JPH1164217A - Component quantity detecting device for spectral analyzer - Google Patents

Abstract

PROBLEM TO BE SOLVED: To detect a component quantity of a to-be-measured sample based on a weighted factor by generating a linear regression model wherein a near- infrared spectrum of an actual target food is weighted with the component quantity, and applying the model to the actually-measured near-infrared spectrum. SOLUTION: The detecting device comprises a linear regression model A wherein a spectral function of a sample is a linear sum with its component spectral function weighted with a component quantity, a spectrum measuring means B wherein a spectrum of a sample is measured with a near-infrared ray spectral analyzer for obtaining an actual-measured data, and a component quantity calculation means C wherein a component quantity weighted by a method of least square by applying the actual-measured data to the linear regression model A.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a spectrometer for analyzing the composition of a sample using an absorption spectrum, and more particularly to a device for detecting the amount of the component.

[0002]

The near-infrared spectrum is a superposition of harmonics based on the non-linearity of molecular vibration, and is extremely complicated and unsuitable for analysis. However, the introduction of statistical methods has made it possible to quantify some substances, and because of their high permeability, they have been applied to nondestructive testing of foods, and they have come to the forefront.

However, an accurate analysis of the relationship between the amount of transmission or reflection and the content of a substance has not been made. Therefore, the content of the target component is accurately analyzed by a conventional method, though it is a statistical method. Using a sample for the preparation of a calibration curve, a multiple regression analysis is only used to detect the dose by trial and error, which quantifies the relationship between the spectrum and the component amount so that the measurement error is minimized. At present, the unacceptably low accuracy of detection for many important food components is believed to be due to the fundamental limitations of methods that rely solely on this statistical method. In statistical analysis of only the detected spectrum, it is in principle impossible to further improve the accuracy.

[0004] In ordinary foods, most of the components are composed of only a few types of substances, and therefore, as shown in FIG. 1, the spectrum of the target substance also includes these types of components A, B, and C. Can be considered as a function of the spectrum of The near-infrared spectrum of the pure substance of the component has a very complicated appearance due to interparticle reflection and diffusion inside, but it is possible to store the shape as it is in the computer memory as a function of frequency or wavelength. it can.

[0005] The near-infrared spectrum of a target food, which is an actual composite substance, is considered to be a linear sum (including interaction) weighted by the component amounts of these functions (which may be called shapes). When this model is applied to the measured near-infrared spectrum by the least square method, a weighted coefficient is calculated, and the component amount is detected from this coefficient.

Therefore, the present invention creates a linear regression model in which the near-infrared spectrum of the actual target food is weighted by the component amount, applies this model to the actually measured near-infrared spectrum, and obtains the data of the object to be measured from the weighted coefficients. The purpose is to detect the component amount.

[0007]

In order to achieve the above object, the present invention is configured as follows.

That is, in a spectroscopic analyzer for analyzing the component composition of a sample using a near-infrared spectrum, a linear regression model in which the spectrum of the sample is a linear sum obtained by weighting the spectrum of the pure substance of the component by the component amount, A spectrum measuring means for measuring a spectrum of the sample to obtain measured data; and a component amount calculating means for calculating a component amount weighted by a least square method by applying the measured data to the linear regression model. Is a component amount detecting device.

[0009]

Embodiments of the present invention will be described below with reference to the drawings.

FIG. 2 shows a configuration diagram of a component amount detection device embodying the present invention. The component amount detection device includes a linear regression model A in which the spectrum function of the sample is a linear sum in which the component spectrum function is weighted by the component amount, and a spectrum in which the spectrum of the sample is measured by a near-infrared spectrometer to obtain measured data. It comprises a measuring means B and a component amount calculating means C for applying the measured data to the linear regression model A and calculating the component amount weighted by the least squares method.

The linear regression model A has a spectrum function f (λ), a spectrum function f (λ), a wavelength λ (nm), and the number of detection spectra of a sample to be analyzed, which is a composite substance of components. n (usually 1,050)
It is assumed that the spectral function f (λ) of the sample to be analyzed has a linear relationship as a linear sum obtained by weighting the spectral function fi (λ) of the component i shown in Expression 1 with the component amount αi.

(Equation 1) Next, a deviation between the straight line and the sample value is represented by ei, and a linear regression model represented by Expression 2 is set.

(Equation 2)

FIG. 3 shows a configuration diagram of a near-infrared spectrometer as spectrum measuring means B embodying the present invention. The near-infrared spectrometer 1 includes a spectroscope 2 that generates a measurement light beam,
It comprises a sample section 3 for storing a sample, a light projecting section 4 for projecting a measurement light beam onto the sample, a light collecting section 5 for condensing the reflected light, and an analyzer 6 for analyzing the collected reflected light. You.

The spectroscope 2 includes a light source 7, a lens 8, a chopper wheel 9, and a slit 10, and these are arranged in a line on the optical axis a of the light source 7. The chopper wheel 9 is arranged at a position where the circumferential end of the chopper wheel 9 blocks the optical axis a.

The light projecting section 4 includes a reflecting mirror 11 and a light projecting lens 12
The reflecting mirror 11 is installed on the optical axis a at an angle of 45 ° with respect to the optical axis a, and the light projecting lens 12 is arranged on the reflected optical axis b of the optical axis a.

The condensing unit 5 is composed of a converging convex mirror 13, a photoelectric sensor 14, and a condensing concave mirror 15, and these are arranged in a line on the reflection optical axis b. The photoelectric sensor 14 is arranged at the focal position of the concave concave mirror 15.

The analyzer 6 includes an amplifier 16 and an A / D converter 1
7 and a CPU 18, and each of them is electrically connected.

The spectroscopic analyzer embodying the present invention has a configuration as described above, and a sample is put into the sample container of the sample section 3 and the sample is measured as follows. First, the light emitted from the light source 7 in the spectroscope 2 is converted into a parallel light by the lens 8.
After the parallel rays are periodically divided by the rotation of the chopper wheel 9, the parallel rays are converted into high-purity monochromatic light through the slit 10.

The monochromatic light is applied to the light projecting surface of the sample section 3,
Diffuse reflected light from the sample at each wavelength is condensed by the condensing concave mirror 15, and the condensed diffuse reflected light is further condensed by the condensing convex mirror 13 and projected to the photoelectric sensor 14.

The diffuse reflection light received by the photoelectric sensor 14 is:
The signal is amplified by the amplifier 16, converted into a digital signal by the A / D converter 17, the absorbance is calculated by the CPU 18, arranged in order of wavelength, and the measured data of the near-infrared spectrum is obtained.

In order to minimize the error of the linear regression model shown in Expression 2, the component amount calculation means C sets the number of components i to m (5 for amylose, amylopectin, protein, fat, and water). The unknown coefficient αi that minimizes the sum of the squares of the error shown is determined.

(Equation 3) That is, the measured data of the near-infrared spectrum measured by the spectrum measuring means B is applied to Expression 3, a coefficient αi that minimizes F is obtained by the least square method, and the component amount is detected from this coefficient. Strictly, since the coefficient αi is proportional to the component amount, it is necessary to determine a proportional constant from the result of chemical analysis of a specific type of component. Note that this method can be applied to spectra processed in various ways.

[0021]

According to the component amount detecting apparatus of the present invention, a linear regression model in which the near-infrared spectrum of an actual target food is weighted by the component amount is created with the above-described configuration, and this model is used to measure the measured near-infrared light. The least squares method is applied to the spectrum, a weighted coefficient is calculated, and the component amount is detected from the coefficient, thereby detecting the component amount of the data to be measured. Therefore,
According to the present invention, there is a rational model for associating the spectrum of the target substance with the spectrum of the component, and there is no trial and error element in the analysis, so that objective and highly accurate detection can be performed. Further, since the entire spectrum of the component is added to the spectrum of the target substance as the data of the analysis, the amount of information regarding the analysis is significantly increased, and the reliability of the result is high. On the other hand, the analysis program has features that it is simple and requires little calculation time. Further, since the present invention can be applied to a processing spectrum, it has an effect of contributing to improvement of a conventional method.

[Brief description of the drawings]

FIG. 1 is a diagram showing that a sample spectrum is a superposition of component spectra.

FIG. 2 is a configuration diagram of a component amount detection device embodying the present invention.

FIG. 3 is a configuration diagram of a spectroscopic analyzer as a spectrum measuring unit embodying the present invention.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Near-infrared spectrometer 2 Spectroscope 3 Sample part 4 Projection part 5 Condensing part 6 Analyzer 7 Light source 8 Lens 9 Chopper wheel 10 Slit 11 Reflection mirror 12 Projection lens 13 Condensing convex mirror 14 Photoelectric sensor 15 Condensing concave mirror Reference Signs List 16 amplifier 17 A / D converter 18 CPU A component amount calculation model B sample spectrum measurement means C component amount calculation means

Claims (1)

[Claims] 1. A spectroscopic analyzer for analyzing a component composition of a sample using a near-infrared spectrum, comprising: a linear regression model in which a spectrum of the sample is a linear sum obtained by weighting a spectrum of a pure substance of the component by a component amount; A spectrum measuring means for measuring a spectrum of the sample to obtain measured data; and a component amount calculating means for applying the measured data to the linear regression model to calculate a component weight weighted by a least squares method. And a component amount detection device.