JPS62239278A - Image recognizing device for agglomerate - Google Patents

Abstract

PURPOSE:To classify and recognize a floc and a background and to detect the floc with good accuracy by executing the binarization after the luminance of the floc is emphasized by a space filtering. CONSTITUTION:After the luminance slope of the floc part of a photographed floc image is emphasized by a luminance slope emphasizing means 60, it is compared with the threshold and binary-coded by a binarization means 70. Since the change of the brightness of a background is very small, the brightness of a background part is close to a '0' level, at the place having the change of the brightness, the luminance slope of the part is emphasized and the border of the shadow can be clarified. Thus, the brightness of a small dark floc is made brighter, made it easier to recognize and the border of the large floc and the background can be clarified.

Description

[Detailed Description of the Invention] [Industrial Application Field] The present invention is applicable to flocs in water purification plants, sewage treatment plants, and other industrial wastewater treatment, as well as flocs in flocculation reactions used in immune reactions. The present invention relates to an image recognition device for aggregates that recognizes particulate objects through image processing.

[Conventional technology]

At a water treatment plant, a flocculant is added to the raw water taken in to flocculate suspended matter to form flocs (hereinafter referred to as flocs).
This floc is settled and removed.

Specifically, after injecting a collector into a rapid mixing pond, the mixture is introduced into a flocculation pond and flocs are formed while being gently stirred. The raw water flowing out of the floc formation pond is led to the settling basin, where the flocs are settled and suspended solids are removed.

Fine particles that do not settle in the sedimentation basin are removed in the filtration basin.

When performing water treatment in this manner, if the microflocs do not coagulate in the floc formation pond, flocs will not be formed, which will accelerate clogging of the filter basin. Therefore,
It is essential to monitor whether flocs are formed.

Conventionally, the state of floc formation has been monitored by visual observation several times a day. For this reason, continuous and quantitative monitoring becomes impossible, and it is inevitable that the detection of abnormal situations such as failure of flocs to occur will be delayed and countermeasures will be delayed. In order to solve this problem, for example,
As described in Japanese Patent No. 143296, a method of monitoring the shape and size of flocs by image processing has been proposed.

Specifically, from a flock image taken with an industrial camera, etc., parts brighter than a predetermined brightness threshold (
A pixel) is set to 1117 levels and is recognized as a flock, and conversely, a portion (pixel) darker than a predetermined value is set to a "0" level and recognized as other than a flock. In this way, the floc images are binarized and image processed, and the floc formation status is monitored.

[Problem that the invention seeks to solve]

In the conventional technology, a part of a flock image whose brightness is brighter than a threshold value is regarded as a flock, and conversely, a part darker than the threshold value is regarded as a background and is binarized. In this case, bright flocks can be binarized even if the threshold is set high, but
Dark flocks will be below the threshold and will be considered as background. On the other hand, if you set the threshold low, dark flocks will also be reduced by 2.
Although it can be converted into a value, the noise existing in the background will be converted into a binary value as a flock. In addition, if the threshold is set low, the background area around bright flocks will also be considered part of the flock, resulting in the flocks being binarized larger than they actually are, or separate flocks nearby. It may also be binarized as one flock.

As described above, the conventional technology has a problem in that it is not possible to binarize flocs with high accuracy.

An object of the present invention is to provide an image recognition device for aggregates that can clearly distinguish and recognize flocs from the background and can detect flocs with high accuracy.

[Means for solving problems]

After the luminance gradient of the flock portion of the photographed flock image is emphasized by a luminance gradient emphasizing means, it is compared with a threshold value and binarized. This makes it possible to clearly distinguish between the flock and the background. This is because the inventor of the present invention found through experiments that the luminance levels of the flocs are different.

[Effect]

Since the background has extremely small changes in brightness, the brightness of the background part is set close to the ``O'' level, and where there is a change in brightness, the brightness gradient of this part is emphasized to clarify the boundary between light and dark. As a result, the brightness of small, dark flocks can be made brighter to make them easier to recognize, and the boundary between large flocks and the background can be made clear.

〔Example〕

FIG. 1 shows an embodiment of the present invention.

FIG. 1 shows an example in which the present invention is applied to a water purification plant.

In FIG. 1, raw water flows into a rapid mixing tank 10, and a liquid polymer flocculant (polyaluminum chloride) or an inorganic flocculant such as aluminum sulfate stored in a flocculant tank 11 is used as a flocculant. Injected by the injection pump 12, an alkaline agent such as calcium hydroxide or sodium carbonate is also injected to promote floc formation. The raw water in the rapid mixing pond 10 is stirred by stirring blades 14. The stirring blades 14 are driven by the stirrer 13. The water into which the flocculant has been injected and stirred is led to a flocculation tank (hereinafter referred to as "floc formation pond") 15.The floc formation pond 15 is partitioned by rectifying walls 16A and 16B having a plurality of holes on the wall surface, and is divided into three ponds. 15A, 15B, and 15G.At each port of the floc formation pond 15, stirring paddles 17Δ, 17B, and 17C are installed, respectively.

Stirring paddles 17A, 17B, 17G are 1 to 10 rpm
It rotates gently around m (paddle peripheral speed = 0.15 to 0.8 m/s).

Aggregate imaging means 18 such as an underwater camera is installed in the pond 15C on the most downstream side of the floc formation pond 15. The grayscale image signal (analog signal) of the aggregate photographed by the aggregate imaging means 18 is input to the image recognition means 3o. Image recognition means 3
0 is the grayscale image storage means 40, the brightness emphasis means 60 and 2
It is composed of a value converting means 70. Details of the aggregate imaging means 18 will be described later. The grayscale image storage means is given a storage command from the timer 35 in units of a predetermined time. Binarization means 7
The image signal binarized with 0 is input to the particle size distribution calculation means 80. The particle size distribution calculating means 80 calculates the particle size distribution of the floc based on the binarized image signal, and stores the calculation result in the volume concentration distribution memory 92. The recognition completion determination means 90 determines whether or not a predetermined number of recognition screens of flock images have been completed. When the number of recognition screens is less than or equal to a predetermined number, the recognition completion determining means 90 instructs the grayscale image storage means 40 to store the grayscale image photographed by the aggregate imaging means 18. When the recognition completion determining means 90 determines that image recognition of a predetermined number of screens (for example, 10 screens) has been completed, the volume concentration distribution stored in the volume concentration distribution memory 92 is input to the aggregation state determination circuit 94. The aggregation state determination circuit 94 determines the number average diameter of the flocs from the volume concentration distribution and applies it to the injection control device 1oO. The injection control device 100 determines the flocculant injection area based on the logarithm qll diameter and controls the injection pump 12.

FIG. 2 shows an example configuration of the image recognition means.

In FIG. 2, the grayscale image storage means 40 is composed of an A/D conversion circuit 41 and a grayscale original image memory 42. The A/D conversion circuit 41 converts the analog grayscale image information obtained by the aggregate imaging means 18 into digital values and adds the digital values to the grayscale original image memory 42 . The grayscale original image memory 42 stores the original image signal when a storage command is given from the timer 35 and the recognition completion determination means 90. 'a The gray image information of the aggregates stored in the gray side image memory 42 is sent to the spatial filtering circuit 6.
Enter 1. The brightness enhancement means 60 is composed of a spatial filtering circuit 61 and a filtering gradation image memory 62. The spatial filtering circuit 61 receives the image signal from the grayscale original image memory 42, performs a spatial filtering operation, and stores the result in the filtering grayscale image memory 62. The stored filtered and shaded images are input to the binarization circuit 71. The binarization means 7o is a binarization circuit 7
1 and a binarization memory 72. Binarization circuit 71
receives the filtering gradation image from the filtering gradation image memory 62, binarizes it, and converts the binarization result into 2 bits.
The data is stored in the value memory 72.

FIG. 3 shows an example configuration of the particle size distribution calculation means 80.

In FIG. 3, the configuration of the particle size distribution calculation means 80 is shown in FIG.

The labeling circuit 81 receives the image signal B from the binarized memory 72 and assigns a number to each of the flocks.

The area calculation circuit 82 calculates the area of each flock for each number, and stores the calculation results in the area memory 82M. The diameter calculation circuit 84 calculates the diameter from the area of the floc and stores the calculation result in the diameter memory 84M. The volume calculation circuit 86 calculates the volume of the floc and stores the calculation result in the counter-product memory 86M. The particle size distribution calculation circuit 88 takes in the floc diameter from the volume memory 86M, calculates the particle size distribution of the floc, and stores it in the particle size distribution memory 88M.

The volume and degree distribution calculation circuit 89 calculates the volume concentration distribution from the memory value of the particle size distribution memory 88M, and when the calculation is completed, it gives a completion signal to the recognition completion determination means 90 and calculates the volume concentration distribution obtained by the calculation as the volume concentration. In addition to the distribution memory 92, input.

FIG. 4 shows an example configuration of the injection control device 100,
Comparison circuit 101, target value setter 102, t? and an injection control circuit 103.

Next, the operation will be explained.

Raw water led from a river or lake (not shown) flows into the rapid mixing tank 10 after sand and coarse particles are settled and removed in a settling tank (not shown). The raw water flowing into the rapid mixing tank 10 contains 2 to 200 microparticles with a diameter of around 1 to 10 μm.
Contains at a concentration of Q. An inorganic flocculant such as a polymer flocculant (polyaluminum chloride) or aluminum sulfate stored in a flocculant tank 11 is supplied to the rapid mixing pond 10 by an injection pump 12 . A stirring blade 14 is provided in the rapid mixing pond 10.
Stirred by This stirring causes the flocculant to diffuse into the raw water. Suspended particles are negative colloids whose particle surfaces are negatively charged, and a positively charged flocculant binds (agglomerates) the numerous suspended particles to each other. The residence time in the rapid mixing tank 10 is 10 to 5 minutes, and during this time suspended fine particles aggregate to form microflocs (floc cores) with a particle size of 10 to 100 μm.The mixed solution containing 6 microflocs is It is guided to the flocculation tank 15. Flock formation pond 15
Then, the water flows down three formation ponds 15A, 15B, and 15G in sequence. The flow regulating walls 16A and 16B prevent the mixed liquid from being sufficiently mixed in the floc formation pond 15, short-circuiting near the water surface, and flowing out from the outlet. Formation ponds 15A, 15B and 1
The residence time of 5C is 5 to 15 minutes each (15 minutes in total in 3 ponds).
minutes to 45 minutes). Each pond has a stirring paddle 17
It is gently stirred by A, 17B and 17C. The flocculant is sufficiently supplied in the rapid mixing pond 10, and the flocculant is attached to the surface of the microflocs. For this reason, the microflocs in the flocculation pond 15 collide or come into contact with each other due to stirring and coagulate. While the flocs remain in the floc formation pond 15 for 15 to 4.5 minutes and are stirred, the flocs grow into flocs with a particle size of 100 to 5000 μm. The condition of the flocs in the formation pond 15C is photographed by the aggregate imaging means 18. The density image signal of the aggregate obtained from the aggregate imaging means 180 is outputted from the image recognition means 30 by D.
/A converter 41.

The D/A converter 41 constantly digitizes the grayscale image signal.
i'';'i and input it to the grayscale original image memory 42. If the D/A converter 41 converts it into a 7-bit digital signal, the brightness of each pixel is digitized into 128 levels. Hereinafter, this embodiment will be explained using an example of a 256-pixel screen with 8 bits in both the horizontal and vertical directions.The grayscale image memory 42 has a storage area as shown in FIG. 5 corresponding to 256 x 258 pixels. .If the horizontal arrangement in Fig. 5 is i row and the vertical arrangement is j column, then
In each storage area in row i and column j in the grayscale image memory 42, the value go(i, j): i=
1-256. j=1 to 256 are stored.

Note that the brightness go(i, j) of each pixel is digitized in 128 steps. The grayscale image information stored in the grayscale original image memory 42 is taken into the spatial filtering circuit 61.

The spatial filtering circuit 61 receives the image signal from the grayscale original image memory 42 and emphasizes the brightness gradient between the flock and the background. The calculation results of the spatial filtering circuit 61 are stored in a grayscale image memory 62. The grayscale image memory 62 has 2
It has a memory storage area corresponding to 56×256 pixels.

The operation of the spatial filtering circuit 61 will be explained in detail.

The local image area to be subjected to the spatial filtering method is an nXn area with n pixels in the horizontal direction and n pixels in the column direction. If n = 3, the local image area will be 3 pixels in the horizontal direction and 3 pixels in the vertical direction. FIG. 6 shows a 3×3 local image area. On the other hand, the weighting coefficient matrix for spatial filtering is similarly defined in an nXn area with n pixels in the X direction and n pixels in the Y direction. If n=3, it becomes a 3×3 matrix.

Let F be a weighted product sum matrix (weighted coefficient matrix) for this spatial filtering. FIG. 7 represents a 3×3 weighted product sum matrix F. The calculation of spatial filtering is as shown in equation (1),
The luminance go(1+,j) of each pixel of the grayscale image and the weighted product sum matrix f(i) of spatial filtering.

j), and then add all the multiplication results.

The spatial filtering calculation result g-umbrella c2.2) is stored in the area of the memory 62 corresponding to the central pixel, as shown in FIG.

g”(2+2)=(go(1,1)xf(1,1)
+go(1,2)xf(1,2) +go(1,3)Xf(1,3) +go(2,1)Xf(2,1) +go(2# 2)x f (2,2) +go(2,
3) Xf (2, 3) + go (3, 1) X f (3, l) + go (3, 2) X f (3, 2) + go (3, 3)・・・
...(1) Here, S is the scaling coefficient, and the calculation result is 1
The number is selected to be within 127 and not to exceed 28.

Equation (1) can be summarized as shown below.

k=-1Ω knee 1 Xf(2+, 2+Q))/S (2) Here, k and Q are symbols for changing the elements of the array.

After the calculation of the pixel g (2, 2) is completed, as shown in Fig. 8, the value of the next pixel g'' (2*3) is calculated by one pixel.At this time, as shown in Fig. 9 go like that
A 3×3 local image area with (2, 3) in the center is the target of calculation, and in the same way as equation (1), g”(2+3)
Calculate. In this way, calculations are performed sequentially in the column direction (horizontal direction) while shifting one pixel at a time, and when the calculation for one column is completed, the same calculation is performed for the second row.

The calculation formula for the pixel in the i-th row and j-th column is as follows.

, °, umbrella (it j) = [Ko(i-1,, j-1
)Xf(1,l)+go(i-1,j)Xf(1,2)+pco(i-1,j+1)Xf(1,3)
, j-1)Xf(2,1)+go(i,j)Xf(2
,2) +go(i,j+1)Xf(2,3) +go(itl,j-1)Xf(3,1)+go(it
1. ．． 1)Xf(3,2)+go(it1.j+1)x
f(3,3))/S...(3) Xf(2+, 2+Q))/S...(4) i=2~255. Calculate from j=2 to 255 and get 225
When all calculations up to the luminance g$(255, 255) of the pixel in row 225 are completed, the calculation for one screen is completed.

Note that spatial filtering calculations are not performed for all pixels in the 1st and 256th rows and all pixels in the 1st and 256th columns.

As described above, the spatial filtering circuit 61 performs (1)
The calculations of Equations and Equations (3) are executed, and the calculated brightness is stored in the filtering grayscale image memory 62.

Here, the weighting factor f (it j) for raw food 1℃2-11, which is an important factor when performing spatial filtering, is
I will explain about it. The weighting coefficient f (ly J) is n
An example of the weighted product sum matrix F when =3 is shown in FIG.

To explain the effect of this example, the luminance go(
The case where te j) is an image as shown in FIG. 11 will be explained. The rectangular cells in Figure 11 represent pixels,
The number inside represents the brightness of the pixel. FIG. 11 shows a flock with a luminance of 3 on a background of luminance 1 and a size of one pixel. In this case, the difference in brightness between the background and the flock is two levels.

When the calculation of equation (3) is executed using the weighted product sum matrix F shown in FIG. 10 on the image shown in FIG. 11, the result is shown in FIG. 12.

In addition, in the calculation, S = 1, and in Fig. 12, U
The brightness value of the outermost pixel shown by cannot be calculated, so U
The value of is 0.

As is clear from FIG. 12, since the luminance of the flock increases to 16, the brightness of the flock is emphasized, while the luminance of the surrounding pixels is 0 or -4. Therefore, the luminance difference between the background and the flock is 16 or more. In this way, it can be seen that the brightness of the flock has become even brighter compared to the background.

As described above, the calculation of the spatial filtering circuit 610 is executed, and the brightness of the calculation result is stored in the filtering gradation image memory 62.

The gray image signal whose brightness gradient has been emphasized by the spatial filtering method as described above is input to the binarization circuit 71. The binarization circuit 71 converts the filtered gray image g"(it) stored in the filtered gray image memory 62.
j) and binarizes this image. In other words, if the binarization threshold is L, then this pixel is 1"
Conversely, if it is 57 or more, this pixel is set to the "0" level. This binarized signal having a value of 110 I+ level or "1" level is defined as b (1+ J).

The binarization circuit 710 executes the following calculation.

If g umbrella (i, j)≧Lt, then b i j = 1・
...(5) If g umbrella (x, j) < Lt, then bij
=Q...(6) As a result, pixels whose gradation luminance go(i, j) as a result of spatial filtering is higher than the threshold Lt are recognized as pixels corresponding to flocks, and are designated as "1".
In contrast, the lower luminance part is recognized as a non-flock pixel and becomes the "0" level.In the end, the set of pixels represented by the -II I 11 level is recognized as a flock. Figure 13 shows Figure 12 with threshold value Lt=
2 and shows the result of binarization.

Assuming that the entire image composed of the binarization result b (1+j) is B, this image B is stored in the binarization memory 72 . Image B is obtained by particle size distribution calculation means 8 as shown in FIG.
0 is entered and the floc particle size distribution is now calculated.

Now, the particle size distribution calculation means 80 calculates the particle size distribution in the following manner.

First, as shown in FIG.
・Add the number m. Here, m is the total number of flocks. The area calculation circuit 82 calculates the area of the flock for each labeled number using the following formula.

A=kl-AP...
(7) Here, A is the projected area of the floc [Tm2], A
P is the number of pixels of each flock, k*
is the conversion constant (re”/pixel). pixa
l is a unit representing a pixel. The calculation of equation (7) is executed for each numbered flock by the labeling circuit 81, and the result is stored in the area memory 82M. The diameter calculation circuit 84 assumes a circle having the same area as each floc and calculates its diameter d using the following formula.

d=J[] ...(8) Calculate this diameter for each area and store the result in diameter memory 8
Store in 4M. The volume calculation circuit 86 has a diameter memory 84M.
Enter the diameter of each floc from
Calculate V using the following formula.

v=xd8/6...
...(9) The calculation result of volume for particle size is stored in volume memory 8.
It is stored in 6M. The particle size distribution calculation circuit 88 takes in the volume V of each floc from the volume memory 86M, determines which classification the particle size of each floc falls into, and stores the volume of each floc in the corresponding memory of the particle size distribution memory 88M. Add to area. Assuming that the particle size classification width is 0.1 m, the classification is performed, for example, into 51 divisions as shown below. The particle size distribution memory 88M also has 51 storage areas.

Dt: O~0.1m+ Dz: 0.1~0.2ff111 Ds: 0, 2~0, 3 tmDso:
4,9~5,0 mmD61:5.01! l11
- As an example, if the diameter of the floc is 0.25rrn, the diameter of the floc is 0.008L8nm3.

If the volume of the particle size DI is Vt, a volume of 0.00 is stored in the storage area corresponding to the particle size D3 of the particle size distribution memory 88M.
818 is stored. In this way, while determining which classification the particle size of each floc belongs to, the particle size distribution of the flocs is obtained by sequentially adding to each area of the particle size distribution memory 88M.

The volume concentration distribution calculation circuit 89 calculates the floc volume concentration distribution V,' (distribution indicating how much floc volume ■1 of each particle size D1 is in a unit volume) from the volume value vi of the particle size distribution memory 88M using the following formula. calculate.

Vl' = Vt/ (N-V-)...
=・(10) Here, N is the number of recognition times (number of processed screens), V
, is the volume captured in one screen.

An example of the obtained volume concentration distribution (vertical axis: two particle diameters DI + horizontal axis: volume concentration ■1') is shown in FIG. Curve a in Figure 15
is a theoretical curve of a lognormal distribution obtained from the histogram of the volume concentration distribution shown in FIG.

In this way, the recognition completion determining means 90 determines whether image recognition of flocs has been completed for N number of screens each time the particle size distribution calculating means 80 finishes calculating the volume concentration distribution for each screen. If the number of recognitions is less than N times, the image captured by the aggregate imaging means 18 at that time is stored in the grayscale image storage means 40, and the above-described image processing of the flocs is repeated. When the number of recognitions reaches N, the value of the volume concentration distribution calculated using equation (10) is stored in the volume concentration distribution memory 92.

In addition, in the above explanation, an example was explained in which the volume concentration distribution is calculated for each recognition screen, but after completing a predetermined number of recognitions (
10) Calculation of formula may be executed.

The agglomeration state determination circuit 94 calculates the logarithmic average diameter I)a of the floc particle size distribution from the value in the 1-volume concentration distribution memory 92 using the following equation.

i=L C1 Logarithmic average diameter determined by the agglomeration state determination circuit 94.

is input to the injection control device 100. The logarithmic mean diameter Da output from the aggregation state determination circuit 94 is input to the comparison circuit 101. The comparator circuit 101 calculates the deviation ΔDa between the target value Dec of the logarithmic average diameter given from the target value setter 102 and the calculated value Da using the following equation.

ΔD, =D・Direct−D, ・・・・・・
(12) The injection control circuit 103 operates the injection pump 12 based on the deviation ΔDt to control the amount of coagulant injection.

Specifically, if the deviation ΔDa is negative, the amount of coagulant injected is increased, and conversely, if the deviation ΔDt is positive, the amount of coagulant injected is decreased. The relationship between the logarithmic average diameter D1 and the flocculant injection MP is as shown in Fig. 16, but the maximum injection amount Pmax and the minimum injection ip, tn are set for the flocculant injection amount to prevent abnormal injection. .

The flocculant injection is controlled as described above, and since the luminance of the flocs is emphasized and recognized by the luminance emphasizing means 60 when the floc image is binarized, the flocs can be image recognized with high accuracy. As a result, by controlling the injection amount of the flocculant, etc., floc formation can always be performed stably.

Further, in this embodiment, the inter-page filtering circuit 61 uses a weighted product-sum matrix to emphasize the luminance gradient of the flock image, and then binarizes the image.

The flock can be recognized by clearly distinguishing it from the background. This will be specifically explained below using FIG. 17.

FIG. 17(a) shows the one-dimensional brightness distribution of a large flock with a diameter of three pixels or more at the brightness go(i, j) of the grayscale image. When such a bright flock image is spatially filtered, the brightness after filtering go
In (iej), as shown in FIG. 17(b), the brightness gradient between the flock and the background is emphasized, so the brightness of the flock becomes even brighter. Similarly, in the case of a minute floc with a diameter of 1 to 2 pixels as shown in FIG. 17(c), the luminance of the floc is emphasized as shown in FIG. 17(d).

Therefore, by performing binarization using the threshold Lt as shown in FIGS. 17(b) and (d), the flocs can be binarized regardless of the size of the flocs.

By the way, in the above embodiment, it was explained that the luminance of the flock can be emphasized using the weighted product sum matrix F as shown in FIG. By setting this value, the brightness of a background with a gentle brightness gradient can be set to zero. The brightness of the background changes depending on the intensity of the lighting, but this brightness level is always zero due to spatial filtering, so it is possible to effectively emphasize the contrast between the flock and the background. This effect does not change even if the temperature changes.

Therefore, even if the intensity of illumination gradually weakens over a long period of measurement, stable floc recognition can be achieved without being affected by this.

Next, although the above-mentioned embodiment has been described with respect to a flock of one pixel, an example of an image in which the grayscale image information go(iej) is a flock of a plurality of pixels as shown in FIG. 18 will be described. FIG. 18 shows a case where floes with a luminance of 3 or 4 exist on a background with a luminance of 1. 10 for the image in Figure 18.
The result of spatial filtering using equation (2) using the weighted product sum matrix F in the figure is as shown in Figure 19.As can be seen from Figures 18 and 19, the spatial Filtering emphasizes the brightness of the flock.

The grayscale image after filtering in Figure 19 is 2 with threshold = 2.
When converted into values, it becomes as shown in Figure 20.

Next, an example in which the size n of the weighted product sum matrix is 3 or more in the present invention will be described. This example assumes n ==5 and n ==
This is an example in which the luminance of the flock can be emphasized more accurately than in case 3.

In this example, it is assumed that the weighted product sum matrix F of the spatial filtering circuit 61 is given as shown in FIG. Using this weighted product-sum matrix F, it is possible to (1) emphasize the brightness of flocs with flat peak portions, (2) remove noise, and (3) emphasize the brightness of *small blocks.

The effects of these (1) to (3) will be explained below.

The effect when applying the weighted product sum matrix F (n=5) shown in FIG. 21 and the weighted product sum matrix (n=3) shown in FIG. 10 will be explained using FIGS. 22 to 24.

FIG. 22(a) shows an example of a one-dimensional luminance distribution of the flocs in which the peak of luminance is flat. Since the brightness change is small at such a peak, when spatial filtering is performed using the weighted product-sum matrix F shown in FIG. 10, the brightness at the peak becomes zero as shown in FIG. 23(b). However, when spatial filtering is performed using the weighted product-sum matrix shown in FIG. 21, the peak portion is emphasized instead of being reduced to zero, as shown in FIG. 24(c). Therefore, this emphasized portion can be effectively binarized as a flock.

Next, the noise removal effect will be explained. FIG. 23(,) represents the one-dimensional luminance distribution of noise generated in the imaging camera in units of one pixel. When such noise is processed using the weighted product-sum matrix F shown in FIG. 10, the brightness of the noise is emphasized as shown in FIG. 23(b). This is the weighted product sum matrix shown in Figure 10.

This is because only the central value is a positive value, so even the brightness of one pixel of noise is strong. However, if a spatial filtering calculation is performed using the weighted product-sum matrix (n=5) shown in FIG. 21, the brightness can be reduced to 0 as shown in FIG. 23(c). That is, the influence of noise can be removed.

Figure 24(a) shows the case where there are minute flocs in the width of the noise, Figure 24(b) shows the result of processing using the weighted product sum matrix shown in Figure 10, and Figure 24(c) is the result of processing using the weighted product sum matrix shown in FIG.

In this way, it is possible to remove noise and at the same time emphasize the brightness of minute flocs.

As explained above, by performing spatial filtering using a weighted product sum matrix of n=5, the luminance of the flocs is
It is possible to emphasize more effectively than in the case of =3.

As shown in Figure 21, the weighted product sum matrix F can effectively emphasize the luminance of the flock (high luminance at the center and low luminance at the surroundings) by setting the center value high and the surrounding values low. . Note that the same effect can be obtained even if the size of the weighted product sum matrix is 5 or more.

〔Effect of the invention〕

According to the present invention, the brightness of the flock can be emphasized and the brightness of the background can be made uniform through spatial filtering, so that the boundary between the flock and the background can be made clear, and the brightness of noise other than the flock can be relatively suppressed. Therefore, it is possible to selectively emphasize only the luminance of the flock without being influenced by gradual changes in background luminance such as uneven illumination or changes over time, or by noise from the imaging system. In the present invention, the luminance of the flocs is emphasized in this manner, and then the flocs are recognized by binarization. Therefore, flocks can be image-recognized with high accuracy.

Note that the present invention can be applied to image recognition of aggregated particles other than flocs and particulate objects in water purification plants. For example, an image needle 111 of activated sludge flocs in a sewage treatment plant
It can be applied to measurement of the particle size of immobilized particles obtained by immobilizing microorganisms using immobilization agents such as 11 and sodium alginate, and measurement of agglutination reaction products in antigen-antibody reactions. It can also be applied to the measurement of all kinds of powdered materials such as pulverized coal and American flour.

[Brief explanation of drawings]

1 is a block diagram showing an example of the present invention, FIG. 2 is a detailed block diagram showing an example of an image recognition means, FIG. 3 is a detailed block diagram showing an example of a particle size distribution calculation means, and FIG. 4 is a block diagram showing an example of a particle size distribution calculation means. is a detailed diagram showing an example of the control device, FIG. 5 is a diagram showing the brightness of a grayscale image, FIGS. 6 to 13 are diagrams for explaining spatial filtering calculations, and FIGS. 14 and 15. 16 is a characteristic diagram for explaining flocculant injection control, and FIGS. 17 to 24 are diagrams showing calculations and effects of spatial filtering. It is. 15... Floc formation pond, 18... Aggregate imaging means. 30... Image recognition means, 40... Grayscale image information storage means, 60... Aggregate brightness enhancement means, 70... Binarization means, 80... Particle size distribution calculation means, 90... - Recognition completion determination means, 100... injection control device.

Claims (1)

[Scope of Claims] 1. A flocculating tank for forming aggregates of suspended substances contained in a liquid, and an aggregate imaging means for photographing the state of the aggregates in the flocculating tank and converting luminance information into an electrical signal. , a gradation image storage means for storing gradation image information of the agglomerate based on the electric signal obtained from the agglomerate imaging means; a brightness emphasizing means for emphasizing the luminance gradient of the agglomerate portion of the gradation image signal; binarizing means for binarizing the grayscale image signal obtained from the luminance emphasizing means according to the luminance level for each pixel; An image recognition device for aggregates, characterized in that it is configured to distinguish between non-aggregates and non-aggregates. 2. The apparatus for image recognition of aggregates according to claim 1, wherein the brightness emphasizing means emphasizes the brightness gradient of the aggregate portion by spatial filtering.