JP2010216890A - Device and method for detection of different kind of product - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To simply detect with a high precision a different kind of product contained in each of the products with a significant quantity for each kind. <P>SOLUTION: When applying a device and method for detection of the different kind of products to the detection of different kinds of drugs, a pulse-like terahertz light 39 is injected to a tablet 7 being conveyed from the one face, and it is utilized that a temporal waveform 32 of an electric signal 25 of the terahertz light 24 transmitted through this drug is different depending on the material and shape of the tablet 7. Further, whether the tablet is an authentic or a different kind is determined using a plurality of statistical quantities 44 (Mahalanobis' generalized distance) for which the difference of the temporal waveform 32 for each tablet is set for the corresponding temporal waveform. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to a heterogeneous product detection apparatus and a heterogeneous product detection device for detecting that a foreign product (a foreign drug, a different raw material, a poison, etc.) is contained in a regular product (drug or food) in a pharmaceutical factory or a food factory. Regarding the method.

  In general, pharmaceutical factories and food factories, etc., carry out work to detect foreign substances taking into account cases where foreign substances (heterologous drugs, raw materials, poisons, etc.) are mistakenly included in legitimate products (drugs and foods). It has been broken. Hereinafter, in the present specification, a drug is used as an example of a product and a different drug is used as an example of a foreign substance, but the product and the foreign substance are not limited to this example.

  Now, a wide variety of drugs are manufactured in pharmaceutical factories and the like. These drugs may have similar shapes, colors, and weights, and there are very few drugs among many regular varieties when packaging many kinds of drugs with PTP (press through packing) on a packaging machine. There is a risk of mixing different drugs. In this case, for example, different types of drugs are mixed in one PTP container (sheet) in which 10 to 20 drugs of the same type are packaged.

  Usually, in the final packing and packing process of a factory, visual inspection is performed by an inspector, and the difference between drugs is selected based on the color and shape of the drug to prevent mixing of other types of drugs, drug components, and substances other than drugs. is doing.

  Recently, a tablet image on a PTP sheet is read with a CCD camera to determine the area, diameter, and color of each tablet, and compared with the area, diameter, and color of the reference tablet to determine whether it is a different tablet. Is adopted.

  The tablet on the PTP sheet is irradiated with light from a light source having an emission spectrum in the near infrared (1 to 2 nm) such as a halogen lamp. The near-infrared spectrum of the reflected light from each tablet is measured by scanning a plurality of tablets with a galvanometer mirror from the viewpoint of the collimated near-infrared spectrometer. A method of detecting a different tablet by comparing this spectrum with a near-infrared spectrum of a reference tablet that has been measured in advance is employed.

  Furthermore, the “terahertz light” having a wavelength between 30 gigahertz and 5 terahertz between the radio wave and the light is irradiated on the object to be measured, and the spectrum waveform of the “terahertz light” transmitted through the object to be measured is measured. A technique has been proposed for detecting foreign matter contained in the measured object by performing spectrum analysis of the waveform of the terahertz light transmitted through the measured object using characteristics that change depending on the material of the measured object. (Patent Document 1).

Furthermore, in the “specimen inspection apparatus and specimen inspection method” of Patent Document 2, the above-described terahertz light is irradiated to the medicine inserted in each recess of the PTP sheet, and the spectrum of the terahertz light transmitted through the medicine is obtained. This spectrum is compared with the reference spectrum of a known tablet to determine whether the drug is a legitimate drug or a different drug.
JP 2005-114413 A International Publication WO2008 / 001785

  However, there is a possibility of oversight in the case of discrimination of medicines by conventional visual inspection. Further, when the medicine has a shape that is difficult to distinguish from the appearance, even if image processing using a CCD camera is performed, 100% safety cannot be ensured.

  In addition, the above-described technique for performing near-infrared analysis has been difficult to apply to film-coated tablets and sugar-coated tablets because of spectrum measurement only on the drug surface.

  There are an enormous number of types and types of drugs, and many types of drugs are often produced on the same production line, so drugs that are formed in almost the same shape and color in the same production line are produced. The number of cases is increasing. However, in the method of spectrum analysis using terahertz light, it is difficult to separate the transmitted light of each drug and the background light of its peripheral part. As a result, the terahertz light mixed with the drug transmitted light and the peripheral background light is difficult. The spectrum is calculated from the time waveform, but due to the interference between the drug transmitted light and the peripheral background light, the waveform 101 is swelled 101 as shown in FIG. .

  Further, there is a problem that data cannot be read as one waveform because the medicine is transported and moved while the time waveform is being read. In particular, in order to increase the wave number resolution of the spectrum, it is necessary to lengthen the measurement time of the time waveform, and there is a problem that the influence of the movement due to the drug delivery becomes large.

  The present invention has been made in view of such circumstances, and an object of the present invention is to provide a heterogeneous product detection apparatus and a heterogeneous product detection method capable of reliably detecting different products while being conveyed. And

  In order to solve the above-mentioned problem, a first aspect of the present invention includes a terahertz light incident portion that emits pulsed terahertz light from one surface of a product in a transported state along a transport path, the product, and the product. A terahertz light detector that receives terahertz light transmitted through the peripheral portion, converts it into an electrical signal, and outputs it, and a time for reading the time waveform of each transmitted terahertz light converted into an electrical signal output from this terahertz light detector A feature amount for extracting a plurality of feature amounts based on a predetermined zero cross point, a waveform peak point, a waveform peak height, and a waveform width in each time waveform read by the time waveform reading portion. An extraction unit, a feature amount data table storing each feature amount extracted by the feature amount extraction unit in a reference product of a type designated in advance, and a plurality of feature amounts from the plurality of feature amounts For each combination of feature quantities generated by the feature quantity combination generation means among the feature quantities stored in the feature quantity data table, the feature quantity combination generation means for enumerating all combinations, each feature quantity of the combination The standard Mahalanobis distance calculating means for calculating the Mahalanobis distance for each combination calculated in step (2) and selecting the feature quantity combination that maximizes the S / N ratio and calculating the Mahalanobis distance corresponding to the selected feature quantity combination. The combination selection means for writing the optimum reference data for the Mahalanobis calculation into the optimum reference data table for the Mahalanobis calculation, and when the product to be inspected is input, the Mahalanobis distance in the reference space of the designated product of the product is stored in the optimum reference data table for the Mahalanobis calculation. Mahalano for inspection calculated using the stored optimum standard data for Mahalanobis calculation A scan distance calculating means is a heterogeneous product detecting device having a heterogeneous determination means for performing a heterogeneous determination of the product based on the calculated Mahalanobis distance.

  In the second aspect of the present invention, the product is inserted into a recess formed in the width direction of the packaging container, and the terahertz light incident portion is arranged in the width direction of the packaging container and is simultaneously terahertz light. The terahertz light detection unit includes a detection head arranged in the transport direction and simultaneously receiving terahertz light, and a delay time initial value of the terahertz light detection unit for each type of product. And a delay time fluctuation range, a trigger signal generating means for generating a signal at a timing when the product reaches a predetermined distance from the center position of the terahertz light beam, and the trigger signal is generated. And a means for recording a time waveform from the determined timing.

  In addition, although the aspect of "apparatus" was demonstrated as each aspect of this invention, you may implement | achieve as an aspect of not only this but a "method".

  In the heterogeneous product detection apparatus and the heterogeneous product detection method of the present invention, the pulsed terahertz light is incident from one side of the transported product, and the time waveform of the electrical signal of the terahertz light transmitted through the product is the product material and It uses different things depending on the shape. Then, using a plurality of feature quantity statistics (Mahalanobis distance) in which the difference in time waveform of each product is set in the corresponding time waveform, it is determined whether the corresponding product is a legitimate product or a different product. Thus, different products can be reliably detected while being conveyed.

  Hereinafter, an embodiment of the present invention will be described with reference to the drawings.

  FIG. 1 is a schematic configuration diagram of a heterogeneous drug detection apparatus to which a heterogeneous product detection method and a heterogeneous product detection apparatus according to an embodiment of the present invention are applied. In this embodiment, the foreign drug detection device is incorporated in the PTP packing machine 1 of the drug manufacturing factory.

  In the PTP packing machine 1, the packaging sheet (hereinafter simply referred to as a sheet) 3 supplied from the sheet supply unit 2 is formed two-dimensionally with recesses 4 at predetermined intervals. It is conveyed in the direction of the middle arrow. The medicine supply unit 6 stores each medicine (hereinafter referred to as a tablet 7) accumulated therein one by one in the concave portion 4 of the sheet 3 that is sequentially delivered to the conveyance path 5. Then, at the right end of the conveyance path 5, the upper surface of the recess 4 in which the tablet 7 is accommodated is covered with an aluminum sheet 8, which is cut into a predetermined length by a cutting machine 9, and a PTP sheet (tablet sheet) 10. As discharged.

  In the heterogeneous drug detection device incorporated in this PTP packing machine 1, the femtosecond laser generator 11 is a near red having a pulse width of 10 fs to 100 fs in a cycle of 50 MHz to 100 MHz based on an instruction from the femtosecond laser control unit 12. External light (700 nm to 1600 nm) 13 is oscillated. Near infrared light 13 emitted from the femtosecond laser generator 11 is branched in two directions by a beam splitter 14. As shown in FIG. 2, one of the branched near-infrared light 13a is further branched by an optical amplifier / distributor 15a and made incident on a plurality of terahertz light emitting units 16.

  In each terahertz light emitting unit 16, a photoconductive antenna 17 is housed. The photoconductive antenna 17 is applied with a bias voltage 33 from the drive control circuit 31 of the control device 27, and the pulse-like near red color described above. When the external light 13a is applied, pulsed terahertz light is emitted. The emitted terahertz light is irradiated by the condenser lens 18 onto the tablet 7 accommodated in the concave portion 4 of the sheet 3 being conveyed on the conveyance path 5 at a constant speed.

  The terahertz light transmitted through the tablet 7 from the upper side to the lower side is incident on the terahertz light receiving unit 21 via the condenser lens 20 in the terahertz light receiving unit 19.

  The other near-infrared light 13b branched by the beam splitter 14 described above is delayed by an optical delay unit 22, and then further branched by an optical amplifier / distributor 15b as shown in FIG. The light enters the terahertz light receiving unit 21 in the light receiving unit 19.

  As shown in FIG. 3, each terahertz light receiving unit 21 receives the near-infrared light 13b when the terahertz light 24 that has passed through the tablet 7 from the upper side to the lower side is incident. The electric signal 25 having a current corresponding to the intensity of the terahertz light 24 is transmitted to the receiving unit 26 of the control device 27 for a certain time.

  The optical delay unit 22 includes a plurality of fixed mirrors 28, a single movable mirror 29, and a mirror driving unit 30, and a timing signal (hereinafter referred to as a trigger) based on a command from the drive control circuit 31 of the control device 27. When the trigger signal generating device 31a emits a trigger signal at a timing when the tablet 7 reaches a position at an arbitrary distance set in advance from the center position of the terahertz light beam, the input signal is also in a standby state. The position is moved sequentially, and as shown in FIG. 3, the peak point of the near-infrared light 13b is sequentially increased from the head position of the waveform of the terahertz light 24, that is, the delay time Δt from the initial delay time, By sampling and adding the electrical signal 25 at that time, the time corresponding to one waveform within the delay time fluctuation range from the initial delay time of the terahertz light 24 Waveform 32 is obtained. In addition, the time waveform 32 changes for every kind of chemical | medical agent. For this reason, the delay time initial value and the delay time fluctuation range of the optical delay unit 22 are preset in the delay time setting unit 31b for each type of medicine.

  The drive control circuit 31 in the control device 27 refers to the setting contents of the delay time setting unit 31b based on the synchronization signal 34 of the output of the pulsed near infrared light 13 input from the femtosecond laser control unit 12. The delay time Δt that is sequentially increased in the optical delay unit 22 is sequentially set in the trigger signal generator 31a. The trigger signal generator 31a generates a trigger signal at the timing indicated by the set delay time Δt. Further, the drive control circuit 31 instructs the data memory 36 to start and end the sampling of the electric signal 25 when the delay time Δt is in the designated section.

  Each time the delay time Δt sequentially increases and advances by a specified interval, each electrical signal 25 is A / D converted into digital data by the A / D converter 34 and written into the data memory 36. In the data memory 36, when the sampling start is instructed from the drive control circuit 31, the sampling of each electric signal 25 is started, and when the sampling end is instructed, the sampling is ended and the tablet 7 shown in FIG. Since the time waveform 32 of the electric signal for one waveform of the transmitted terahertz light 24 is obtained, the obtained time waveform 32 is written into the waveform memory 37.

  The time waveform 32 converted into an electric signal for one waveform of the terahertz light 24 transmitted through the tablet 7 written in the waveform memory 37 of the control device 27 is sent to a data processing device 38 comprising a computer.

FIG. 5 is a cross-sectional view of the terahertz light beam, and shows a state when the beam spot center csp of the terahertz light beam is at a position 24 (center c7 of the tablet 7) in FIG. 4B. Since the beam spot diameter d _{sp is} larger than the diameter d _{7 of the} tablet 7, the background light and the tablet transmitted light reach the terahertz light receiving unit 21 and are detected.

  FIG. 6 is a schematic diagram showing how the tablet 7 moves in the terahertz light beam during conveyance. In the production line, since the tablet 7 moves while measuring the time waveforms 32a, 32 of the terahertz light 40, 24 detected by the terahertz light receiving unit 21, the tablet 7 is in the terahertz light beam as shown in FIG. Move in the transport direction.

  FIG. 4C shows a terahertz light emission in a state where the disc-like tablet 7 having the thickness H shown in FIG. 4A is inserted into the recess 4 of the sheet 3 as shown in FIG. It is a figure which shows the time waveform 32 converted into the electrical signal of the terahertz light 24 which permeate | transmitted this tablet 7 when the pulse-form terahertz light 39 is irradiated to the tablet 7 from the unit 16. FIG.

In this figure, when the beam spot diameter of the terahertz light 39 is larger than the tablet 7, the time waveform 32 a caused by the background light in the peripheral portion that does not pass through the tablet 7 indicated by the dotted line in FIG. Detected before waveform 32. The time difference T _H between the time waveform 32 due to the tablet transmitted light and the time waveform 32 a of the background light is proportional to the thickness H of the tablet 7.

  FIG. 7 is a diagram showing the time waveform 32 of the terahertz light 24 transmitted through the tablet 7 out of the two time waveforms 32 and 32a shown in FIG. The horizontal axis is the time t indicated by the number of samples in the sampling in the A / D converter 35 with the peak point of the time waveform 32a caused by the background light as the reference position (t = 0) at the time t, and the vertical axis is , Electric field strength E.

  FIG. 8 is a functional block diagram showing a different drug detection function performed by the data processing device 38 formed of a computer.

Each time waveform 32 input from the waveform memory 37 of the control device 27 is subjected to moving average by the moving average unit 41 and then sent to the feature amount extracting unit 42. The value E _t obtained by averaging at time t of the time waveform 32 shown in FIG. 7, using the value E _t at time t before the averaging is as follows. However, L is the number of data used for one averaging.

E _t = (E _{tL / 2} + E _{tL / 2 + 1} +... + E _t +... + E _{t + L / 2} ) / (L + 1)
Thus, the noise component contained in this time waveform 32 is removed by carrying out the moving average of the measured time waveform 32. FIG. The time waveform 32 from which the noise component has been removed is sent to the feature quantity extraction unit 42.

  In this embodiment apparatus, in order to create the reference data used for the determination before performing the determination of the different drug for each tablet 7, the time waveform 32 of each reference tablet of the 11 types of tablets 7 has one type each. A total of 2200 pieces for 200 pieces are read by the control device 27 in FIG. 1, time-averaged, and sent to the feature amount extraction unit 42. The reference tablet 7 is a normal tablet without defects. The feature quantity extraction unit 42 extracts the feature quantity 44 of the time waveform 32 of each reference tablet of the inputted tablet 7 and writes it in the feature quantity data table 43 shown in FIG.

Specifically, a total of 14 feature quantities 44 (hereinafter also referred to as feature quantities x _ji ) shown below obtained from the individual time waveforms 32 in the time waveform 32 shown in FIG. 7, where j is the waveform number, i: Feature quantity number) is stored in the feature quantity definition memory 45. Note that the feature quantity x _ji (= x _j1 , x _j2 ,..., X _j13 , x _j14 ) of the feature quantity numbers i = 1 to 14 described below is an example, and can be extracted from the time waveform 32 and used as the type of drug. Accordingly, if the feature amount has a different value, an arbitrary feature amount x _ij may be used.

(A) Zero cross point The time t when the intensity E of the time waveform 32 becomes zero is defined as the zero cross point, and the value t ₀ is obtained. Since the data is discrete, the point where E = 0 is crossed when the section where the sign of the data Et changes is linearly approximated is defined as a zero cross point. In this algorithm, the zero cross point immediately before the minus peak is the first zero cross point (t ₀₁₎ , and the second zero cross point that appears next to the first zero cross point is the second zero cross point (t ₀₂ ), the third zero cross point that appears next to the second zero cross point is defined as the third zero cross point (t ₀₃ ), and each zero cross point t ₀₁ , t ₀₂ , t ₀₃ is defined as The feature amounts are x _j1 , x _j2 , and x _j3 .

(B) Peak point (waveform peak point)
In the time waveform, the tablet peripheral portion passing light corresponding to the portion 32a in FIG. 7 and the tablet transmitting light corresponding to the portion 32 in FIG. Here, the portion 32a means a waveform observed at the same time as the waveform observed when terahertz light is received in the absence of the tablet 7, and the portion 32a appears after the waveform peak of 32a. A waveform with a large peak.

First, determine the time t = t _r the time waveform of the portion of 32a becomes the maximum, to the time t _r and the base peak.

Next, a time point t _{pp at} which the slope of the time waveform 32 becomes zero between two adjacent zero cross points is defined as a peak point, and a peak point t _pp is obtained. In this algorithm, the first and second peak points t _pp1 and t _pp2 are _set as feature quantities x _j4 and x _j5 . Further, a feature amount obtained by converting each peak point into a reference peak relative value t _{pp1 -r} (= t _pp1 -t _r ), t _{pp2 -r} (= t _pp2 -t _r ) is calculated. Further, a feature amount obtained by converting the second and subsequent peak points into the first peak point relative value t _{pp2 -pp1} (= t _pp2 -t _pp1 ) is calculated.

The feature amount obtained by converting the first peak point into the reference peak relative value t _pp1-r is a feature amount related to the thickness H of the tablet 7. The feature amount obtained by converting the second and subsequent peak points into the first peak point relative value t _{pp2 -pp1} is a feature amount that is not easily affected by fluctuations in the tablet thickness H.

(C) Peak height (waveform peak height)
The value E _tpp of the time waveform 32 at each peak point described above is defined as the peak height, and the value E is obtained. The peak height is not the absolute value of the value E of the time waveform 32, but is obtained separately on the positive side and the negative side. In this example, the heights E _tpp1 and E _tpp2 of the first and second peaks are _assumed to be feature quantities x _j6 and x _j7 .

(D) Waveform width The width between two points t _b1 and t _b2 having the same value E _b (= θ · E _tpp ) in a region sandwiched between two adjacent zero cross points t ₀ is set to the waveform width t _w (= t _b2 −t _b1 ) and its value is obtained. In the present embodiment, the following values E _b1, ..., the waveform width t _w1 in E _b7, ..., wherein t _w7 amount x _j8, x _j9, ..., and x _j13, x _j14.

E _b1 (θ ₁ , E _tpp2 ) = 0.2E _tpp2
E _b2 (θ ₂ , E _tpp2 ) = 0.4E _tpp2
:
E _b6 (θ ₆ , E _tpp2 ) = 0.8E _tpp2
E _b7 (θ ₇ , E _tpp2 ) = 0.9E _tpp2
That is, in the present embodiment, θ times {θ ₁ , θ _{2 ,.} . . , Θ _s | 0.0 ≦ θ ₁ <θ ₂ <,. . . , < _{W s} , t _w2 ,..., T _w6 , t _w7 at the value E _b of <θ _s ≦ 1.0} are feature quantities x _j8 , x _j9 ,..., X _j13 , x _j14 . In this example, the value of the multiple θ is set to θ ₁ = 0.2, θ ₂ = 0.4, θ ₃ = 0.5, θ ₄ = 0.6, θ ₅ = 0.7, θ ₆ = 0.8 and θ ₇ = 0.9.

The feature amount extraction unit 42 extracts 14 feature amounts 44 in a total of 2200 time waveforms 32 in each of the eleventh reference tablets sent from the moving average unit 41, and classifies each feature amount data table 43. Write. In the feature amount data table 43, as shown in FIG. 9, for each of the 11 types (type number p), for each of the 200 time waveforms 32 (waveform numbers j = 1 to 200), the total of 14 described above. A writing area for each feature quantity x _ji of the feature quantity (feature quantity number i = 1 to 14) is formed.

In the feature quantity combination table 46, as shown in FIG. 10, a combination number g is assigned and stored to a combination in which the selection / non-selection of the 14 feature quantities 44 described above is represented by 0/1. There are 2 ¹⁴ −1 = 16383 combinations, and all combinations are listed. Therefore, the combination number g = 1 to 16383. Note that the feature quantity combination table 46 is not limited to the above-described all combinations of the 14 feature quantities. For example, the feature quantity combination table generation unit (not illustrated) may perform all of the m feature quantities from any m feature quantities. It is possible to generate combinations so as to enumerate combinations (2 ^m −1 combinations).

  The combination feature amount extraction unit 47 extracts a feature amount selected by a combination of feature amounts specified by each combination number g set in the feature amount combination table 46 in the feature amount data table 43, and calculates the Mahalanobis distance. The data is output to the reference data creation unit 48.

  As a result, it is assumed that k feature values 44 are sent to the Mahalanobis distance calculation reference data creation unit 48 for n (= 200) waveforms 32 (j = 1 to n). Then, the Mahalanobis distance calculation reference data creation unit 48 calculates Mahalanobis distance calculation reference data for the input feature value data according to the following procedure.

For each input feature quantity x _ji (j = 1 to n, i = 1 to k), the following standardization process is performed, and the obtained feature quantity X _ji after the standardization process is revised. The feature amount is X _ji . That is, if the average of the i-th n feature quantities x _{1i to} x _ni is m _i and the standard deviation is σ _i ,
Feature quantity X _ji = (x _ji -m _i ) / σ _i
It becomes.

Next, a correlation matrix R of the feature amount X _ji is obtained. The correlator coefficient r _iq between the i-th feature quantity and the q-th feature quantity is given by the following equation.

Correlator coefficient r _iq = 1 / k (X _i1 X _q1 + X _i2 X _q2 +... + X _in X _qn )
i = 1 to k, q = 1 to k
Thereby, the correlation matrix R is shown below.

Next, an inverse matrix A of the correlation matrix R is obtained. The inverse matrix A of the correlation matrix R is expressed by the following equation, where the element is a _ji .

It becomes.

The inverse matrix A, m _{i, and} σ _i (i = 1 to k) are the Mahalanobis distance calculation reference data for the combination g and are recorded in the Mahalanobis distance calculation reference data table 49.

The reference Mahalanobis distance calculation unit 50 takes in feature quantity data of the reference drug from the feature quantity data table 43, reads out the Mahalanobis distance calculation reference data of the corresponding feature quantity combination from the Mahalanobis distance calculation reference data table 49, and standardizes the features. The amount X _ji is calculated, and the Mahalanobis distance of one type of the reference tablet 7 for the combination of the feature amounts of the combination number g is obtained using the inverse matrix A. A general expression for the Mahalanobis distance Dj of the jth time waveform 32 is given below.

However, u, v = 1 to k, j = 1 to n
The Mahalanobis distance data calculated in this way is recorded in the Mahalanobis distance data table 51.

The optimum feature quantity selection unit 52 reads the Mahalanobis distance data from the Mahalanobis distance data table 51, obtains the Mahalanobis distances D _{1 to} D _n of the respective time waveforms 32 where j = 1 to n, and calculates the lower limit in this combination (combination number g). The S / N ratio η shown by the equation is calculated.

η = −10 log 1 / n (1 / D ₁ ² + 1 / D ₂ ² +... + 1 / D _n ² )
... (2)
Thus, the calculation process of the S / N ratio η of one type of the reference tablet 7 for the combination of the two feature amounts 44 with the combination number g is completed.

  In this way, the Mahalanobis distance Dj in the equation (1) and the S / N ratio η in the equation (2) are calculated for a total of 16383 combinations. The optimum feature quantity selection unit 52 searches for a set of feature quantities having the maximum value from the S / N ratio data for 16383 sets. The result is written in the optimum feature value combination table 53, and further, the average value, standard deviation, and inverse matrix calculation data of the feature values for this optimum combination are written in the Mahalanobis distance calculation optimum reference data table 54.

  The above processing is carried out for 11 types of tablets 7 as reference drugs. The setting is completed by the above processing.

  At the time of inspection, the type of the reference tablet is designated before the inspection. The inspection Mahalanobis distance calculation unit 55 calculates the Mahalanobis distance in the type of the reference tablet of the tablet 7 to be inspected. That is, when the time waveform 32 of the tablet 7 to be inspected is input, the time waveform 32 is subjected to a time averaging process by the moving average unit 41 and then sent to the feature amount extraction unit 42.

The feature quantity extraction unit 42 calculates 14 feature quantities 44 for the time waveform 32. The result is output to the combination extraction unit 47. The combination extraction unit 47 outputs only the feature quantity 44 selected in the optimum feature quantity combination table 53 among the 14 types of input feature quantities 44 to the inspection Mahalanobis distance calculation unit 55. The inspection Mahalanobis distance calculation unit 55 calculates the standardized feature quantity X _i of the input feature quantity using the average value and standard deviation of the reference tablets stored in the Mahalanobis distance calculation optimum reference data table 54.

Then, the inverse matrix (inverse matrix A) of the reference tablet stored in the optimum reference data table 54 for calculating the Mahalanobis distance is substituted for the Mahalanobis distance D of the one-time waveform 32, and j = The Mahalanobis distance D ₁ when 1 is set as the Mahalanobis distance D. The inspection Mahalanobis distance calculation unit 55 sends the calculated Mahalanobis distance D to the different type determination unit 56.

The different type determination unit 56 compares the different type determination threshold value stored in the different type determination threshold value memory 57 with the calculated Mahalanobis distance D. And
When Mahalanobis distance D <differential determination threshold value Mahalanobis distance D> = differential determination threshold value, it is determined that the type of the tablet 7 is not a specified type but a different type. The determination result is displayed and output on the display unit 8.

  In addition, this invention is not limited to embodiment mentioned above.

  For example, the optimum reference data for Mahalanobis distance calculation such as the average value, standard deviation, and inverse matrix (inverse matrix A) of plural types of reference tablets need not be stored in the optimum reference data table 54 for Mahalanobis distance calculation. . Immediately before starting the evaluation of a new type of tablet, calculation of the optimum reference data for Mahalanobis distance calculation for the reference tablet 7 of the corresponding type and registration in the optimum reference data table 54 for Mahalanobis distance calculation may be performed.

  In the present embodiment, detection of a foreign drug in a pharmaceutical factory has been described. However, in a food factory or the like, foreign substances (heterogeneous raw materials, poisonous substances, etc.) may be mixed in regular food, and such foreign substances are detected. It is also possible to do. That is, in this embodiment, a drug is used as an example of a product, and a different drug is used as an example of a foreign object. However, the product and the foreign object are not limited to this example, and the time of electrical characteristics of terahertz light transmitted through the product As long as the waveform is different between the product and the different substances, it can be extended or generalized to any product such as a food, or any other different substance such as a different raw material or poison. In addition, with expansion or generalization to any product, the PTP packaging container in the case of a drug can be expanded or generalized to any packaging container according to the product.

1 is a schematic configuration diagram of a heterogeneous drug detection device according to an embodiment of the present invention. The elements on larger scale which take out and show the principal part of the different chemical | medical agent detection apparatus of the embodiment. The figure which shows the preparation procedure of the time waveform in the terahertz light of the same different chemical | medical agent detection apparatus. The figure which shows the relationship between a time waveform and the shape of a tablet. The schematic diagram which shows a terahertz light beam cross section. The schematic diagram which shows a mode that a tablet moves in a terahertz light beam at the time of conveyance. The figure which shows each feature-value defined in the time waveform. The functional block diagram which shows the function of the data processor of the same kind chemical agent detection apparatus. The figure which shows the memory content of the feature-value data table formed in the same data processor. The figure which shows the memory content of the feature-value combination table formed in the same data processing apparatus. The schematic diagram which shows the conventional spectrum waveform and wave | undulation.

  DESCRIPTION OF SYMBOLS 1 ... PTP packing machine, 2 ... Sheet supply part, 3 ... Sheet for packaging, 4 ... Recessed part, 5 ... Conveyance path, 7 ... Tablet, 10 ... PFP sheet, 11 ... Femtosecond laser oscillator, 12 ... Femtosecond laser control part , 13, 13a, 13b ... near infrared light, 14 ... beam splitter, 15a, 15b ... optical amplifier / distributor, 16 ... terahertz light emitting unit, 17 ... photoconductive antenna, 19 ... terahertz light receiving unit, 21 ... terahertz Optical receiving unit, 22 ... optical delay unit, 26 ... receiving unit, 27 ... control device, 31 ... drive control circuit, 32 ... time waveform, 38 ... data processing device, 41 ... moving average unit, 42 ... feature quantity extraction unit, 43 ... feature value data table, 44 ... feature value, 45 ... feature value definition memory, 46 ... feature value combination table, 47 ... combination feature value extraction unit, 48 ... Mahalanobis distance calculation Reference data creation unit 49: Mahalanobis distance calculation reference data table 50 ... Reference Mahalanobis distance calculation unit 51 ... Mahalanobis distance data table 52 ... Optimal feature quantity selection unit 53 ... Optimal feature quantity combination table 54 ... Mahalanobis Optimal reference data table for distance calculation, 55... Mahalanobis distance calculation unit for inspection, 56. Different kind determination part, 57. Different kind determination threshold memory, 58.

Claims (3)

A terahertz light incident part that injects pulsed terahertz light from one side of the product in a transported state along the transport path;
A terahertz light detector that receives the terahertz light transmitted through this product and the periphery of the product, converts it into an electrical signal, and outputs the electrical signal;
A time waveform reading unit that reads a time waveform of each transmitted terahertz light converted into an electrical signal output from the terahertz light detection unit;
Feature amount extraction means for extracting a plurality of feature amounts based on a predetermined zero cross point, waveform peak point, waveform peak height, waveform width in each time waveform read by the time waveform reading unit;
A feature value data table for storing each feature value extracted by the feature value extraction means in a reference product of a type designated in advance;
A feature quantity combination table generating means for enumerating all combinations of a plurality of feature quantities from the plurality of feature quantities;
For each combination of feature amounts generated by the feature amount combination generation unit among the feature amounts stored in the feature amount data table, a Mahalanobis distance for each combination calculated with each feature amount of the combination is calculated. A reference Mahalanobis distance calculating means;
A combination selection unit that selects a combination of feature amounts that maximizes the S / N ratio and writes the optimum reference data for Mahalanobis distance calculation corresponding to the selected combination of feature amounts into an optimum reference data table for Mahalanobis calculation; ,
Mahalanobis distance calculation means for inspection that calculates the Mahalanobis distance in the reference space of the product when the product to be inspected is input using the optimum reference data for Mahalanobis calculation stored in the optimum reference data table for Mahalanobis calculation;
A heterogeneous product detection apparatus, comprising: a heterogeneous determination unit that performs a heterogeneous determination of the product based on the calculated Mahalanobis distance. The product is inserted into a recess formed in the width direction of the packaging container,
The terahertz light incident portion has a plurality of incident heads that are arranged in the width direction of the packaging container and simultaneously enter terahertz light,
The terahertz light detection unit is a detection head that is arranged in the transport direction and simultaneously receives terahertz light;
Means for setting a delay time initial value and a delay time fluctuation range of the terahertz light detection unit for each type of the product;
Trigger signal generating means for generating a signal at a timing when the product reaches a position at a predetermined distance from the center position of the terahertz light beam;
2. The heterogeneous product detection apparatus according to claim 1, further comprising means for recording a time waveform from the timing at which the trigger signal is generated. A terahertz light incident step for injecting pulsed terahertz light from one surface of the product in a transported state along the transport path;
A terahertz light detecting step for receiving the terahertz light transmitted through this product and the peripheral portion of the product, converting it to an electrical signal, and outputting the electrical signal;
A time waveform reading step for reading a time waveform of each transmitted terahertz light converted into an electrical signal output from the terahertz light detection step;
A feature amount extraction step for extracting a plurality of feature amounts based on a predetermined zero cross point, waveform peak point, waveform peak height, waveform width in each time waveform read in this time waveform reading step;
A step of calculating a first peak point of a time waveform corresponding to light passing through a product periphery as a reference peak, and a step of converting a zero cross point and a waveform peak point into values based on the reference peak among the feature values Storing each feature quantity extracted in the feature quantity extraction step in a feature product data table in a reference product of a type designated in advance;
A feature quantity combination table generating step for enumerating all combinations of a plurality of feature quantities from the plurality of feature quantities;
For each combination of feature amounts generated in the feature amount combination generation step among the feature amounts stored in the feature amount data table, a Mahalanobis distance for each combination calculated only with each feature amount of the combination is calculated. A reference Mahalanobis distance calculating step,
A combination selection step of selecting a combination of feature amounts that maximizes the S / N ratio and writing the optimum reference data for Mahalanobis distance calculation corresponding to the selected combination of feature amounts into an optimum reference data table for Mahalanobis calculation; ,
A Mahalanobis distance calculation step for inspection that calculates a Mahalanobis distance in the reference space of the product to be inspected using the optimum reference data for Mahalanobis calculation stored in the optimum reference data table for Mahalanobis calculation when a product to be inspected is input; ,
A heterogeneous product detection method comprising: a heterogeneous determination step for performing a heterogeneous determination of the product based on the calculated Mahalanobis distance.

Images