknown substance with the standard spectrum to find one having a high degree of similarity. This requires an enormous collection of standard spectral data (tens of thousands to hundreds of thousands) and means for quickly selecting only a specific spectrum from the collection. Although the former method has been and is still used for qualitative analysis, computer utilization in the field of chemistry has rapidly progressed due to the increase in computer memory requirements and computational speed. It became more widely used.

In spectrum analysis using this computer,
To identify unknown substances, a file search method that compares and collates a file (database, called a spectrum library) that encodes a huge amount of spectral data of known substances with the spectral data of unknown materials is used. As the identification method by the file search method using a computer, a correlation method, a maximum likelihood estimation method, a linear discriminant function method, a shortest distance method, etc. are known. However, the method of
(A) A priori information (such as noise variance) is required,
(B) A large amount of calculation (In order to search a huge library, it is desirable that the amount of calculation is small and high speed is possible)
However, there is a problem in this respect, and the correlation method of is now the mainstream.

Then, in order to further increase the calculation speed and reduce the storage capacity, i) a method of using a part of the interferogram (LVAzarrge, RRWilliams, and JAdeHaseth, App
l.Spectrosc. 35, 466 (1981) , PMOwens, and TLIsen
hour, Anal.Chem. 55, 1548 ( 1983)) and, ii) a method using encoding data on the absorption peak position (CSRann, Ana
l.Chem. 44 , 1669 (1972)), for the purpose of improving accuracy,
iii) A method that uses peak position and intensity data of the spectrum (K. Tanabe, et al. Anal. Chem. 47 , 118 (1975)) is used, and the high density of recent semiconductor memories (ROM and RAM) is used. And iv) Method of using all spectrum data regardless of data compression due to rapid expansion of storage capacity such as optical memory (CD-ROM) (for example, G. Hangac, RCWieboldt,
RBLam, and TLIsenhour, Appl.Spectrosc. 36 , 40 (198
2)) etc. have been proposed.

The conventional correlation method calculates the correlation between two spectral waveforms h (ν) and g (ν) in order to know the relationship between them. The cross-correlation Υ _gh (ν) is given by Υ _gh (ν) = g (ν) ★ h ^* (ν) = ∫g (ν ′ + ν) · h ^* (ν ′) dν ′ (1). However, ^* represents a complex correlation and ^* represents a complex conjugate. In equation (1), when the Foue transforms of g (ν) and h (ν) are represented by uppercase letters, and t = 1 / ν, and the inverse Fourier transform operation is represented by F ⁻¹ (·), Υ _gh (ν ) = F ⁻¹ [G (t) · H ^* (t)] (2) However, the above g (ν) and h (ν) are standardized in advance by the square norm. In the conventional correlation method, as shown in FIG. 1, the value at the origin of the correlation function of equation (2), η = Υ _gh (0) = ∫g (ν) · h ^* (ν) dν = ∫G Let (t) · H ^* (t) dt (3) be the similarity between g (ν) and h (ν), and use the spectral library for the spectral data h (ν) of the test material, which is an unknown substance. Reference spectral data gi in
= 1,2, ..., N specifies the substance i that gives the maximum η.

However, the conventional methods of i), ii), iii), and iv) above have a drawback that the discrimination ability is inferior to the similar spectral waveforms, and the methods of ii), iii), and iv) are used. File search error, that is, search due to peak position or fluctuation or peak intensity fluctuation due to spectrophotometer accuracy, cell thickness or slit width, or peak detection error due to fluctuation of peak detection sensitivity due to threshold level There is a problem that a lot of noise is contained in, and since there are almost no peaks for a broad spectrum waveform where the maximum peak position is not clear,
There is a problem that it is difficult to apply this method.

The object of the present invention is to solve the problems of the conventional method,
It is an object of the present invention to provide a novel method and apparatus for identifying a substance, which has excellent discrimination ability even for similar spectral waveforms and is not affected by the measurement conditions of a sample of an unknown substance.

SUMMARY OF THE INVENTION The present invention focuses on the finding that a large amount of characteristic information of a spectral waveform is contained in the phase portion, and as shown in the principle diagram of FIG. 2, the spectral data g (ν) of a known substance.
Fourier transform is performed to extract the phase component G _φ (t) from this Fourier transform information G (t), and the Fourier transform of the spectral data h (ν) of the unknown substance is performed to obtain the phase from this Fourier transform information H (t). Extract the component H _φ (t),
The basic feature is that only the phase components G _φ (t) and H _φ (t) are correlated.

When formulated, it is given by the following equations (4), (5), and (6).

_{Υ φφ (ν) = g φ} (ν) ★ h * φ (ν) = F -1 [G φ (t) · H φ * (t)] = F -1 [exp {jφ g (t) · exp {-Jφ _h (t)] = F ^-1 [exp {jφ _g (t) -φ _h (t))}] (5), the correlation function Υ _φφ (ν) of only the phase term is defined. To do. Note that g _φ (ν) and h _φ (ν) are exp {jφ
_It is an inverse Fourier transform of _g (t)}, exp {jφ _h (t)} and is generally a complex function. Υ _φφ (ν) is hereinafter referred to as a phase correlation function.

The similarity η _φ is defined by a value at the origin of the phase correlation function, for example, as in the following equation.

_{η φ = Υ φφ (0)} = ∫exp {jφ g (t) -φ
_h (t)) (6) As described above, the square of the object of correlation is squared because the amplitude is set to 1 (| G _φ (t) | = | H _φ (t) | = 1). Normalization by the norm is performed as shown in equation (7).

∫ | G _φ (t) | ² dt = n, ∫ | H _φ (t) | ² dt = n (7) Here, n is the number of measurement points. Note that this standardization is automatically performed by a computer.

Action According to the method of the present invention, the spectrum g of the known substance
If (ν) and the spectrum h (ν) of the unknown substance are different, the phase correlation function Υ _φφ (ν) becomes white, while if they are the same, it becomes a Dirac delta function, and the substance difference can be identified very easily. .

At the time of identification, either human inspection or machine identification may be used. In the former case, the phase correlation function Υ _φφ
(Ν) may be displayed on a display device in a graph format or a numerical format, or the result data may be printed out by a printing device. In the latter case, the similarity η _φ = Υ _φφ (0) is obtained as in the above equation (6), and the magnitude of this value may be determined to make the determination. The latter allows the series of processes to be fully automated.

Also, in equation (6) above, the similarity η _φ is taken as the value at the origin of the phase correlation function, but the observed spectrum has a wavelength (or wave number) due to the response delay of the observation system (electrical and mechanical system).
If there is an error due to deviation, the peak position of the phase correlation function may deviate from the origin, but in consideration of this deviation, the maximum value of Υ _φφ (ν) may be defined as the similarity. Further, the sum of the phase correlation function and energy near the origin may be used as the measure of similarity.

EXAMPLES Hereinafter, the present invention will be specifically described with reference to examples shown in the accompanying drawings.

FIG. 3 is a process explanatory diagram of one example, showing a method of identifying a substance using an infrared absorption spectrum.

(11) is an infrared absorption spectrum library with different N
The individual spectrum data are stored. (21) is an infrared spectrophotometer for measuring a sample (20) of unknown substance.

An unknown sample (20) is measured by an infrared spectrophotometer (21) to obtain spectral data (22). Spectral data (2
2) Fourier transform is performed by the Fourier transform means (23) to obtain Fourier transform data (24). In the Fourier transform data (24), only the phase term is extracted by the phase extracting means (25), and becomes the phase component data (26). This data (26) is conjugated by the means (29) for taking the following complex conjugate to become complex conjugate data (30).

On the other hand, one spectral data (12) is selected from the library (11). The selected spectrum data (12) is Fourier transformed by the Fourier transform means (13) to obtain Fourier transform data (14). Next, this Fourier transform data (14) is used as phase extraction means (15).
Only that phase term is extracted by and the phase component data (1
6)

The phase component data (16) is multiplied by the complex conjugate data (30) by the multiplier (31). The data of the multiplication result is subjected to the inverse Fourier transform by the inverse Fourier transform means (32) to obtain the phase correlation function Υ _φφ (ν). Then, the processing means (33) calculates the value of the phase correlation function at the origin to obtain the similarity η _φ . The similarity data is output by the output means (34). Further, as is clear from the expression (6), the value Υ _{φφ at} the origin of the phase correlation function
To obtain only (0), the inverse Fourier transform means (32)
May be omitted and the multiplication result by the multiplier (31) may be directly integrated.

Additional spectral data (1
2) is read out, and the same process is repeated until the similarity η _φ close to 1.0 is detected. Output means (3
In 4), the substance name corresponding to the spectrum data hit by the search may be output without the need to output the similarity data.

FIG. 3 shows a specific example of this embodiment.

As shown in FIGS. 4 (a1) and 4 (b1), this is an example of Benzene (benzene) and Toluene (toluene) having similar spectral waveforms. The spectral data of benzene is known as g (ν), and the spectral data of toluene is unknown as h (ν).

Fig. 4 (a2) and (b2) show the respective Fourier amplitude distributions | G
(T) |, | H (t) | It can be seen that both have low-frequency filter structures. As a result, FIG. 4 (d1), (d
As shown in 2), in the conventional correlation method using the equation (1),
Both autocorrelation (Υ _hh ) and cross-correlation (Υ _gh ) have similar shapes, and Υ _hh (0) = 1.00 and Υ _gh (0) = 0.
Since the difference is 72, the discrimination ability is poor because the difference is not large.

On the other hand, if correlation is obtained using only the phase information exp {jφ _g (t)} and exp {jφ _h (t)} shown in (c1) and (c2) of FIG. As shown in (d2),
The autocorrelation is almost a delta function, while the crosscorrelation is white. Υ _hφhφ (0) = 1.00 for Υ
_{Since gφhφ} (0) = 0.12, the difference is extremely large and the substances can be easily identified.

The difference in slit width, sampling interval, etc. affects only the Fourier amplitude information (FIGS. 4 (b1) and (b2)), and the phase correlation function remains unchanged. Further, deviation of the peak position is not a no problem by the maximum value of _Υ gφhφ (ν) as the similarity eta _phi. Further, even with respect to broad spectrum data in which the peak position is not clear, the broadness appears only in the amplitude term, and therefore the discrimination sensitivity due to the phase correlation does not decrease.

In the above embodiment, no consideration is given to noise. However, when noise is added to the spectrum, in this phase correlation method, the amplitude is replaced with 1 in a region (generally, high frequency Fourier region) where the amplitude of the Fourier transform component of the reference spectrum or the input spectrum is small, so that the noise component is It can be emphasized and thereby influence the correlation result.

Therefore, the noise resistance of the phase correlation method is improved by means of suppressing noise in the Fourier region where the amplitude is small.

As an example of the noise suppressing means, a filter for performing an operation to make the amplitude zero is added in a Fourier region having an amplitude smaller than a certain threshold value Lth as in the following equations (8) and (9).

An embodiment in which the operation is added is shown in FIG. The same reference numerals as in FIG. 3 indicate the same or similar elements. In FIG. 5, a noise suppression filter (17) based on the equation (8) is added, and a noise suppression filter (27) based on the equation (9) is added. Is taken. Similarity η ′
_φ is given as a value at the origin of the phase correlation function _Υ'φφ (ν), for example.

FIG. 6 shows an example of the noise removal filter processing using the equations (8) and (9), and the case of using benzene. In the figure, (a) is the infrared absorption spectrum of benzene, (b) is the amplitude distribution of its Fourier transform, (c) is the phase distribution, and (d) is the amplitude distribution after processing according to equation (9).

Even if the noise due to measurement is negligible, the rounding error of the computer becomes a problem, but this operation is also effective in terms of this problem. Further, when performing control based on this amplitude information, it is necessary to consider normalization, which is slightly different from the phase correlation method shown in FIG. That is,
The frequency band equation for correlation has only a region where G ′ _φ (t) is not 0 and the input H ′ _φ (t) is not 0. If this is n, η ′ _φ needs to be standardized by n. is there.

Next, a specific example of the method shown in FIG. 5 will be shown.

Eight types of liquid samples having the following numbers 1 to 8 were used as samples.

1.o-xylene (ortho-xylene) 2.m-xylene (meta-xylene) 3.p-xylene (para-xylene) 4.Benzene (benzene) 5.Toluene (toluene) 6.Hexane (hexane) 7. Cyclohexane 8.Methyl ethyl ether For measurement of infrared absorption spectrum, M manufactured by Perkin-Elmer Co.
ODEL983 Out-of-seat absorption spectrophotometer, manufactured by MODEL 3600
Via DATA STAION and RS-232C interface,
The measurement data was sent to NEC-9801F manufactured by NEC, stored in a disk medium, and each processing operation was performed by a program on PC-9801F. The PC-9801F is equipped with an 8087 Numerical Processor (NDP), and FFT is designed to use this NDP as much as possible.
Machine language subroutines such as (Fast Fourier Transform) were created to speed up the calculation process. As a result, 1
The time required to perform the correlation calculation between the spectra of one unknown sample and the spectrum data of eight kinds of libraries was about 25 seconds.

The similarity η, η ′ _φ between the conventional correlation method based on the equation (1) and the phase correlation method according to the present invention was calculated for all combinations using the above eight kinds of spectra. The results of this calculation are shown in Table 1 and Table 2. The numerical values are shown in%.

Then, the numerical data in Tables 1 and 2 are visualized three-dimensionally,
FIG. 7 shows a matrix form.

In FIG. 7, when the spectrum of the same substance is on the diagonal line at the center, that is, g (ν) and h (ν) are the boundaries, the left half shows η and the right half shows η ′ _φ , and the front is respectively From the column of, the substances of sample numbers 1, 2, ..., 8 correspond. The output of the phase correlation method eta _'phi autocorrelation (central diagonal) very small value is outside, the identification sensitivity of the substances according to the spectrum compared to the conventional correlation method is very high.

Tables 3 and 4 show the calculation results by the correlation method between the spectrum data measured with different slit widths. Two slit widths, 0.43 mm and 1.20 mm, samples are o-, m-, p
-Is a xylene isomer. Numerical values in the table are shown in%.

FIG. 8 shows the numerical data of Table 3 and Table 4 visualized and summarized as a three-dimensional matrix. In FIG. 8, even in the phase correlation method, some decrease in the value is recognized between the spectrum data under the measurement conditions in which the slit widths are different, but when considering the result of FIG. ), It is a sufficiently negligible amount, and the phase correlation method can identify substances with sufficient accuracy even between spectral data measured with different slit widths.

In the examples shown in FIGS. 3 and 5 above, the library was a library of spectral data, but a library of phase information was constructed and the identification search of unknown substances was performed by this database. Good. This greatly improves the search speed. Also, by quantizing the phase part of the Fourier component of the spectrum data in this way, it is possible to significantly compress the data. It is necessary to allocate 10 to 12 bits to the peak information of conventional spectrum data, but if only phase information (of course standardized and stored), 1 to 2 bits are needed, about 1/5 to 1/10 of the conventional Can be compressed to.

In the above embodiment, the spectrum information is the infrared absorption spectrum, but the spectrum information is not limited to this. In emission spectrum, mass spectrum, Raman spectrum, N
The method of the present invention can be similarly applied to the MR spectrum and any other spectrum.

Further, the substance referred to in the present invention is not limited to a single substance,
A substance obtained by mixing two or more single substances is also included. However,
The mixed substances are subject to a constant mixing ratio.
It goes without saying that the substance may be solid, liquid or gas.

EFFECTS OF THE INVENTION As described above, according to the present invention, the substance can be identified regardless of the measurement conditions of the spectrum, and since the identification ability is extremely high, it is possible to clearly discriminate between similar spectra, and further, the peak information and the amplitude can be distinguished. Since we do not use information,
Broad spectra can be treated exactly the same as those that do not, and there is an excellent effect that accurate results can be obtained without a nozzle particularly in a search for identifying a substance.

[Brief description of drawings]

FIG. 1 is an explanatory view of a conventional method, FIG. 2 is an explanatory view of the principle of the method of the present invention, FIG. 3 is a process explanatory view of an embodiment according to the present invention, and FIG. 4 is a specific example of the embodiment. FIG. 5 is a waveform diagram for explaining, FIG. 5 is a process explanatory diagram of another embodiment according to the present invention, FIG. 6 is a waveform diagram for explaining noise suppression processing, and FIG. 7 is contrasted with a conventional correlation method. FIG. 8 is an explanatory diagram of the effect of the phase correlation method according to the present invention, and FIG. 8 is an explanatory diagram of the effect of the present invention shown in contrast to the conventional method when the slit width is changed among the measurement conditions. 12 …… Spectrum data of known substance, 22 …… Spectral data of unknown substance, 13,23 …… Fourier transform means, 15,
25 ... Phase information extraction means, 16, 26 ... Phase data, 31
...... Multiplying means or multiplier, 17,27 …… Noise suppressing filter means, 18 ′, 28 ′ …… Noise suppressing phase data, 3
4 …… Similarity output means.

 ─────────────────────────────────────────────────── --Continued from the front page (56) References JP-A-60-98335 (JP, A) JP-A-62-38324 (JP, A) JP-A-61-148331 (JP, A)

Claims (4)

[Claims] 1. A step of Fourier-transforming spectral information of a known substance to extract its phase information, a step of Fourier-transforming spectral information of an unknown substance to extract its phase information, and a phase of the known substance. A method of identifying a substance, which comprises a step of correlating information and phase information of the unknown substance, and discriminates whether the unknown substance is the same as or different from the known substance according to a result of the step of obtaining the correlation. 2. A step of Fourier transforming spectral information of a known substance to extract its phase information, a step of suppressing a noise component of the phase information, and a Fourier transform of spectral information of an unknown substance to obtain its phase. A step of extracting information, a step of suppressing a noise component of the phase information of the unknown substance, and a step of correlating the noise-suppressed phase information of the known substance with the noise-suppressed phase information of the unknown substance. A method for identifying a substance, which comprises the step of identifying the unknown substance as the same or different from the known substance according to the result of the step of obtaining the correlation. 3. A means for Fourier-transforming spectral information of a substance to extract phase information thereof, a means for cross-correlating the phase information extracted by the means, and a part or all of an operation result of the means. A device for identifying a substance, which comprises a means for outputting. 4. A means for Fourier transforming spectral information of a substance to extract phase information thereof, a means for suppressing a noise component of the phase information extracted by the means, and a phase for suppressing the noise component by the means. An apparatus for identifying a substance, comprising means for cross-correlating information, and means for outputting a part or all of an operation result of the means.