JPH04286924A - Color-detecting type classifying apparatus - Google Patents

Abstract

PURPOSE:To easily set reference information which has been obtained in advance for determining a type in an apparatus for classifying objects to be measured such as leaf tobacco based on a color. CONSTITUTION:A color space value from a color sensor 1 is input from a measurement controller 3 to a non-linear arithmetic part 4, where a determination value is output. A coefficient of the nonlinear arithmetic part 4 is set based on the color space value and an evaluated value as for a plurality of objects to be measured which have been classified into a plurality of types which are input from an input means 6. The coefficient is set so that an error between the evaluated value and the determination value of the non-linear arithmetic part 4 is minimum.

Description

[Detailed description of the invention]

0001

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a raw material sorting device suitable for use in grading the quality of harvested and dried leaf tobacco and various agricultural products.

0002

[Prior Art] Conventionally, the quality of leaf tobacco has been evaluated based on quality factors such as leaf position, color, ripeness, mesophyll, and texture, and has been classified into quality appraisal categories as shown in Table 1 below. be identified. Then, prices are set according to each appraisal category (categories marked with "○"). (Hereafter, margin.)

[Table 1]

[0003] Among the purchasing categories in Table 1, color loss type P
This type of leaf tobacco is called quick-drying leaf tobacco, which is dehydrated quickly after yellowing during the post-harvest drying process, and the natural browning of leaf tobacco is not sufficient and the yellowish tinge remains. Also,
Among the poorly ripened mesophylls (autumn leaves, middle leaves, and lower leaves),
The white-blurred leaves, which have a dull, whitish hue with no luster in appearance, occur together with rapidly drying leaves, so the whitish-blurred leaves are also classified as type P.

[0004] These P type leaf tobaccos have problems such as not only giving a bad taste to the tobacco but also not having sufficient flavoring effect in the tobacco manufacturing process. Particular attention is paid to type identification.

[0005] However, since the above-mentioned P-type leaf tobacco is similar in hue to the ordinary A-type leaf tobacco, there is a problem in that it is difficult to grade even an expert using the conventional visual grading method.

For this reason, the color of a specimen such as leaf tobacco is detected, this color is converted into coordinate information in the color space of a predetermined color system, and the leaf tobacco, etc. is determined based on this coordinate information and pre-stored judgment conditions. A color detection type leaf tobacco sorting device that can automatically distinguish the type of leaf tobacco has been proposed (Japanese Patent Application Laid-Open No.
3-30657).

[0007]

[Problems to be Solved by the Invention] In the above-mentioned color detection type leaf tobacco type device, for example, a distribution in a color space corresponding to the colors of a plurality of leaf tobaccos classified in advance into a plurality of types is stored, and the characteristics within this distribution region are stored. The type of leaf tobacco is determined by comparing information indicating the boundary with coordinate values in a color space detected by a color sensor or the like.

[0008] For this reason, it is necessary to calculate in advance the shape of the three-dimensional distribution area of the color space for a plurality of leaf tobaccos that serve as the judgment criteria, the position information of the characteristic boundaries of this distribution area, etc. in the form of an equation, etc. . Note that setting such calculation formulas requires a lot of effort.

[0009] In addition, even for leaf tobacco of the same variety, the color distribution may vary depending on the production area and harvest year.
Since it is necessary to set the determination criteria again each time, there is a problem that the above-mentioned conventional method, which requires time and effort to set, is not easy to use.

0010

[Means for Solving the Problems] The color detection type classification device of the present invention, which has been made to solve the above problems, detects the color of an object to be measured and uses a brightness index of 1 and a perceptual color index of 2. A color measuring means that outputs color space values of a expressed color system, and color space values of a plurality of objects classified in advance into a plurality of types, and the classified types of the objects. an input means for inputting a judgment value indicating the plurality of color space values, a nonlinear calculation means for multiplying and adding the plurality of color space values by a coefficient, and calculating a nonlinear function value of these added values; coefficient setting means for setting the coefficients of the non-calculating means so that the error between the judgment value and the calculation result of the non-linear calculating means for the color space value is minimized with respect to the color space value of the object to be measured and the judgment value; The type of object to be measured is determined based on the calculation result of the nonlinear calculation means for the color space value obtained by the color measurement means.

[0011]

[Operation] In the color detection type classification device of the present invention, the nonlinear calculation means multiplies a plurality of color space values by a coefficient and adds them, calculates a nonlinear function value of these added values, and calculates a nonlinear function value based on this calculation value. The type of object to be measured is determined. Further, the coefficients of the nonlinear calculation means are set by the coefficient setting means so that the error between the calculation results for the color space values of the plurality of objects to be measured and the determination value of the object to be measured is minimized. Therefore, by inputting color space values and judgment values using the input means for a plurality of objects to be measured that have been classified into a plurality of types in advance,
Coefficients can be easily set so that the error in the judgment value is reduced.

0012

Embodiment FIG. 1 is a block diagram of a color detection type classification device according to an embodiment of the present invention. The color sensor 1 irradiates an object to be measured such as leaf tobacco with illumination light, receives the reflected light, and calculates the lightness index L* of the color system L* a* b* which indicates the color of the object to be measured.
and the perceived chromaticity indexes a*, b*. Each color space value is output as digital data, and the color space value from the color sensor 1 is input to the measurement control unit 3 via the interface 2. Note that as this color sensor, for example, "Color Difference Meter CR-110" manufactured by Minolta can be used.

The measurement control section 3 is composed of a microcomputer or the like, and FIG. 1 is a functional block diagram showing functions obtained based on a control program stored in a memory (not shown).

In the measurement control unit 3, the color space value reading means 31 takes in the color space values L, a, b from the color sensor 1 via the interface 2, and the storage means 32 stores the color space values L, a, b. is stored in a hard disk (not shown) or the like. Further, the control means 33 inputs the color space values L, a, b taken in by the color space value reading means 31 to the nonlinear calculation section 4.
Output to.

The nonlinear calculation unit 4 calculates a nonlinear function based on the coefficient Wjkm set by the measurement control unit 3, and calculates a judgment value D for the input color space values L, a, and b from the measurement control unit 3. Output.

The control means 33 takes in the judgment value D obtained by the nonlinear arithmetic unit 4, and compares this judgment value D with a preset judgment standard Ds to determine whether the tobacco is a regular leaf tobacco or a P-type leaf tobacco. The type of the object to be measured is determined, and the determination result is displayed on the display means 5 such as a display via the output means 34.

At the time of initial setting of the apparatus of this embodiment, color space values Li for a plurality of objects to be measured are determined in advance.
, ai , bi and a judgment value (hereinafter referred to as an appraisal value) Ti determined by an appraiser or the like, the coefficient Wjkm of the nonlinear calculation unit 4 is set, and the nonlinear calculation unit 4 is thereby inputted. A nonlinear function means is configured to output a judgment value D corresponding to a color space value.

[0018] Also, the appraisal value Tr obtained by actual appraisal of the measured object and the color space value L at the time of measurement.
The coefficient Wjkm can be further updated by r , ar , and br to improve the determination accuracy. In addition,
The color space values Li, ai, bi and the appraisal value Ti at the time of initial setting, and the new appraisal value Tr at the time of updating the coefficients are inputted from the input means 6 such as a keyboard and stored in the storage means 22.

FIG. 2 is a block diagram of the nonlinear arithmetic unit 4. The nonlinear arithmetic unit 4 includes a plurality of multipliers G whose multiplication coefficients are rewritable and which multiplies the input value by a coefficient and outputs the multiplier, and a total sum of the plurality of inputs. A plurality of adders A that take and output , and the following equation 1
and a plurality of sigmoid function units S having the input/output relationship shown in FIG.

[Math 1]

Note that in FIG. 2, a part of the multiplier G is omitted for the sake of simplicity, and the same multiplier G as shown by the solid line is arranged in the part indicated by the broken line in the figure. The first intermediate calculation unit 41 and the second intermediate calculation unit 42 each include three pairs of adders A and sigmoid function units S, and color space values L, a, and b are input to input terminals IL and Ia, respectively. ,Ib
is input.

Color space values L, a, and b input from input terminals IL, Ia, and Ib are input to each adder A1 of the first intermediate calculation section 41 via a multiplier G0, respectively, and
The outputs of each adder A1 are input to a sigmoid function unit S1, and the outputs from each sigmoid function unit S1 of this first intermediate calculation unit 41 are respectively inputted to each addition unit of the second intermediate calculation unit 42 via a multiplier G1. The signal is input to device A2.

Further, in the second intermediate calculation section 42, the outputs of each adder A2 are inputted to a sigmoid function unit S2, and the outputs from each sigmoid function unit S2 of this second intermediate calculation unit 41 are inputted via a multiplier G2. Adder A3
is input. The output of the adder A3 is input to a sigmoid function unit S3, and the output of this sigmoid function unit S3 is input as the determination value D to the measurement control unit 2.

Note that the coefficients of each multiplier G0, G1, G2 are expressed as "wjkm", and the input terminals IL, Ia
, Ib and the first intermediate calculation unit 41, the coefficients of the multiplier G0 are w110 , w120 , w130 , w210
, w220 , w230 , w310 , w320
, w330, the first intermediate calculation unit 41 and the second intermediate calculation unit 4
2 and the coefficients of multiplier G1 between w111 and w121
, w131 , w211 , w221 , w231
, w311 , w321 , w331 , the coefficients of the multiplier G2 between the second intermediate calculation unit 42 and the adder A3 are w
112, w212, w312.

Here, the sigmoid function unit step S3
has output characteristics as shown in Equation 1 and FIG.
≦1.

[0025] Now, let us explain the case of classifying leaf tobacco into color-loss type P type or normal type A type.
This judgment value D is a value that indicates which type the leaf tobacco is closer to when classifying the object into two types, such as whether the leaf tobacco is closer to the color loss type P type or the normal type A type. Now, the actual appraisal value T by the appraiser is expressed as 0.0 and 1.0, and the appraisal value T for discolored P type is 0.0, and the appraisal value T for normal type A is 0.0.
The type appraisal value T is 1.0. Then, each coefficient Wjkm of the nonlinear calculation unit 4 is set based on the appraisal value T and color space values L, a, and b for a plurality of leaf tobaccos.

It should be noted that the judgment value D output from the linear calculation section 4 at the time of measurement may have a judgment residual when setting the coefficient Wjkm, and may be a leaf tobacco with a color space value different from the data used for setting the coefficient. is sometimes used, the determination value from the nonlinear calculation unit 4 is not separated into two values of "0" and "1" but is a continuous value of 0.0 to 1.0. Therefore, a reference value Ds is set in advance for the judgment value D, and this reference value Ds
For example, regular leaf tobacco and P-type leaf tobacco are determined using the threshold value.

Here, the color space value Li of N leaf tobaccos is
, ai, bi [i=1 to N] and the appraisal value Ti of each leaf tobacco is determined, and the method of setting the coefficient Wjkm will be explained.

First, a random number is set for each coefficient Wjkm to determine the judgment value Di of the nonlinear calculation unit 4 for each color space value Li, ai, bi, and the error Err is determined by the following equation 2.

[Equation 2] At this time, the optimal value of the coefficient Wjkm is the value that minimizes the error Err, but it is generally difficult to find the optimal solution for a nonlinear function, so a numerical analysis method called the least square error method is used. apply.

The least square error method is a method that involves repetitive calculations, and for example, as shown in the following equation 3, the coefficient Wjkm
The direction of inclination of the error Err with respect to the coefficient Wjkm when the error Err is a certain value is calculated, and the value of the coefficient Wjkm is changed so that the error Err becomes a smaller value.

[Equation 3] However, the left side (Wjkm') is the updated coefficient, and the second term (μ) on the right side is a small value. By repeatedly calculating this equation 2, Err becomes smaller, and Err
When r takes the minimum value, the partial differential term of the second term on the right side is “0”
Then, the coefficient Wjkm is no longer updated, and the final value is taken as the value of the coefficient Wjkm.

In the actual calculation by the measurement control section 3, the error E when the value of the coefficient Wjkm is slightly changed is
A partial differential term is obtained from the difference in rr, and when this partial differential term reaches a preset small value, the iterative calculation is terminated.

FIG. 4 is a flowchart showing the control of the measurement control unit 3. When the power is turned on, a mode for setting coefficients (setting mode) and a mode for determining leaf tobacco (measurement mode) are selected. Display (step S1)
), the selected mode is determined based on the operating state of a predetermined key on the keyboard 6 (step S2).

When the setting mode is selected, the color space values Li, ai, bi and the appraisal value T are input from the input means.
i and are sequentially stored in the storage means 22 (step S3, step S4), and the calculations are performed as described above to obtain each coefficient Wjk.
m (step S5), and monitors the operation of a predetermined end key (step S6). If the end is not completed, the process returns to step S1. If the end is not completed, the coefficient Wjkm and the measured color space values Li, ai, The data such as bi is stored in a hard disk or the like and the process ends (step S7).

When the measurement mode is selected in step S2, it is detected that the measurement value has been determined by the color sensor 1 based on the interrupt signal from the color sensor 1, and the color space value Lr,
ar , br are taken in via the interface 3 (step S8), and the color space values Lr , ar , br
is output to the nonlinear calculation unit 4 and the judgment value D is taken in (step S9).

Next, the preset reference value Ds and the judgment value D are compared (step S10), and when D≧Ds, it is judged as a regular type leaf tobacco, and when D≦Ds, it is judged as a P type leaf tobacco. Determination is made (step S11). Then, this determination result is displayed (step S12), and the process of step S6 is performed.

[0035] As described above, a judgment result is obtained, but if this judgment differs from the actual judgment made by the appraiser or sufficient measurement accuracy cannot be obtained, the appraiser's judgment of leaf tobacco that was incorrectly determined is Measurement accuracy can be improved by adding the data of the appraisal value T and the color space values L, a, and b of this leaf tobacco to the data used in the initial setting of the coefficients and setting the coefficients again. can.

The nonlinear calculation section 4 of the above embodiment has two stages of intermediate calculation sections, and each intermediate calculation section has three pairs each of adders and sigmoid function units. The number of stages of the intermediate arithmetic unit and the number of adders and sigmoid function units are not limited to these. Further, in the above embodiment, a sigmoid function is used as the nonlinear function, but a nonlinear function such as the following expression may be used as the nonlinear function.

[Math 4]

[0037]

[Effects of the Invention] As explained above, according to the color detection type classification device of the present invention, the color space value of the color system indicating the color of the object to be measured is multiplied by a coefficient. nonlinear calculation means for adding values and calculating a nonlinear function value of these added values; input means for inputting color space values and judgment values indicating types of a plurality of objects to be measured that have been classified in advance into a plurality of types; a coefficient setting means for setting coefficients of the non-linear calculation means so that errors in the calculation results of the non-linear calculation means for color space values of a plurality of objects to be measured inputted by the input means are minimized; Since the type of the object to be measured is determined based on the calculation result of the nonlinear calculation means for the obtained color space value, the error of the calculation value of the nonlinear calculation means from the judgment value is reduced, and the Coefficients can be easily set. As a result, when grading objects to be measured using color measurement means, it is possible to eliminate the trouble of configuring discrimination logic for each category for characteristics that vary depending on the type, grade, place of origin, etc. of the object to be measured. It is possible to automatically generate a rating logic using a uniform method regardless of each characteristic, thereby improving diagnostic accuracy, reducing the cost of determining a discriminant logic, and improving operability.

[Brief explanation of the drawing]

FIG. 1 shows a color detection type quality determination device according to an embodiment of the present invention.

FIG. 2 is a block diagram of a nonlinear calculation section in an embodiment of the present invention.

FIG. 3 is a flowchart in an embodiment of the present invention.

FIG. 4 is a diagram showing input/output characteristics of a sigmoid function according to an embodiment of the present invention.

[Explanation of symbols]

1 Color sensor 3 Measurement control section 4 Nonlinear operation section G Multiplier A Adder S Sigmoid function unit

Claims (1)

[Claims] 1. Color measuring means for detecting the color of an object to be measured and outputting a color space value of a color system expressed by a brightness index of 1 and a perceptual color index of 2; input means for inputting a color space value of the plurality of objects to be measured and a judgment value indicating the classified type of the object to be measured, and multiplying the plurality of color space values by a coefficient and adding them and a nonlinear calculation means for calculating a nonlinear function value of these added values, and a calculation by the nonlinear calculation means for the color space values of the plurality of objects to be measured and the determination value inputted by the input means. coefficient setting means for setting the coefficients of the non-calculating means so that the error between the result and the judgment value is minimized, based on the calculation result of the non-linear calculating means for the color space value obtained by the color measuring means. 1. A color detection type classification device, characterized in that the type of an object to be measured is determined by