JPH0484280A - High precision peak recognition method - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

[Detailed description of the invention] [Industrial application field]

The present invention relates to a method for identifying spectral peaks with high accuracy (high reliability) from spectral data obtained by EPMA or the like.

[Conventional technology]

In addition to the detection signal obtained from the characteristic X-rays from the measurement sample, the spectrum obtained by EPMA etc. includes detection signals from noise and other X-rays incident from the surroundings, and also contains detection signals from the characteristic X-rays. Even in the case of a signal, there are statistical fluctuations due to random incidence, and unless a sufficient integration time is taken, spectral data with a smooth curve will not be obtained. Usually, measurement is performed with an integration time long enough to determine the peak, so the resulting spectrum has a jagged shape. Therefore, the spectral data was first subjected to smoothing processing to make it into an ideal curved shape without jagged edges, and then the portion that protruded from the background was recognized as the spectral peak.

[Question H that the invention attempts to solve]

In the conventional example described above, a prominent part of smoothed svector data is simply recognized as a peak, so a noise peak that happens to be present may be mistaken as a real peak, or conversely, it may be recognized as a peak. The present invention, which was not reliable enough because it sometimes missed true peaks below the screening level, could be improved by checking again.
The purpose is to further increase the reliability of detected peaks.

[Means to solve the problem]

Differentiate the smoothed spectrum data obtained by X-ray spectroscopy, etc., search for the point where the differential value changes from 10 to - (differential inflection point), and identify that differential inflection point as the peak apex. , the detected signal strength value at the peak apex is (PK), the half-width is estimated from the same (PK), and the detected signal strength values at points n times the half-width on both sides from the peak apex are calculated as background. Value [(BG-), (
BG+)], and the above peak signal strength value (PK) is (F'K) - [kX E G-), (BG
+)J/2±+ (B(,-) + (BG+) )
/2] When >0 is satisfied, (PK) is recognized as the peak value.

[For use]

The spectral data is smoothed, the smoothed spectral data is differentiated, a differential inflection point is searched, and it is determined whether a peak present at the searched differential inflection point is a detected peak. This judgment is made by comparing the baseline of the smoothed spectrum with a constant value (
Rather than detecting peaks by recognizing them as peaks if they stand out above (selection level), peaks are detected by inverting the sign of the differential signal, so even minute peaks are not overlooked. On the other hand, this may lead to excessive detection of peaks. Therefore, in the present invention, the following peak verification is further performed to confirm the peak by double detection. In radiation measurement, in which a material is irradiated with an electron beam and the generated radiation is detected by a detector, there is a statistical fluctuation error in the detected signal. This fluctuation error has a Gaussian distribution with respect to the direct value. Therefore, the error standard deviation σ for the radiation measurement value is σ = F10

[It becomes ``Tama-.'' From this, it becomes about 90% (87%) certainty if it is within the range of 1.5σ from the measured value.
When the measured value within the 1.5σ range no longer intersects with the background within the 1.5σ range, it is determined that the measured value is significantly different from the background, so the determination can be made using the following determination formula. . [<PK) -1,5f- and Go ``x-T] -E (
(BG-) + (BG 10) / 2 + 1.5 , J (PK) Misaki G-10 BG+))
/2, and the above formula is (PK) −[3X −10 BG+
)'7/2+ + (BG-)+ (BG+))/2]
>0, and based on experience such as experiments, this formula has the highest determination accuracy. As mentioned above, the detection signal PK at a point that has been recognized as a peak is added to the background at the peak position, which is interpolated from the detection signals at positions on both sides of the peak that are considered to be background parts. By determining whether it is more prominent than the square root of the ground plus times,
We are checking again to see if the initially recognized peak is the true peak. [Example] Figures 1 to 4 show an example of the data processing method of the present invention. Figure 1 shows an electron beam microanalyzer (EPMA).
The unprocessed spectral data obtained in the above steps is subjected to smoothing processing and converted into spectral data as shown in FIG. Differentiating this spectral data results in a differential curve (increase/decrease function curve) as shown in Figure 3. Since the top of the peak is the point where the change changes from increase to decrease, the differential inflection point is the point where the differential curve changes from 10 to -. ). Therefore, in the differential curve, the differential inflection point is searched, the detected signal strength value (PK) at the differential inflection point is checked and stored, and the detected signal strength (PK) at the peak point is determined as shown in FIG. Estimate the half-width corresponding to PK),
Check the detected signal strength value at a point three times the premature value width from the peak apex, and use the same strength value as the background value [(BG-)
, (BG0)], the above peak signal strength value (PK)
(PK) − [3X (BG −+ BG+>7/
2+ + (BG =) + CBG +>
If the inequality is satisfied by substituting 1/2 ko>0, (PK) is recognized as the peak value. The background measurement position will be explained. A normal spectral peak waveform also has a Gaussian distribution, but in terms of intensity, even at a point 1°5σ away from the direct value, it has an intensity of about 32% of the peak intensity, 3
At a point σ away, the intensity is about 1% of the peak intensity, so when measuring X-ray intensity using EPMA, measurement errors are usually reduced by taking into account issues such as device stability, sample smoothness, and cleanliness. The target is 1% or less. Therefore, by setting the background position at a point 3σ apart when there is no other peak in between, the influence of the peak can be almost eliminated. Note that the half width is approximately 1.2σ.

【effect】

According to the present invention, detection peak recognition has become more accurate.

[Brief explanation of drawings]

Figure 1 shows the unprocessed spectrum data of one embodiment of the present invention, Figure 2 shows the smoothed spectrum data of the above embodiment, Figure 3 shows the differential curve (increase/decrease function curve) of the above embodiment, and Figure 4 FIG. 2 is an explanatory diagram of data processing in the above embodiment.

Claims (1)

[Claims] Smoothed spectrum data obtained by X-ray spectroscopy, etc. is differentiated, a point where the differential value changes from + to - (differential inflection point) is searched, and the differential inflection point is peaked. The detection signal intensity value at the peak apex is set as (PK), the half-width is estimated from the peak apex, and the detected signal at a point n times the half-width on both sides from the peak apex is determined. The intensity value is divided into the background value [(BG-), (
BG+)], and when the above peak signal strength value (PK) satisfies ▲There are mathematical formulas, chemical formulas, tables, etc.▼, (PK) is recognized as the peak value. Certification law.