JP2889518B2 - High-speed quality evaluation method for extruded granulated products - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a high-speed extruded granulated product for rapidly evaluating the quality of the above-mentioned noodle-shaped molded product which is extruded from a lattice-shaped hole and then cut into pellets and granulated. Regarding the evaluation method.

[0002]

2. Description of the Related Art In the field of visual inspection for judging the quality of a manufactured object, when an object to be inspected has a known shape, such as a rigid body, a so-called model is determined from its geometrical characteristics. Pattern classification is possible by matching or the like. However, when a non-rigid object whose shape is not always constant is to be inspected, it is difficult to use the shape as an inspection parameter, and it is necessary to use another method such as texture analysis.

In this type of inspection field, it is a necessary condition that processing can be performed at high speed in real time without affecting the inspection system, regardless of whether the object is rigid or non-rigid. In particular, in the noodle-shaped molded product extruded from the lattice holes, a non-rigid body having a substantially circular cross section is curved at an arbitrary curvature, and is in a state of contacting or overlapping with each other in an arbitrary direction.

[0004] In such a quality inspection of the target object, a skilled worker evaluates the quality visually, that is, without contact. That is, the degree of quality of such an object has a strong correlation with the roughness of its surface, and the above-described visual evaluation estimates the quality from this surface state. As described above, conventionally, the quality evaluation of a noodle-like molded product is performed by a skilled worker who visually observes the surface state, and there is no known case in which such operation is automated. On the other hand, in the field of image coding processing,
Although a wavelet transform capable of separating an image for each frequency band is known, in the field of the present invention, a wavelet transform is used to extract or enhance a surface state of a two-dimensional image. No scheme has been proposed. Also, a steerable filter is known as a device that can extract a high frequency component of an outline in an arbitrary direction with a small filter.
In this field, there has not been proposed a method of separating objects that are in contact with each other at high speed by using a steerable filter.

Further, in order to inspect a crack of an object, a method of measuring the crack itself by a hierarchical method has been proposed. This is to extract various cracks by binarizing them so that the width of the cracks and the like can be accurately measured. The above wavelet transform is described in S. Gee. Murray, "A Theory for Multiresolution Signal Decomposition: The Wavelet Representation" (SG Ma
llet "A Theory for Multiire
solution Signal Decomposi
Tion: The Wavelet Represen
station, IEEE Trans. Pttern
Anal. & Mach. Intel. , PAMI-
11,7, pp. 674-693, July 198
9) and regarding the steerable filter, see WW. tea. Freeman and e. H. Andersen, "The Design and Youth of Sterable Filter" (WT Freeman &
E. FIG. H. Andelson "The Design a
nd Use of Steerable Filter
r "(IEEE Trans. Pattern Ana)
l. & Mach. Intel. , PAMI-13,
9, pp. 891-906, Sep. 1991) and Japanese Patent Laid-Open Publication No. 3-138775
There is a disclosure of Japanese Patent Publication No.

[0006]

In the above prior art, no technology for evaluating the surface condition of the target object has been developed, and no technology for high-speed separation of contour data using a steerable filter has been developed. . Further, in the case of measuring a crack by layering image data, the width of the crack can be assumed with high accuracy, but the surface state of the target object cannot be evaluated.

According to the present invention, the degree of smoothness of the surface of a target object is quantitatively evaluated, and the degree of cracks or irregularities on the surface of the target object is determined for each object on the basis of an evaluation criterion of a worker learned in advance. In addition, classification is possible for each quality. In particular, some noodle-shaped molded products extruded from a lattice-like hole are clearly separated on the image, but they are in contact with each other, and are linear or have any curvature. In general, the size of the object varies, so that the object must be forcibly separated on the image.

For this reason, it is not possible to apply a model matching method for abutting this kind of object with a typical model. Although the quality of an object can be estimated from the surface state of the object, processing using the high-pass filter for extracting a crack or the like requires a long processing time and is not suitable for high-speed processing.

Further, when the variety to be evaluated changes,
Each time the filter coefficient must be changed, the coping ability is lacking. Furthermore, in order to save processing time and suppress unnecessary noise, it is necessary to exclude the area outside the area of the target object from being processed. However, conventional filter processing saves processing time and handles real space and frequency space simultaneously. It is also difficult.

[0010] The relationship between the quality of the target object and its surface state slightly varies depending on the type of the object, and the quality and the surface state do not have a linear relationship. There is no formula for this relationship, and at present it depends on the experience of the worker.
For this reason, it is not appropriate to apply the conventional linear pattern classification method to the quality of an object which is extruded and granulated from a grid-like hole.

That is, an object of the present invention is to solve the above-mentioned problems of the prior art, and to automatically evaluate the quality of a noodle-like molded product extruded from a lattice-like hole at a high speed by image processing and classify it. System.

[0012]

In order to achieve the above object, the present invention firstly takes an image of an object to be evaluated (a noodle-shaped molded product) with a video camera, and converts the image data into a hierarchical image by wavelet transform. Become Background separation and feature extraction are performed based on the frequency components of the hierarchized images.

Further, the contour of the object is extracted based on the hierarchized image of the upper layer, the objects that are in contact with each other are separated on the image, and the luminance distribution of the image whose features are extracted for each object is obtained. The quality is estimated based on this. That is, as shown in FIG. 1, a video camera 2 for imaging a noodle-shaped molded product 1 and converting the surface state into two-dimensional image data as shown in FIG.
A layering means 3 for creating a layered structure for each spatial frequency band having a different spatial frequency by a two-dimensional discrete wavelet transform of an image signal of the video camera 2; and the noodle-shaped molded article 1 using the layered structure. A background separating / characteristic extracting means 4 for extracting image data of the noodle-shaped molded article 1 by separating the background of the noodle-shaped molded article 1 and performing the inverse wavelet transform based on the upper layer of the hierarchical structure to form the noodle-shaped molded article 1 Contour extracting means 5 for extracting the contour of the article 1, and inverse wavelet transform for pixels in the contour extracted by the contour extracting means 5 up to the resolution of the lowermost layer of the hierarchical structure to obtain the noodle shaped article 1 An image resynthesizing means 6 for obtaining separated image data and an image reconstructed by the image resynthesizing means 6 are used to evaluate the surface condition of the noodle-shaped molded article 1. And a quality estimating unit 7 having a multilayer perceptron neural network, and evaluating the quality of the quality of noodle-like molded article is extruded from the grid-shaped hole from the surface state.

The object to be inspected in the above invention is
It is not limited to the noodle-shaped molded product extruded from a lattice-shaped hole or the like, but the extruded noodle-shaped molded product is chopped into pellets, and other objects whose quality can be evaluated from the surface state. The same can be applied to

[0015]

DESCRIPTION OF THE PREFERRED EMBODIMENTS A video camera 2 has a CCD image pickup device or an image pickup tube and a photoelectric conversion device, and has a noodle-like molded product 1 which is molded by extrusion through holes such as a lattice.
And converts the surface state into two-dimensional image data. The layering means 3 creates a hierarchical structure for each spatial frequency band having a different spatial frequency from the image signal of the video camera 2 by two-dimensional discrete wavelet transform.

The layering by the wavelet transform is executed as follows. Here, one-dimensional description will be given for simplicity of description. That is, the input signal from the video camera 2 is i (x), and the output having the low frequency component is O (x).
_{Assuming that L} (x) and the output having the high frequency component are O _H (x), O _L (x) and O _H (x) are

[0017]

(Equation 1) ... (1) However,

(Equation 2) Is the wavelet transform coefficient,

(Equation 3) Is the low frequency characteristic,

(Equation 4)

Has high frequency characteristics. FIG. 2 is a schematic diagram for explaining the configuration of the layering means by the wavelet transform. In this example, the number of layers is set to 3, and 1 is the original image, 2
a, 2b are horizontal downsamplers, 3a, 3b, 3c,
3d is a vertical down sampler, 4 is the first layer, 5a, 5b
Is a horizontal downsampler, 6a, 6b, 6c, 6d are vertical downsamplers, and 7 is a second layer.

In FIG. 1, assuming that the number of pixels of the original image 1 is N, the original image 1 is converted into a horizontal downsampler 2 by using low-frequency and high-frequency wavelet transform coefficients.
a, 2b, each sampled image is further similarly converted to a vertical downsampler 3a, 3b.
Sampled by b, 3c, 3d and the number of pixels is N / 2
To obtain a first layer composed of four images.

Then, the horizontal low-frequency image and the vertical low-frequency image of the image of the first layer are subjected to wavelet transformation and passed through horizontal downsamplers 5a and 5b, and similarly subjected to wavelet transformation to perform vertical downsampler. By sampling at 6a, 6b, 6c and 6d, a second layer 7 composed of four images having N / 2 ² pixels is obtained.

FIG. 3 is a schematic diagram for explaining the structure of an image re-combining means by inverse wavelet transform. In FIG. 3, each of the second layers 7 of the four images described in FIG. 2 is used as a vertical upsampler 8a, 8b. , 8c, and 8d, and inverse wavelet transform these with low-frequency and high-frequency wavelet transform coefficients, and add them by adders 9a and 9b.

The inversely transformed two sheets are further passed through horizontal upsamplers 10a and 10b, which are inversely wavelet-transformed with low-frequency and high-frequency wavelet transform coefficients, added by an adder 11, and recombined. A layer 12 is obtained. Then, the two images of the recombined first layer are sampled by the vertical upsamplers 13a, 13b, 13c, and 13d, and the two images subjected to the inverse wavelet transform are similarly added by the adders 14a and 14b.

The two images after the conversion are sampled by the horizontal upsamplers 15a and 15b, and the two images subjected to the inverse wavelet conversion in the same manner are added by the adder 16 to obtain a recomposed image 1
Get 7. FIG. 4 is an explanatory diagram of the frequency components of each layer of the hierarchical structure. The images obtained by the wavelet transform have independent frequency bands, and the number of layers with the original image as the lowest layer is In the case of 3, the frequency components of each image are as illustrated.

[0024] In the figure, HL ^j is the high frequency range the horizontal frequency component in the hierarchy j, the vertical frequency component low frequency band,
omega _x is the horizontal frequency band, omega _y denotes a vertical frequency band.
The background separation / feature extraction means 4 separates the background of the noodle-shaped molded article 1 using the hierarchical structure and extracts image data of the noodle-shaped molded article 1. The highest level of the hierarchical structure has the lowest resolution, and the lower layers have higher resolution. The low-frequency image in a certain hierarchical J (corresponding to O _L in the formula (1)), the high frequency component image having the lower layers than the hierarchy does not have.

Therefore, when performing image resynthesis, an image corresponding to a necessary frequency component of a necessary hierarchy J is validated,
By invalidating the others, information consisting only of the necessary frequency band is extracted. In layer J where cracks and irregularities on the surface of the target object are sufficiently attenuated in the low-frequency image,
The low-frequency image ^{_{I LL j (x, y)}} , and by the threshold ^{_{TH, I LL j (x,}} y) = 0, when I LL j (x, y) < and TH ··· (2), lower than that In the layer of, I _LL ^j-1 (x, y) = 0, when I _LL ^j (x, y) <TH, and I _LL ^j (x / 2, y / 2) <TH (3) ).

By performing re-synthesis while performing this conversion,
To separate the surface and background or shadow of the target object. Next, the hierarchy
j corresponding to the frequency band included in the image,
_HL ^j, I_LH ^j, I_HH ^j, I_LL ^jAnd

The bandwidth ω _HL ^j , ω _LH ^j , ω _{HH of} each image
^j and ω _LL ^j are the horizontal frequency band ω _x and the vertical frequency band ω
_{Using y} , ω _HL ^j = ((1/2) ^j π ≦ ω _x <(1/2) ^j-1 π, 0 ≦ ω _y <(1/2) ^j π) ω _LH ^j = (0 ≦ ω _x <(1/2) ^j π, (1/2) ^j π ≦ ω _y <(1/2) ^j-1 π) ω _HH ^j = ((1/2) ^j π ≦ ω _x <( 1/2) ^j-1 π, (1/2) ^j π ≦ ω _y <(1/2) ^j-1 π) ω _LL ^j = (0 ≦ ω _x <(1/2) ^j π, 0 ≦ ω _y <(1/2) ^j π) (4) Also, ω _LL ^j has the following relationship with the image one level _higher ^: ω _LL ^j−1 = ω _HL ^j + ω _LH ^j + ω _HH ^j + ω _LL ^j (5)

Therefore, the spatial region to be removed in the image having the frequency band to be removed is set to 0, and the image is recombined to emphasize features due to cracks and irregularities. The contour extracting means 5 extracts the contour of the noodle-shaped molded article 1 by performing inverse wavelet transform based on the upper layer of the hierarchical structure, and the image reconstructing means 6 extracts the contour by the contour extracting means 5. Inverted wavelet transform is performed on the pixels in the contour thus formed up to the resolution of the lowermost layer of the hierarchical structure to obtain reproduced image data in which each of the noodle-shaped products 1 is separated.

FIG. 5 is an explanatory diagram of an example of the configuration of the stabilizing filter.
°, 120 ° two-dimensional filters, 21a to 21c are multipliers, 22 is an adder, and 23 is a look-up table. FIG. 6 is a conceptual diagram of a window and inclination estimation according to the inclination of the edge, and FIG. 7 is a conceptual diagram of a window perpendicular to the edge.

[0030] Here, to emphasize stay trouble filter edge orientation theta, using the N derivative G _N of the Gaussian function G (x, y), [ G N (θ)] = [Σ (M) _{(r = 1)} ] k _r (θ) [G _N ⁽ θ _r ⁾ ] (6)

Here, [G _N ⁽ θ ⁾ ],
[Σ ^(M) _{(r = 1)} ] and [G _N ⁽ θr ⁾ ] are

(Equation 5) Means For example, when N = 2, the steerable filter has the configuration shown in FIG. First, the following processing is performed in the hierarchy J. [G _N ⁽ θ ⁾ ] and its Hilbert transform pair [H
_N ⁽ θ ⁾ ]], the direction θm (0 ° ≦ θm ≦ 360 °
Is determined by the following equation.

[0032] ^{[E (θm)] = (} (I LL J ※ [G (θm)]) 2 + (I LL J ※ [H (θ m)] 2) 1/2 ····· (7) Here, * indicates convolution integral, [E ⁽ θm ⁾ ],
[G ⁽ θm ⁾ ] and [H ⁽ θm ⁾ ] are respectively

(Equation 6) Means

In the fan-shaped region WH whose center axis is the direction θm shown in FIG. 6, the energy concentration S (θm)
Is determined by the following equation. S (θm) = ∫ _WH [E ⁽ θm ⁾ ] dxdy (8) Let θm at which S becomes the maximum be the estimated edge direction θe.

Regarding the direction θ⊥ perpendicular to θe, if the energy is locally maximum and is equal to or more than a certain threshold value TH in an area WV having the center axis θθ shown in FIG.

The spatial phase [φ ⁽ θe ⁾ ] of each pixel is represented by [φ ⁽ θe ⁾ ] = tan ^-1 (I * [H ⁽ θe ⁾ ] / I * [G ⁽ θe ⁾ ] (... ⁾ 9) where [φ ⁽ θe ⁾ ], [H ⁽ θe)
⁾ ] And [G ⁽ θe ⁾ ] are

(Equation 7) Means

When the calculation result is Δ / 2 rad. And 3χ / 2ra
d. If it is near, it is regarded as a step edge. Also, 0
rad. And @rad. If it is near, it is regarded as a pulse edge. At the edge point, in the case of a step-like edge, it is a contour. When the edge has a pulse shape, [E ⁽ θe
⁾ ] Is greater than or equal to a fixed threshold value TH2, S (θm) is sufficiently large, and the distance from the nearest edge existing in the direction θ⊥ is greater than or equal to L. Here, L is a threshold value of the distance set according to the shape of the target.

Here, [E ⁽ θe ⁾ ] is

(Equation 8)

Means For an area WH in an arbitrary direction θ centered on an arbitrary image, there are contour points on both sides, and if the average of the directions of the contours on each side is θ ± Δθ, the image is regarded as an edge. With respect to θ⊥ perpendicular to the edge direction θ, the edge at the center is defined as a contour.

Next, a layer j lower than the layer J (0 ≦ j <
In J), the above is executed at coordinates (x, y) where I _LL ^{j + 1} (x / 2, y / 2) is a contour. This is repeated up to the lower layer (j = 0). The quality estimating means 7
Quality of the noodle-shaped product extruded from the lattice holes using a multilayer perceptron type neural network for evaluating the surface state of the noodle-shaped granulated product 1 with respect to the image reconstructed by the image reconstructing means 6 Is evaluated from its surface condition.

That is, the quality estimating means 7 uses a perceptron-type neural network to estimate and classify the quality for each surface state of the noodle-like molded article as the target object. The input of the quality estimating means 7 is a luminance distribution for each object, and the output is a quality rank. First, the number (rank number) for classifying quality is set in advance. Then, the input image frame is presented to the operator who has conventionally performed the visual classification, and 1
Let the pixel average quality be classified.

Next, the average luminance distribution of the object region in one image is input, and the classification result is used as a teacher signal. This learning is performed uniformly from the worst quality to the good quality. Using the learning results obtained in this way, the quality of the noodle-like molded product extruded from the lattice holes is automatically evaluated from the surface state.

[0042]

Embodiments of the present invention will be described below in detail with reference to the drawings. In this embodiment, a high-speed quality evaluation system in which an organic rubber chemical is used as an example of a target object will be described. This organic rubber chemical is extruded from a large number of lattice holes to form a noodle-shaped molded product having a diameter of about 1 mm, which is magnified by a video camera.

FIGS. 8 and 9 are system configuration diagrams for explaining one embodiment of a high-speed quality evaluation method of an extruded granulated product according to the present invention, wherein both figures are divided for convenience of illustration. , Circle B and circle C are combined to form one drawing. 8 and 9
, 30 is a layering unit (wavelet layering unit) using a wavelet transform for layering image data of a target object obtained by imaging with a video camera, 31 is a background separation / feature extraction unit, and 33 is an image. Recombining / luminance distribution calculation unit, 3
Reference numeral 4 denotes a neural network quality estimating unit.

The input image is decomposed into multiple resolutions by a wavelet transform in a wavelet layering unit 30. Here, the input image is 512 × 512 pixels, and the highest layer is 64 × 64 pixels. The resolution is 25
Six gradations are assumed. The image signal enlarged and captured by the video camera is first hierarchized by the wavelet hierarchization unit 30, but in the present embodiment, there are three hierarchies except for the original image.

Each image I of each layer_HL ^j, I_LH ^j, I_HH ^jYou
And the top level I_LL ^jIs the background separation / feature extraction unit 31
(J = 1 to 3). Background separation / feature extraction unit 31
Identifies the object and the background, and does not need the area of the object
Frequency components are removed. The result is the mask bit MB
^j(MB^j _LL, MB ^j _HH, MB^j _LH, MB^j _HL, J = 1
To 3) are output to the image resynthesis / luminance distribution calculation unit.
You.

The contour extraction unit 32 extracts the contour of the object.
And the result is the mask bit MC^j(J = 1-3)
Is output to the image resynthesis / luminance distribution calculation unit 33.
In the image resynthesis / luminance distribution calculation unit 33, the mask bit MB
^jAnd mask bit MC ^jIs "1", the image
The corresponding frequency component of the pixel is removed
You.

From the result obtained by the image resynthesis, a frequency distribution with luminance as a variable is calculated. The calculation result is input to the neural network quality estimation unit 34,
The quality of the object is estimated from the relationship between the luminance distribution and the quality that have been learned in advance. Hereinafter, each unit in the above system configuration will be further described. FIG. 10 and FIG. 11 are configuration diagrams of the background separation / feature extraction unit.
Hierarchical resynthesizer, 31c, 31d, 31e are comparators, 31
f, 31g, 31h are AND circuits, 31i is a rejection band selector, and 31j to 31s are OR circuits.

In the figure, a hierarchical image is input and
Then, the one-level re-combining units 31a and 31b^j _LL(J ≧
1) is required. FIG. 12 shows a configuration example of the one-layer re-synthesis unit.
It is a circuit diagram for explaining, 31d₁, 31d_ThreeIs horizontal h
Filter, 31d_Two, 31d_FourIs the horizontal g filter, 31
d_Five, 31d₆, 31d₁₁Is an adder, 31d₇, 31d
₈Is a horizontal up sampler, 31d ₉Is the vertical h-fill
TA, 31d_TenIs a vertical g filter, 31d₁₂Is vertical up
It is a sampler.

In I ^J _LL , a pixel whose luminance is lower than the threshold value TH is recognized as a background, and a background flag B ^J _LL (J = 1 to 3) is set in a lower-resolution image if the background is present. The background flag B ^J _LL and the band rejection selector 31i
The flag MB ^J _LL (J = 1 to
3) is set. In this case, the whole area is masked in the image corresponding to the selected band.

FIG. 13 is a circuit diagram for explaining an example of the configuration of the contour extraction unit, where 32a, 32c and 32f are hierarchical selectors,
32b is a one-level resynthesis unit, 32d is an AND circuit, 32e
Is a one-layer outline extraction unit, and 32g is a layer management unit that outputs a layer signal for selecting a layer. In this contour extraction unit, first, the hierarchy selectors 32a, 32c, 32f
Is the layer selection signal Layer from the layer management unit 32g and is the third
The layer is specified.

Thus, I ³ _ll , I ³ _LH , I ³ _Hl ,
I ³ _HH is input to the one-layer re-combining unit 32b. I ^j-1 _LL, that is, I ² _LL, is output from the one-layer recombining unit 32b. On the other hand, since the input of the layer selector 32c is always 1 in the current layer according to the layer selection signal Layer, I ³
_ll is output to the one-layer outline extraction unit 32e as it is. 1
From the hierarchical outline extracting unit 32e are extracted contour is outputted as the flag MC ³ for each pixel.

When the processing of this hierarchy is completed, the hierarchy management unit 32g sends the layer selection signal Layer to the hierarchy selectors 32a, 32
2c and 32f to select the second layer. The I ² _LH , I ² _Hl , I ² _HH and I ² _ll synthesized in the previous layer are input to the input unit of the one-layer re-combining unit 32b. Further, only the pixels corresponding to the contour extracted in the previous layer are input to the one-layer contour extracting unit 32e through the AND circuit 32d.

Then, the processing up to the first layer is repeated in the same manner as described above, and the flag MC ² and the flag MC ¹ are output. FIGS. 14, 15 and 16 are circuit diagrams for explaining an example of the configuration of the one-layer contour extraction unit, which is divided into three for convenience of illustration. In each figure, 321a is the real part of the steerable filter, 321b is the imaginary part, 322a ₁ to
322a _m is squared and the square of the square root operation circuit of the second input of the first input, 322b ₁ ~322b _m arithmetic circuit of the first input and the second input of tan ^-1, 323a, 323b faces θ
A selector 324 is a circuit for calculating an average value of θ in the window WH, 325 is a local maximum energy detection circuit in the window WV, 326 is a phase determination circuit, and 327a _{1 to} 327a.
₃ is a threshold value TH1, TH2, TH3 determination circuit,
28a ₁ and 328a ₂ are AND circuits, 328a ₃ is an OR circuit, 329 is an edge search circuit within the distance L, 330 is a contour detection circuit within the window WH, 331 is a center determination circuit within the window WV, and 332 is a logic. The product circuit 333 is a logical sum circuit.

FIG. 17 is a circuit diagram illustrating an example of the configuration of the real part of the steerable filter of the one-layer contour extraction circuit in FIG. In the figure, 3210a, 3210b,
3210c has directions θ of 0 °, 60 °, and 120 °, respectively.
3211a, 3211b... 32
An operation unit 11m has the same operation as described with reference to FIGS. 5, 6, and 7.

FIG. 18 shows the structure of the image resynthesis / luminance distribution calculation unit.
3310, 3310a, 3311a to 3311c,
3312a to 3312c and 3313a to 3313c
Logical product circuit, 3314 is a layer selector, 3315 is one layer recombining
Component, 3316 is a hierarchy management unit, 3317 is luminance distribution calculation
Department. The image resynthesis / luminance distribution calculation unit is hierarchical
Image and flag MB^J, MC ^JIs entered and the hierarchical tube
Flag MB according to the layer selection signal from the^J, M
C ^JThe set pixel has a luminance of 0.

First, the layer selector 3314 selects the third layer according to the layer selection signal Layer from the layer management unit 3316, and the image of the third layer is input to the one-layer re-synthesis unit 3315. Next, the second layer is selected by the layer selector 3314,
The image of the second layer is input to the one-layer re-synthesis unit 3315.
Hereinafter, the process is repeated up to the first layer, and the one-layer recombining unit 3315 is performed.
Output a feature extraction image.

This feature extraction image is output to the luminance distribution calculation unit 33.
17 and the appearance frequency of each of the 256 gray levels is calculated. The calculated result is given to the quality estimating means as a luminance distribution using the gradation as a variable, and is evaluated by the neuro network. Since the configuration and operation of the neural network itself are known, detailed description is omitted here. In the present embodiment, learning using the visual evaluation result of a skilled worker who has been engaged in the quality evaluation of the noodle-shaped molded product as teacher data is performed.

In operation of the present system, the luminance distribution output from the image resynthesis / luminance distribution calculation unit is input and the quality is evaluated using the neural network constructed as a result of the learning. The operation of the evaluation is as described in the embodiment of the present invention. FIG. 19 is a schematic diagram for explaining the overall configuration of the high-speed quality evaluation method of an extruded granulated product according to the present invention, wherein 100 is an extruded granulator, 101 is a raw material supply pipe, 102 is an additive A supply pipe, and 103 is Additive B
A supply pipe, 104 is a molded article conveying device, 105 is a video camera, 106 is a high-speed quality evaluation device, 107 is a fuzzy operation device, 108 is a control panel, and 109 and 110 are electromagnetic valves.

The extruding granulator 100 has a granulating mechanism for kneading the raw materials and additives and forming into granules, and has a large number of holes arranged in a grid, and extrudes the kneaded material from the grid holes. To produce a noodle-shaped molded product. The noodle-shaped molded product extruded from the extrusion granulator 100 is conveyed to the next step by the granulated product conveying device 104. The molded article transporting device 104 has a function of chopping the noodle-shaped molded article into an appropriate length to form a pellet, or a function of aligning the noodle-shaped molded article or the pellet-shaped molded article under a video camera. It can also be done.

The video camera 105 is used for the molded article conveying device 10.
In step 4, the granulated product conveyed is magnified and imaged, and its image signal (image data) is provided to the high-speed quality evaluation device 106. The high-speed quality evaluation device 106 non-quantitatively estimates and executes a quality evaluation equivalent to the evaluation of a skilled worker in a non-contact manner based on the configuration described above. The quality evaluation value estimated by the high-speed quality evaluation device 106 is given to the fuzzy operation device 107, and the flow rate of the additive is calculated by a so-called ambiguous process.
8 is output.

The control panel 108 controls the amount of the additive by controlling the opening and closing of the electromagnetic valves 109 and 110 based on the input from the fuzzy arithmetic unit 107, and also controls the kneading conditions of the extrusion granulator 100. The control panel 108 also controls the transfer of the granulated product transfer device 104. FIG. 20 is a schematic view of a photograph for explaining an example of the surface state of a good-quality noodle-like molded product and an explanatory view of a luminance distribution obtained by image processing. FIG. 20A is a schematic view of a photograph of a noodle-like molded product. , (B) is a luminance distribution diagram of one of them.

As shown in FIG. 6A, no cracks or irregularities were visually recognized on the surface of the noodle-shaped molded product evaluated as good, and the luminance obtained by the image processing is shown in FIG. Distributed as shown. FIGS. 21A and 21B are a schematic view of a photograph illustrating an example of the surface state of a defective noodle-shaped molded article and an explanatory view of a luminance distribution obtained by image processing. FIG. FIG. 3B is a diagram showing one luminance distribution.

As shown in FIG. 7A, many cracks and irregularities are visually recognized on the surface of the noodle-shaped molded product evaluated as defective, and the luminance obtained by the image processing is shown in FIG.
Are distributed as shown in FIG. Based on such a difference in luminance distribution, the high-speed quality evaluation device 106 including the quality estimation means having the multilayer perceptron-type neural network estimates the quality of the noodle-shaped molded product transported through the granulated product transport device 104. .

As a result, the quality evaluation is performed at a high speed without the need for an operator who performs the quality evaluation, and the evaluation result is fed back to the amount of the additive supplied to the extrusion granulator or the granulation conditions. Thereby, a high-quality granulated product can be manufactured. In FIG. 19, the surface state of the noodle-shaped molded product is imaged while being transported by the molded product transporting device 104 or in a state where the transport is stopped.
The field of view of the video camera may be pointed at the molded product outlet of (1) and an image of the noodle-shaped molded product immediately after molding may be taken.

Further, the present invention is not limited to the noodle-shaped molded article described in the above embodiment, but may be used to determine whether the quality is good or not based on the noodle-shaped object cut into a pellet shape by cutting it further shorter, or the surface condition. It goes without saying that it can be applied to the evaluation of all possible objects.

[0066]

As described above, according to the present invention,
An operator who quantitatively evaluates the degree of smoothness of the surface of a target object, particularly a noodle-like object extruded and granulated from a grid-like hole, and previously learned the degree of cracks and irregularities on the surface of the target object for each object. And a high-speed quality evaluation method for extruded granulated products that can be evaluated based on the evaluation criteria described above, automatically evaluated for each quality at a high speed, and classified and evaluated.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a basic configuration of a high-speed quality evaluation method of an extruded granulated product according to the present invention.

FIG. 2 is a schematic diagram illustrating a configuration of a layering unit using wavelet transformation.

FIG. 3 is a schematic diagram illustrating a configuration of an image resynthesis unit using inverse wavelet transform.

FIG. 4 is an explanatory diagram of frequency components of each layer of the hierarchical structure.

FIG. 5 is an explanatory diagram of a configuration example of a steerable filter.

FIG. 6 is a conceptual diagram of a window and inclination estimation according to an edge inclination.

FIG. 7 is a conceptual diagram of a window perpendicular to an edge.

FIG. 8 is a partial view of a system configuration for explaining an embodiment of a high-speed quality evaluation method of an extruded granulated product according to the present invention.

FIG. 9 is a partial view connected to FIG. 8 of a system configuration for explaining one embodiment of a high-speed quality evaluation method of an extruded granulated product according to the present invention.

FIG. 10 is a partial configuration diagram of a background separation / feature extraction unit.

11 is a partial configuration diagram of the background separation / feature extraction unit connected to FIG. 10;

FIG. 12 is a circuit diagram illustrating a configuration example of a one-layer resynthesis unit.

FIG. 13 is a circuit diagram illustrating a configuration example of a contour extraction unit.

FIG. 14 is a partial circuit diagram illustrating a configuration example of a one-layer contour extraction unit.

FIG. 14 illustrates a configuration example of a one-layer contour extraction unit.
FIG.

FIG. 16 illustrates a configuration example of a one-layer contour extraction unit.
FIG.

17 is a circuit diagram illustrating a configuration example of a real part of a steerable filter of the one-layer contour extraction circuit in FIG. 14;

FIG. 18 is a configuration diagram of an image resynthesis / luminance distribution calculation unit.

FIG. 19 is a schematic diagram illustrating an overall configuration of an embodiment of a high-speed quality evaluation method of an extruded granulated product according to the present invention.

FIG. 20 is a simulated view of a photograph illustrating an example of the surface state of a good-quality noodle-like granulated product, and an explanatory diagram of a luminance distribution obtained by image processing.

FIG. 21 is a simulated view of a photograph illustrating an example of the surface state of a defective noodle-like granulated product and an explanatory diagram of a luminance distribution obtained by image processing.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Noodle-like granulated product 2 Video camera 3 Hierarchical means 4 Background separation / feature extraction means 5 Contour extraction means 6 Image resynthesis means 7 Quality estimation means.

──────────────────────────────────────────────────続 き Continuation of the front page (72) Inventor Toshiaki Sakai 316, Ohnokita, Hirao-cho, Kumage-gun, Yamaguchi Prefecture (56) References JP-A-9-108564 (JP, A) JP-A-6-274615 (JP) , A) (58) Fields surveyed (Int. Cl. ⁶ , DB name) G06T 1/00-7/00

Claims (1)

(57) [Claims] 1. A high-speed quality evaluation method of an extruded granulated product for evaluating the quality of a noodle-shaped molded product which is extruded into a noodle shape and then cut into pellets and granulated from the surface state, A video camera that images a noodle-shaped molded product and converts the surface state into two-dimensional image data; and creates a hierarchical structure for each spatial frequency band having a different spatial frequency by two-dimensional discrete wavelet transform of the imaging signal of the video camera. A background separating / feature extracting means for separating a background of the noodle-shaped molded product by using the hierarchical structure and extracting image data of the noodle-shaped molded product, and an upper layer of the hierarchical structure Contour extracting means for extracting the contour of the noodle-like granulated product 1 by performing inverse wavelet transformation based on Image resynthesizing means for performing inverse wavelet transform up to the resolution of the lowermost layer of the hierarchical structure for the pixels in the section to obtain reconstructed image data in which each of the noodle-shaped molded articles is separated, and reconstructing by the image resynthesizing means Quality estimation means having a multilayer perceptron-type neural network for evaluating the surface state of the noodle-shaped molded product with respect to the obtained image, and determining whether the quality of the noodle-shaped molded product extruded from the lattice holes is good or bad from the surface state. A high-speed quality evaluation method for extruded granulated products characterized by evaluation.